audit2: cli + modelregistry + dead-code cleanup (2 med, rest low)

cli: --reasoning-effort notice evaluates the EFFECTIVE model (works without --model) (MED); unsafe warning fires for --auto high too (MED); --auto value validated even with --skip-permissions-unsafe (LOW); DefaultRegistry built once per exec, not 3x (LOW); --profile kept as intentional legacy-compat accept-and-ignore (contested finding resolved, doc-commented). modelregistry: removed dead FallbackFor. tools: removed dead TaskNonInteractiveMessage + dead OnSandboxDecision hook (no setter anywhere); fixed the now-stale loop.go comment. TDD where behavioral; build/vet/full-suite green.

Refs #102

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -1,54 +0,0 @@
-name: Bug report
-description: Report a reproducible problem in Zero
-labels: ["bug"]
-body:
-  - type: markdown
-    attributes:
-      value: |
-        Thanks for filing a bug. Clear, reproducible reports are the most useful way to help the
-        project. Please do **not** report security vulnerabilities here — see SECURITY.md.
-  - type: input
-    id: version
-    attributes:
-      label: Version / branch / commit
-      description: Output of `zero version`, or the branch/commit you built from.
-      placeholder: "zero dev (commit abc1234) / main"
-    validations:
-      required: true
-  - type: input
-    id: env
-    attributes:
-      label: OS and environment
-      description: Operating system, terminal, and any relevant environment details.
-      placeholder: "macOS 14 / iTerm2 / Go 1.25.1"
-    validations:
-      required: true
-  - type: textarea
-    id: repro
-    attributes:
-      label: Steps to reproduce
-      description: A minimal reproduction case when possible.
-      placeholder: |
-        1. Run `zero ...`
-        2. ...
-    validations:
-      required: true
-  - type: textarea
-    id: expected
-    attributes:
-      label: Expected behavior
-    validations:
-      required: true
-  - type: textarea
-    id: actual
-    attributes:
-      label: Actual behavior
-    validations:
-      required: true
-  - type: textarea
-    id: logs
-    attributes:
-      label: Relevant logs, screenshots, or error messages
-      description: Redact any secrets (API keys, tokens) before pasting.
-    validations:
-      required: false

.github/ISSUE_TEMPLATE/config.yml

@@ -1,5 +0,0 @@
-blank_issues_enabled: false
-contact_links:
-  - name: Security vulnerability
-    url: https://github.com/Gitlawb/zero/security/advisories/new
-    about: Report security issues privately via GitHub Security Advisories (see SECURITY.md). Do not file a public issue.

.github/ISSUE_TEMPLATE/feature_request.yml

@@ -1,41 +0,0 @@
-name: Feature request
-description: Propose a new capability or change
-labels: ["enhancement"]
-body:
-  - type: markdown
-    attributes:
-      value: |
-        Please do not begin implementation until the issue has been reviewed and marked with the
-        `issue-approved` label (see CONTRIBUTING.md).
-  - type: textarea
-    id: problem
-    attributes:
-      label: Problem
-      description: The problem you are trying to solve.
-    validations:
-      required: true
-  - type: textarea
-    id: proposal
-    attributes:
-      label: Requested behavior
-      description: The behavior or functionality you are requesting.
-    validations:
-      required: true
-  - type: textarea
-    id: why
-    attributes:
-      label: Why it belongs in Zero
-    validations:
-      required: true
-  - type: textarea
-    id: alternatives
-    attributes:
-      label: Alternatives or workarounds considered
-    validations:
-      required: false
-  - type: textarea
-    id: context
-    attributes:
-      label: Examples, screenshots, or prior discussion
-    validations:
-      required: false

.github/PULL_REQUEST_TEMPLATE.md

@@ -1,16 +0,0 @@
-## Summary
-
-<!-- What does this change do, and why? -->
-
-## Linked issue
-
-<!-- Per CONTRIBUTING.md, PRs need an approved parent issue with the `issue-approved` label. -->
-Fixes #
-
-## Checklist
-
-- [ ] The linked issue already has the `issue-approved` label.
-- [ ] `go build ./...`, `go vet ./...`, and `go test ./...` pass locally.
-- [ ] `gofmt` clean.
-- [ ] Tests added/updated for the change (and run under `-race` where relevant).
-- [ ] UI changes include screenshots or a short recording where possible.

.github/workflows/ci.yml

@@ -20,9 +20,7 @@ jobs:
 
     steps:
       - name: Checkout
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
-        with:
-          persist-credentials: false
+        uses: actions/checkout@v4
 
       - name: Setup Go
         uses: actions/setup-go@40f1582b2485089dde7abd97c1529aa768e1baff
@@ -30,21 +28,6 @@ jobs:
           go-version-file: go.mod
           cache: true
 
-      - name: Check formatting
-        if: matrix.os == 'ubuntu-latest'
-        shell: bash
-        run: |
-          unformatted="$(gofmt -l .)"
-          if [ -n "$unformatted" ]; then
-            echo "gofmt needed on:" >&2
-            echo "$unformatted" >&2
-            exit 1
-          fi
-
-      - name: Vet
-        if: matrix.os == 'ubuntu-latest'
-        run: go vet ./...
-
       - name: Test
         run: go test ./...
 
@@ -60,9 +43,7 @@ jobs:
 
     steps:
       - name: Checkout
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
-        with:
-          persist-credentials: false
+        uses: actions/checkout@v4
 
       - name: Setup Go
         uses: actions/setup-go@40f1582b2485089dde7abd97c1529aa768e1baff
@@ -78,56 +59,8 @@ jobs:
 
       - name: Upload performance report
         if: always()
-        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
+        uses: actions/upload-artifact@v4
         with:
           name: zero-performance-smoke
           path: dist/perf/perf-bench.json
           if-no-files-found: warn
-
-  security:
-    name: Security & code health
-    runs-on: ubuntu-latest
-
-    steps:
-      - name: Checkout
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
-        with:
-          persist-credentials: false
-
-      - name: Setup Go
-        uses: actions/setup-go@40f1582b2485089dde7abd97c1529aa768e1baff
-        with:
-          go-version-file: go.mod
-          cache: true
-
-      # govulncheck, deadcode, and golangci-lint each resolve a toolchain from
-      # their own modules, which can drop below the version go.mod requires and
-      # then fail to load our packages. Pin GOTOOLCHAIN to the go.mod toolchain
-      # so all three run under it.
-      - name: Pin toolchain from go.mod
-        run: |
-          toolchain="$(awk '/^toolchain /{print $2}' go.mod)"
-          echo "GOTOOLCHAIN=${toolchain:-auto}" >> "$GITHUB_ENV"
-
-      # Hard gate: fails the build when code reaches a known vulnerability. A stdlib
-      # CVE is cleared by a toolchain bump (see go.mod). May also flag a newly
-      # published advisory on an unrelated PR — intentional: do not ship known vulns.
-      - name: govulncheck
-        run: go run golang.org/x/vuln/cmd/[email protected] ./...
-
-      # Advisory: reports functions unreachable from any cmd/* main so dormant
-      # code is visible in CI. Non-blocking while the dormant subsystems are
-      # still being wired or removed.
-      - name: deadcode (advisory)
-        continue-on-error: true
-        run: go run golang.org/x/tools/cmd/[email protected] -test=false ./...
-
-      # Advisory: catches what deadcode's whole-program reachability analysis
-      # doesn't, unused private functions/assignments reachable within a
-      # package but never actually called, plus staticcheck-style correctness
-      # and readability issues. Scoped to a few linters rather than the full
-      # default battery, and non-blocking, while the existing findings across
-      # the repo are cleaned up incrementally (see #527).
-      - name: golangci-lint (advisory)
-        continue-on-error: true
-        run: go run github.com/golangci/golangci-lint/v2/cmd/[email protected] run --enable-only unused,ineffassign,staticcheck ./...

.github/workflows/pr-auto-review.yml

@@ -21,7 +21,7 @@ jobs:
 
     steps:
       - name: Checkout
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
+        uses: actions/checkout@v4
         with:
           fetch-depth: 0
 
@@ -80,7 +80,7 @@ jobs:
 
       - name: Post review summary
         if: always()
-        uses: actions/github-script@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7
+        uses: actions/github-script@v7
         env:
           ZERO_REVIEW_BODY_PATH: ${{ steps.review-summary.outputs.path }}
         with:

.github/workflows/release-artifacts.yml

@@ -15,29 +15,22 @@ jobs:
   package:
     name: Package (${{ matrix.os }})
     runs-on: ${{ matrix.os }}
-    permissions:
-      contents: write # create/upload the GitHub Release on a tag push
     strategy:
       fail-fast: false
       matrix:
-        # `zero-release package` builds + smoke-tests natively (host==target), so each
-        # target platform needs its own native runner. These five cover the platforms
-        # install.sh / install.ps1 / the npm wrapper resolve.
         os:
-          - ubuntu-latest      # linux-x64
-          - ubuntu-24.04-arm   # linux-arm64
-          - macos-latest       # macos-arm64 (Apple Silicon)
-          - macos-15-intel     # macos-x64 (Intel)
-          - windows-latest     # windows-x64
+          - ubuntu-latest
+          - macos-latest
+          - windows-latest
 
     steps:
       - name: Checkout
         if: ${{ github.event_name != 'workflow_dispatch' || inputs.ref == '' }}
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
+        uses: actions/checkout@v4
 
       - name: Checkout requested ref
         if: ${{ github.event_name == 'workflow_dispatch' && inputs.ref != '' }}
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
+        uses: actions/checkout@v4
         with:
           ref: ${{ inputs.ref }}
 
@@ -50,23 +43,6 @@ jobs:
       - name: Test
         run: go test ./...
 
-      # zero-release names assets zero-v<package.json version>-… while the release
-      # is published under the pushed git tag. If they disagree, install.sh/install.ps1
-      # and the npm postinstall all build a download URL that 404s. Fail fast so a
-      # forgotten version bump can never publish mis-versioned, un-installable assets.
-      # Matches on the ref (not event_name) because release-please dispatches this
-      # workflow at the tag ref — its GITHUB_TOKEN-created tag cannot fire `push`.
-      - name: Check tag matches package.json version
-        if: ${{ startsWith(github.ref, 'refs/tags/v') }}
-        shell: bash
-        run: |
-          pkg_version="$(node -p "require('./package.json').version")"
-          if [ "v$pkg_version" != "$GITHUB_REF_NAME" ]; then
-            echo "::error::tag $GITHUB_REF_NAME does not match package.json version v$pkg_version — bump package.json before tagging so the release assets and the npm package agree." >&2
-            exit 1
-          fi
-          echo "tag $GITHUB_REF_NAME matches package.json version v$pkg_version."
-
       - name: Package release artifact
         run: go run ./cmd/zero-release package
 
@@ -74,169 +50,8 @@ jobs:
         run: go run ./cmd/zero-release verify
 
       - name: Upload package
-        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4
+        uses: actions/upload-artifact@v4
         with:
           name: zero-${{ runner.os }}-${{ runner.arch }}
           path: dist/release/*
           if-no-files-found: error
-
-      # On a `v*` tag ref (pushed manually, or dispatched at the tag by the
-      # release-please workflow), publish the packaged archive + .sha256 as GitHub
-      # Release assets so the documented `scripts/install.sh` / `install.ps1` path
-      # (which reads /releases/latest) works. Uses the preinstalled `gh` (no new
-      # action dependency); each matrix OS creates the release if absent
-      # (best-effort — release-please normally creates it first) then uploads its
-      # own assets with --clobber. Tags come from merging the release-please PR.
-      - name: Publish to GitHub Release
-        if: ${{ startsWith(github.ref, 'refs/tags/v') }}
-        shell: bash
-        env:
-          GH_TOKEN: ${{ github.token }}
-        run: |
-          tag="${GITHUB_REF_NAME}"
-          gh release create "$tag" --repo "$GITHUB_REPOSITORY" --title "$tag" --generate-notes 2>/dev/null || true
-          gh release upload "$tag" dist/release/* --repo "$GITHUB_REPOSITORY" --clobber
-
-  # Publishes @gitlawb/zero to npm: first the five platform payloads (versions
-  # of the SAME package at suffixed versions like X.Y.Z-linux-x64, carrying the
-  # native binary + vendored helpers), then the wrapper at X.Y.Z whose
-  # optionalDependencies alias those platform versions. Platform versions MUST
-  # publish first — the wrapper pins them exactly, so the reverse order would
-  # open a window where installs resolve aliases that 404. See
-  # docs/NPM_PACKAGING.md for the full contract.
-  #
-  # Auth is npm OIDC trusted publishing: no NPM_TOKEN secret. The package's
-  # npmjs.com settings must list this repo + workflow file as a trusted
-  # publisher, and provenance is then generated automatically. Note npm only
-  # allows configuring a trusted publisher on an EXISTING package, so the very
-  # first @gitlawb/zero publish is a one-time manual `npm publish` bootstrap.
-  # Because the platform payloads are versions of the same package, they reuse
-  # that single trusted-publisher configuration.
-  publish-npm:
-    name: Publish npm package
-    needs: package
-    if: ${{ startsWith(github.ref, 'refs/tags/v') }}
-    runs-on: ubuntu-latest
-    # Manual approval gate: this environment carries a required-reviewer rule
-    # (Settings -> Environments -> npm-publish), so the job pauses until a
-    # maintainer approves it in the Actions UI before anything reaches npm.
-    environment:
-      name: npm-publish
-      url: https://www.npmjs.com/package/@gitlawb/zero
-    permissions:
-      contents: read
-      id-token: write # OIDC token exchange for npm trusted publishing + provenance
-    steps:
-      - name: Checkout
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
-
-      # No registry-url here on purpose: setup-node's registry-url writes an
-      # .npmrc line referencing $NODE_AUTH_TOKEN, and npm hard-fails on the
-      # unset env var. Trusted publishing needs no token at all.
-      - name: Setup Node
-        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
-        with:
-          node-version: 24
-
-      # OIDC trusted publishing needs npm >= 11.5.1. Use the npm bundled with
-      # the pinned Node toolchain (Node 24 ships >= 11.5.1) rather than pulling
-      # npm@latest from the registry inside the publish path — an unpinned
-      # install there would be avoidable supply-chain exposure. Fail fast if a
-      # future runner/toolchain change ever regresses the bundled npm.
-      - name: Check npm supports trusted publishing
-        run: |
-          npm --version | node -e '
-            const version = require("fs").readFileSync(0, "utf8").trim();
-            const [major, minor, patch] = version.split(".").map(Number);
-            const ok =
-              major > 11 ||
-              (major === 11 && (minor > 5 || (minor === 5 && patch >= 1)));
-            if (!ok) {
-              console.error(
-                `::error::bundled npm ${version} is too old for OIDC trusted ` +
-                  `publishing (needs >= 11.5.1) — bump node-version in this job.`,
-              );
-              process.exit(1);
-            }
-            console.log(`npm ${version} supports trusted publishing.`);
-          '
-
-      # Probe the exact URLs the wrapper's first-run downloader builds
-      # (internal/release/release.go asset-name scheme) as an anonymous client
-      # — install.sh/install.ps1 and `zero update` read them too. This also
-      # blocks publishing while the repo is still private: private release
-      # assets 404 without auth, exactly as they would for an installing user.
-      - name: Verify release assets are downloadable
-        shell: bash
-        run: |
-          version="$(node -p "require('./package.json').version")"
-          base="https://github.com/${GITHUB_REPOSITORY}/releases/download/v${version}"
-          for asset in \
-            "zero-v${version}-linux-x64.tar.gz" \
-            "zero-v${version}-linux-arm64.tar.gz" \
-            "zero-v${version}-macos-x64.tar.gz" \
-            "zero-v${version}-macos-arm64.tar.gz" \
-            "zero-v${version}-windows-x64.zip"; do
-            for url in "${base}/${asset}" "${base}/${asset}.sha256"; do
-              if ! curl -fsSLI -o /dev/null "$url"; then
-                echo "::error::${url} is not publicly downloadable — the wrapper's fallback downloader and install.sh would fail. Is the repo public and are all assets uploaded?" >&2
-                exit 1
-              fi
-            done
-          done
-          echo "all release assets for v${version} are publicly downloadable."
-
-      - name: Download release artifacts
-        uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
-        with:
-          pattern: zero-*
-          path: dist/artifacts
-          merge-multiple: true
-
-      - name: Assemble npm packages
-        run: node scripts/npm/build-platform-packages.mjs --artifacts-dir dist/artifacts --out-dir dist/npm
-
-      # Platform versions are semver PRERELEASES of the wrapper version
-      # (X.Y.Z-linux-x64), so they must never publish under the default
-      # `latest` dist-tag — that would make `npm install @gitlawb/zero`
-      # deliver a platform payload as the CLI. Publish them under a dedicated
-      # tag and assert `latest` survived before the wrapper goes out.
-      - name: Publish platform packages to npm
-        shell: bash
-        run: |
-          version="$(node -p "require('./package.json').version")"
-          for dir in dist/npm/platforms/*/; do
-            echo "publishing ${dir}"
-            (cd "$dir" && npm publish --access public --tag platform)
-          done
-          # Tolerate a failed lookup (registry lag, or no latest tag yet on a
-          # bootstrap publish): under set -e a bare failure here would abort
-          # AFTER the platform payloads went out, leaving no wrapper and a
-          # re-run that fails on republishing existing versions. An empty
-          # value falls through the guard below.
-          latest="$(npm view @gitlawb/zero dist-tags.latest || true)"
-          case "$latest" in
-            "${version}-"*)
-              echo "::error::dist-tag latest points at platform version ${latest} — a platform publish clobbered it. Fix the dist-tag before the wrapper publish." >&2
-              exit 1
-              ;;
-          esac
-          echo "dist-tag latest is ${latest} (unchanged by platform publishes)."
-
-      - name: Publish wrapper to npm
-        shell: bash
-        run: |
-          version="$(node -p "require('./package.json').version")"
-          (cd dist/npm/wrapper && npm publish --access public)
-          # Confirm the wrapper owns `latest` (retry: registry metadata can lag
-          # a publish by a few seconds).
-          for attempt in 1 2 3 4 5; do
-            latest="$(npm view @gitlawb/zero dist-tags.latest || true)"
-            if [ "$latest" = "$version" ]; then
-              echo "dist-tag latest is ${version}."
-              exit 0
-            fi
-            sleep 5
-          done
-          echo "::error::dist-tag latest is ${latest}, expected ${version} after the wrapper publish." >&2
-          exit 1

.github/workflows/release-please.yml

@@ -1,35 +0,0 @@
-name: Release Please
-
-on:
-  push:
-    branches:
-      - main
-
-permissions:
-  contents: write # push the release-PR branch, create the tag + GitHub Release
-  pull-requests: write # open/update the release PR
-  issues: write # create/apply release-tracking labels on the release PR
-  actions: write # dispatch release-artifacts.yml once a release is cut
-
-jobs:
-  release-please:
-    name: Release PR / tag
-    runs-on: ubuntu-latest
-    steps:
-      - name: Run release-please
-        id: release
-        uses: googleapis/release-please-action@a02a34c4d625f9be7cb89156071d8567266a2445 # v4.2.0
-
-      # Tags created with the default GITHUB_TOKEN never fire `on: push: tags`
-      # workflows (GitHub suppresses workflows triggered by workflow-created
-      # events), so release-artifacts.yml would not run for a release-please
-      # tag. workflow_dispatch invoked through the API is exempt from that
-      # suppression — dispatch the artifacts build at the new tag explicitly.
-      - name: Build and publish release artifacts
-        if: ${{ steps.release.outputs.release_created }}
-        env:
-          GH_TOKEN: ${{ github.token }}
-        run: |
-          gh workflow run release-artifacts.yml \
-            --repo "$GITHUB_REPOSITORY" \
-            --ref "${{ steps.release.outputs.tag_name }}"

.github/workflows/zero-action-smoke.yml

@@ -1,105 +0,0 @@
-name: ZERO Action Smoke
-
-on:
-  pull_request:
-    paths:
-      - action.yml
-      - docs/GITHUB_ACTION.md
-      - .github/workflows/zero-action-smoke.yml
-  push:
-    branches:
-      - main
-    paths:
-      - action.yml
-      - docs/GITHUB_ACTION.md
-      - .github/workflows/zero-action-smoke.yml
-
-jobs:
-  validate:
-    name: Validate action.yml
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout
-        uses: actions/checkout@34e114876b0b11c390a56381ad16ebd13914f8d5 # v4
-        with:
-          persist-credentials: false
-
-      - name: Parse action.yml and assert documented inputs
-        shell: bash
-        run: |
-          python3 - <<'PY'
-          import sys, yaml
-
-          with open("action.yml", "r", encoding="utf-8") as handle:
-              action_raw = handle.read()
-          doc = yaml.safe_load(action_raw)
-
-          assert isinstance(doc, dict), "action.yml must be a mapping"
-          assert doc.get("name"), "action must declare a name"
-          assert doc.get("description"), "action must declare a description"
-
-          runs = doc.get("runs") or {}
-          assert runs.get("using") == "composite", "action must be a composite action"
-          steps = runs.get("steps") or []
-          assert len(steps) >= 2, "composite action must define steps"
-
-          inputs = doc.get("inputs") or {}
-          expected_inputs = [
-              "prompt", "prompt-file", "provider", "api-key", "api-key-env",
-              "model", "mode", "auto", "self-correct", "add-dir", "worktree",
-              "output-format", "post-to", "slack-webhook-url", "github-token",
-              "working-directory", "zero-version", "zero-repo",
-          ]
-          missing = [name for name in expected_inputs if name not in inputs]
-          assert not missing, f"action.yml is missing inputs: {missing}"
-
-          # Conservative CI default: autonomy must default to 'low'.
-          assert inputs["auto"].get("default") == "low", "auto must default to low"
-          # stream-json is the documented default capture format.
-          assert inputs["output-format"].get("default") == "stream-json", \
-              "output-format must default to stream-json"
-
-          outputs = doc.get("outputs") or {}
-          for name in ("exit-code", "output-file", "summary"):
-              assert name in outputs, f"action.yml is missing output: {name}"
-
-          # The action must never enable unsafe mode implicitly.
-          serialized = yaml.safe_dump(doc)
-          assert "--skip-permissions-unsafe" not in serialized, \
-              "action must never pass --skip-permissions-unsafe"
-
-          run_zero = next((step for step in steps if step.get("name") == "Run ZERO"), None)
-          assert run_zero is not None, "action.yml must contain a Run ZERO step"
-          run_zero_script = run_zero.get("run") or ""
-          assert 'export "${ZERO_INPUT_API_KEY_ENV}=${ZERO_INPUT_API_KEY}"' in run_zero_script, \
-              "Run ZERO must export the provider key for the current step"
-          assert "later steps inherit" not in action_raw.lower(), \
-              "provider keys must not be documented as inherited by later steps"
-          for line in run_zero_script.splitlines():
-              if "GITHUB_ENV" in line and (
-                  "ZERO_INPUT_API_KEY" in line or "ZERO_INPUT_API_KEY_ENV" in line
-              ):
-                  raise AssertionError("provider key must not be persisted to GITHUB_ENV")
-
-          print(f"action.yml OK: {len(inputs)} inputs, {len(outputs)} outputs, {len(steps)} steps")
-          PY
-
-      - name: Shell-syntax-check embedded run blocks
-        shell: bash
-        run: |
-          set -euo pipefail
-          python3 - <<'PY'
-          import yaml
-          with open("action.yml", "r", encoding="utf-8") as handle:
-              doc = yaml.safe_load(handle)
-          steps = (doc.get("runs") or {}).get("steps") or []
-          for index, step in enumerate(steps):
-              run = step.get("run")
-              if run:
-                  with open(f"/tmp/zero_step_{index}.sh", "w", encoding="utf-8") as out:
-                      out.write(run)
-          PY
-          for script in /tmp/zero_step_*.sh; do
-            echo "bash -n ${script}"
-            bash -n "${script}"
-          done

.gitignore

@@ -27,10 +27,6 @@ report.[0-9]_.[0-9]_.[0-9]_.[0-9]_.json
 .cache
 *.tsbuildinfo
 
-# Go build/module caches (GOCACHE/GOMODCACHE pointed here for local runs)
-.gocache/
-.gomodcache/
-
 # IntelliJ based IDEs
 .idea
 
@@ -41,17 +37,11 @@ report.[0-9]_.[0-9]_.[0-9]_.[0-9]_.json
 
 /zero
 /zero.exe
-/zero-windows-command-runner.exe
-/zero-windows-sandbox-setup.exe
-/zero-skills
-/.zero-binary-version
-/snake-game.html
 
-# Superpowers skill artifacts (brainstorm mockups, specs, plans) — local only
-.superpowers/
+# local launcher with API key — never commit
+start-zenline.sh
+
+# user test artifact — never commit
+shop.html
+docs/audit-2026-06-07-deep-audit.md
 docs/superpowers/
-
-# Termux/Android build artifacts
-zero-termux-arm64
-zero-linux-sandbox
-zero-seccomp

.release-please-manifest.json

@@ -1 +0,0 @@
-{".":"0.3.0"}

AGENTS.MD

@@ -0,0 +1,16 @@
+---
+description: Use Go-native project commands; Node is only for the npm wrapper.
+globs: "*.go, *.js, package.json, .github/workflows/*.yml, docs/*.md"
+alwaysApply: false
+---
+
+Default to Go commands in this repository.
+
+- Use `go run ./cmd/zero` to launch the local CLI.
+- Use `go test ./...` for the repository test suite.
+- Use `go run ./cmd/zero-release build` to build the release-facing binary.
+- Use `go run ./cmd/zero-release smoke` to smoke-test the built binary.
+- Use `go run ./cmd/zero-release package` and `go run ./cmd/zero-release verify` for release artifact checks.
+- Use `go run ./cmd/zero-perf-bench` for performance smoke work.
+- Do not add Bun or TypeScript build/test dependencies for repository validation.
+- Use Node only when directly exercising the npm wrapper at `bin/zero.js`.

AGENTS.md

@@ -1,27 +0,0 @@
----
-description: Project guidelines for the open-source CLI coding agent (zero) repository.
-globs: "*.go, *.js, *.md, *.json, *.toml, *.yaml, *.yml"
-alwaysApply: false
----
-
-# Repository Conventions for Zero
-
-This file outlines the project conventions and repository guidelines for coding agents when working on the `zero` repository. For the general guide on how to extend Zero (write specialist sub-agents, hooks, plugins, MCP, skills), see [docs/EXTENDING.md](docs/EXTENDING.md).
-
-## 1. Project Conventions
-
-- Build with `make`, not `go build` directly.
-- Tests live next to the source file (`foo_test.go` next to `foo.go`).
-- Run `make lint` before opening a PR.
-- Never edit files under `third_party/` — those are vendored.
-- Unify functions/methods where possible to prevent codebase inflation. Prefer a single cross-platform function with conditional checks over duplicating helpers per platform.
-
-## 2. Guidelines for Coding Agents
-
-If you are an AI coding agent executing tasks in this repository, you **MUST** run all Go code quality and security checks before committing code or completing your task:
-
-1. **Format & Vet**: Run `go fmt ./...` and `go vet ./...`.
-2. **Lint**: Run `go run github.com/golangci/golangci-lint/v2/cmd/[email protected] run --enable-only unused,ineffassign,staticcheck ./...`.
-3. **Security**: Run `go run golang.org/x/vuln/cmd/[email protected] ./...`.
-
-If any check fails or cannot be run, do not ignore it. Prompt the user for instructions or setup assistance before proceeding.

CHANGELOG.md

@@ -1,161 +0,0 @@
-# Changelog
-
-All notable changes to this project are documented in this file.
-
-The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project
-aims to follow [Semantic Versioning](https://semver.org/spec/v2.0.0.html) once the first release is
-tagged. Until then, source builds report the version `dev`.
-
-## [0.3.0](https://github.com/Gitlawb/zero/compare/v0.2.0...v0.3.0) (2026-07-09)
-
-
-### Features
-
-* gate project-scoped hooks, plugins, and MCP servers behind workspace trust ([#529](https://github.com/Gitlawb/zero/issues/529)) ([a880ce8](https://github.com/Gitlawb/zero/commit/a880ce80a6ec72da511fb9bdf6dd69291c72a64b))
-* **modelregistry:** infer reasoning efforts for Hunyuan and vendor-prefixed model ids ([#599](https://github.com/Gitlawb/zero/issues/599)) ([92a92ce](https://github.com/Gitlawb/zero/commit/92a92ceb29dc63f5216ab140ef9a6dd9afe17df8))
-* **providers:** OAuth login profiles and list-first /provider manager ([#560](https://github.com/Gitlawb/zero/issues/560)) ([1655056](https://github.com/Gitlawb/zero/commit/16550569c1be615cfaf244dab25909ec37f6dee6))
-* **tui:** remember recent provider+model selections in /model picker ([#568](https://github.com/Gitlawb/zero/issues/568)) ([d0c4e62](https://github.com/Gitlawb/zero/commit/d0c4e62cd429a0614c15296934c740a08bc0e07b))
-* **tui:** show CLI version on the startup home screen ([#538](https://github.com/Gitlawb/zero/issues/538)) ([fd69233](https://github.com/Gitlawb/zero/commit/fd69233e334f1823a06b5794085a9255b3abdfa8))
-* voice dictation (speech-to-text) ([#557](https://github.com/Gitlawb/zero/issues/557)) ([87158a1](https://github.com/Gitlawb/zero/commit/87158a1c90b4f91fc5f2bb8178ebaf46d7654680))
-
-
-### Bug Fixes
-
-* address bugs found in a multi-agent codebase audit ([#481](https://github.com/Gitlawb/zero/issues/481)) ([008bc9b](https://github.com/Gitlawb/zero/commit/008bc9b3f3ba13c7d4822b9559b020f381ff555b))
-* **agent:** keep tools exposed for max-turn finalization ([#533](https://github.com/Gitlawb/zero/issues/533)) ([3f0503b](https://github.com/Gitlawb/zero/commit/3f0503bc2312ae29d5ade784d8824dc9a3524958))
-* **auth:** persist OpenRouter API key after login ([#595](https://github.com/Gitlawb/zero/issues/595)) ([2a062aa](https://github.com/Gitlawb/zero/commit/2a062aa4014c6cd5e20a57dad4a685f88966f109))
-* bump Go to 1.26.5 for crypto/tls fix (GO-2026-5856) ([#607](https://github.com/Gitlawb/zero/issues/607)) ([a7cfb99](https://github.com/Gitlawb/zero/commit/a7cfb99fed7b88ebc09a2f251cb82864d3c2cade))
-* gitignore Windows sandbox helpers and npm version marker ([#578](https://github.com/Gitlawb/zero/issues/578)) ([25653f6](https://github.com/Gitlawb/zero/commit/25653f686c016f95b81ceb7ff5d5452d37c4d4f3))
-* **mcp:** silence startup warning for unconfigured default servers ([#563](https://github.com/Gitlawb/zero/issues/563)) ([302f58b](https://github.com/Gitlawb/zero/commit/302f58bb5f2a03ec7230354ed4747e4e55c16c50))
-* **mcp:** skip RFC 8414 discovery when OAuth endpoints are preconfigured ([#586](https://github.com/Gitlawb/zero/issues/586)) ([8a52d98](https://github.com/Gitlawb/zero/commit/8a52d98cad7cd0086dee9aede4ce477e432bd385))
-* **modelregistry:** reject oversized models.dev cache responses ([#602](https://github.com/Gitlawb/zero/issues/602)) ([66a6396](https://github.com/Gitlawb/zero/commit/66a63964149fb2f07e646e5f1987627c5cd9ac28))
-* **provider:** stop dropping custom no-auth providers on restart ([#558](https://github.com/Gitlawb/zero/issues/558)) ([ba99fa8](https://github.com/Gitlawb/zero/commit/ba99fa8d487fb28f2700e0ff10b2a25c75303cf7))
-* **sandbox:** Windows-appropriate suggestions when blocking interactive commands ([#414](https://github.com/Gitlawb/zero/issues/414)) ([ba4c007](https://github.com/Gitlawb/zero/commit/ba4c00755dc1c31b3dcca18e50b20f62c6bf5d1f))
-* **tools:** block MSYS and WSL shells under the Windows sandbox ([#587](https://github.com/Gitlawb/zero/issues/587)) ([0666818](https://github.com/Gitlawb/zero/commit/066681855b80e3baf5d07d7397610b25f724e353))
-* **tools:** platform-specific pager suggestions, quote/caret-safe cd detection ([#543](https://github.com/Gitlawb/zero/issues/543)) ([8b248f4](https://github.com/Gitlawb/zero/commit/8b248f4e1198dc86ab332a697fba4cf520823cbd))
-* **tools:** Windows cmd.exe quoting guidance and clipboard escaping fix ([#468](https://github.com/Gitlawb/zero/issues/468)) ([f10ed0c](https://github.com/Gitlawb/zero/commit/f10ed0c893ce6de08923f143d681ba96f0fcfe3a))
-* **tui:** bypass toggleSidebar and toggleMouse global shortcuts when composer is non-empty ([#576](https://github.com/Gitlawb/zero/issues/576)) ([c7346fb](https://github.com/Gitlawb/zero/commit/c7346fbcf01fe7e70ed2fcfccbf9965e5985727b))
-* **tui:** paste protection for Termux char-by-char paste ([#573](https://github.com/Gitlawb/zero/issues/573)) ([8e9149f](https://github.com/Gitlawb/zero/commit/8e9149f1a296bebebf95cae9dd2a7c5156c9dbb6))
-* **tui:** update picker_test for switchProviderModel 4-value return ([#589](https://github.com/Gitlawb/zero/issues/589)) ([8f15650](https://github.com/Gitlawb/zero/commit/8f156506dc92449bd24caa14e36f28477ba00fff))
-* **update:** clearer error on unsupported release platform (android/termux) ([#603](https://github.com/Gitlawb/zero/issues/603)) ([1fc9b2d](https://github.com/Gitlawb/zero/commit/1fc9b2d25c79e089f34cb7b5b6a7f7c7b8233123))
-* **update:** support safe symlink extraction during updates ([#575](https://github.com/Gitlawb/zero/issues/575)) ([ce9cb91](https://github.com/Gitlawb/zero/commit/ce9cb912958a3d9ae0b052bfebb849c28e0e719b))
-* warn about untracked scratch files left behind after a run ([#571](https://github.com/Gitlawb/zero/issues/571)) ([062328b](https://github.com/Gitlawb/zero/commit/062328b632a2d27353a6d47c522bbd22d7282539))
-
-
-### Performance Improvements
-
-* **grep:** stop content scan after head limit ([#601](https://github.com/Gitlawb/zero/issues/601)) ([8a05e64](https://github.com/Gitlawb/zero/commit/8a05e6486f7a63281b869064db03dfc5531e6a04))
-
-## [0.2.0](https://github.com/Gitlawb/zero/compare/v0.1.0...v0.2.0) (2026-07-06)
-
-
-### Features
-
-* add --auto flag for LLM-generated commit messages ([#423](https://github.com/Gitlawb/zero/issues/423)) ([b0abde7](https://github.com/Gitlawb/zero/commit/b0abde7d0697e808480cd59d69a6f4d0c6320475))
-* add zero changes push and pr subcommands, and extra repo-info metrics ([#391](https://github.com/Gitlawb/zero/issues/391)) ([2312abe](https://github.com/Gitlawb/zero/commit/2312abe5ddd95f4c6ef373cfb61cc03092f48cdd))
-* agent quality, caching, retry, and tooling upgrades ([#506](https://github.com/Gitlawb/zero/issues/506)) ([3c81fea](https://github.com/Gitlawb/zero/commit/3c81fea22873ee3df7fc97b10cb4f77792706c4b))
-* **agent:** curb over-engineering the solution in the editing discipline ([#517](https://github.com/Gitlawb/zero/issues/517)) ([f4c998a](https://github.com/Gitlawb/zero/commit/f4c998ac30c4f07ff313a2d706791e857293be49)), closes [#516](https://github.com/Gitlawb/zero/issues/516)
-* **agent:** inject per-user config.UserConfigDir()/zero/ZERO.md guidelines into system prompt ([#475](https://github.com/Gitlawb/zero/issues/475)) ([7b10aab](https://github.com/Gitlawb/zero/commit/7b10aab74bf14a01166b2cea22deab79bba9850b))
-* **openai:** forward prompt_cache_key for server-side prefix cache routing ([#515](https://github.com/Gitlawb/zero/issues/515)) ([87e7e69](https://github.com/Gitlawb/zero/commit/87e7e69afd18b5539579856f3a61c6a95bc445ae))
-* **providers:** add `zero providers models` to discover a provider's models ([#386](https://github.com/Gitlawb/zero/issues/386)) ([0bc8074](https://github.com/Gitlawb/zero/commit/0bc8074c97b0310e4a9d70c3f967003ee5e8a59f))
-* **providers:** add KiloCode and OpenCode provider support ([#388](https://github.com/Gitlawb/zero/issues/388)) ([b1ccb6d](https://github.com/Gitlawb/zero/commit/b1ccb6d9c1875377f5e5ea81a1304edd1e41ab4f))
-* **providers:** add Meituan LongCat catalog preset ([#424](https://github.com/Gitlawb/zero/issues/424)) ([b4275e3](https://github.com/Gitlawb/zero/commit/b4275e350472b2490212bf814709819d354c1216))
-* **providers:** split minimax zai into intl cn ([#398](https://github.com/Gitlawb/zero/issues/398)) ([aaad4d2](https://github.com/Gitlawb/zero/commit/aaad4d271270f41af837b6f3b60ae80beba0c645))
-* require manual approval before npm publish + drop release-as pin ([#369](https://github.com/Gitlawb/zero/issues/369)) ([bd89a1f](https://github.com/Gitlawb/zero/commit/bd89a1f451643c1b65ec803070abc7b116631ebe))
-* **sandbox:** unelevated Windows fallback tier instead of prompts-only degrade ([#427](https://github.com/Gitlawb/zero/issues/427)) ([b9ddd6f](https://github.com/Gitlawb/zero/commit/b9ddd6f42138312a1fee8d8bb67c46c8eb1dea2f))
-* support shift enter for composer newlines ([#462](https://github.com/Gitlawb/zero/issues/462)) ([daf65e0](https://github.com/Gitlawb/zero/commit/daf65e0af9a040314d4ab337b0ad59c55416b7bc))
-* **tui:** /loop — repeat a prompt or command on an interval or self-paced ([#502](https://github.com/Gitlawb/zero/issues/502)) ([387fe67](https://github.com/Gitlawb/zero/commit/387fe67ee7cd81317c9c969f5906a4437080fea3))
-* **tui:** add search/filter to provider picker in setup wizard ([#400](https://github.com/Gitlawb/zero/issues/400)) ([2fcea71](https://github.com/Gitlawb/zero/commit/2fcea71778d23e050c93409c471aef45b68c1621))
-* **update:** add zero upgrade command to apply self-updates ([#461](https://github.com/Gitlawb/zero/issues/461)) ([5f36349](https://github.com/Gitlawb/zero/commit/5f36349c1884e81fa9bc66bb5fe813b627e897b7))
-
-
-### Bug Fixes
-
-* **action:** keep provider key scoped to zero step ([#448](https://github.com/Gitlawb/zero/issues/448)) ([407a927](https://github.com/Gitlawb/zero/commit/407a92739ff508cba32d2c12b3f36f0efcdd54c3))
-* add android platform support for Termux npm install ([#455](https://github.com/Gitlawb/zero/issues/455)) ([9bd93c6](https://github.com/Gitlawb/zero/commit/9bd93c62f8d57fb74057284aa66a1b6e1429dcdd)), closes [#449](https://github.com/Gitlawb/zero/issues/449)
-* **agent:** reject a malformed additional_permissions payload before prompting ([#453](https://github.com/Gitlawb/zero/issues/453)) ([e4f760e](https://github.com/Gitlawb/zero/commit/e4f760ee8bd57299cd2fcb37e8e23130037c2607))
-* allow non-TLS connections to private-network provider endpoints ([#444](https://github.com/Gitlawb/zero/issues/444)) ([1d86384](https://github.com/Gitlawb/zero/commit/1d8638466ca31517eb9db2b9353d3dce1cbeeabc))
-* **auth:** route zero auth login chatgpt to the dedicated ChatGPT flow ([#443](https://github.com/Gitlawb/zero/issues/443)) ([305a62c](https://github.com/Gitlawb/zero/commit/305a62c954ca6cec00bc58d5398f933415156aff))
-* **config:** fall back to a usable saved provider instead of forcing full re-onboarding ([#410](https://github.com/Gitlawb/zero/issues/410)) ([c60ad87](https://github.com/Gitlawb/zero/commit/c60ad8729f79bb841114d352ee2d2fe29d5d0e41))
-* **config:** let a gateway ANTHROPIC_BASE_URL resolve as anthropic-compatible ([#497](https://github.com/Gitlawb/zero/issues/497)) ([30dd7c3](https://github.com/Gitlawb/zero/commit/30dd7c3112ad22d42fa12b5addd4e38f4beda42a)), closes [#479](https://github.com/Gitlawb/zero/issues/479)
-* **config:** unbrick first-run setup — default google/anthropic models, enter setup on fixable config errors ([#385](https://github.com/Gitlawb/zero/issues/385)) ([72eed06](https://github.com/Gitlawb/zero/commit/72eed06b4f94c43d75d31fe54a58d2f566de059e))
-* **config:** use ~/.config on macOS and enter setup when no provider ([#371](https://github.com/Gitlawb/zero/issues/371)) ([#372](https://github.com/Gitlawb/zero/issues/372)) ([027a8f2](https://github.com/Gitlawb/zero/commit/027a8f2768b17b89f5c8270887f156e2ccda69ea))
-* **docs:** rename AGENTS.MD > AGENTS.md ([#438](https://github.com/Gitlawb/zero/issues/438)) ([4266baf](https://github.com/Gitlawb/zero/commit/4266baf222df583ed2078b776687f12d496475b5))
-* **gemini:** strip unsupported JSON Schema fields from tool declarations ([#374](https://github.com/Gitlawb/zero/issues/374)) ([39e7100](https://github.com/Gitlawb/zero/commit/39e7100674150144a1152e3110c64c7cf0321d64)), closes [#373](https://github.com/Gitlawb/zero/issues/373)
-* **install:** persist install dir to user PATH on Windows ([#407](https://github.com/Gitlawb/zero/issues/407)) ([bdb1b0e](https://github.com/Gitlawb/zero/commit/bdb1b0ecd15859b1712a6037d296dace7f9c3c3f))
-* **mcp:** block cross-origin credential redirects ([#396](https://github.com/Gitlawb/zero/issues/396)) ([f915f70](https://github.com/Gitlawb/zero/commit/f915f70e5a3096e2419fa8d961a0f84a626fa4a9))
-* **oauth:** treat Windows ERROR_ACCESS_DENIED as lock contention in createSecretFile ([#445](https://github.com/Gitlawb/zero/issues/445)) ([d05e914](https://github.com/Gitlawb/zero/commit/d05e9148a7f79f67d1d3c31fca2775f21fbd331e))
-* **openai:** handle Ollama reasoning stream deltas ([#486](https://github.com/Gitlawb/zero/issues/486)) ([f6c0606](https://github.com/Gitlawb/zero/commit/f6c060631e18e082dda24cc4dc0903c31c2120d6))
-* preserve conversation context in exec prompts ([#460](https://github.com/Gitlawb/zero/issues/460)) ([949ee43](https://github.com/Gitlawb/zero/commit/949ee43f71e5cb7fab4695c5cb7b442fe4ecfbf7))
-* **provider-wizard:** allow multiple custom OpenAI-compatible providers ([#403](https://github.com/Gitlawb/zero/issues/403)) ([3fbbd28](https://github.com/Gitlawb/zero/commit/3fbbd28e4c586822cc4312c86232d94befe56e87))
-* **sandbox:** fix nested pipe creation under the Windows restricted token ([#456](https://github.com/Gitlawb/zero/issues/456)) ([563a6db](https://github.com/Gitlawb/zero/commit/563a6dbe91e65d5daeefd7626e8a77e30a6d8fb2))
-* **sandbox:** gate /tmp test assertions on GOOS, not path existence ([#426](https://github.com/Gitlawb/zero/issues/426)) ([f653dca](https://github.com/Gitlawb/zero/commit/f653dcac363fb69ad7be5b35e6e0fa6d2bce476d))
-* **sandbox:** self-heal a corrupt unelevated setup marker ([#437](https://github.com/Gitlawb/zero/issues/437)) ([8d0c5fe](https://github.com/Gitlawb/zero/commit/8d0c5feccb8bdbfb015df0508aa6e3bcbd1fd0e8))
-* **specialist:** cap max specialist nesting depth ([#491](https://github.com/Gitlawb/zero/issues/491)) ([177442c](https://github.com/Gitlawb/zero/commit/177442cfe4015bd8df04cc9894f98b468ee796d4))
-* Termux/Android support — PRoot scroll, SIGSYS sandbox, build docs ([#509](https://github.com/Gitlawb/zero/issues/509)) ([0f69d99](https://github.com/Gitlawb/zero/commit/0f69d995e9b586b774f66c066b21abab5e03024a))
-* **tools:** Block MSYS coreutils under Windows sandbox ([#476](https://github.com/Gitlawb/zero/issues/476)) ([81aad58](https://github.com/Gitlawb/zero/commit/81aad58d97839e51c068e2f08907618991fdc3fb))
-* **tools:** CRLF line ending mismatch in edit_file tool on Windows ([#378](https://github.com/Gitlawb/zero/issues/378)) ([33dc7ae](https://github.com/Gitlawb/zero/commit/33dc7ae2cc82c5389675531e1416856dae7151ce))
-* **tools:** fix cmd.exe /S/C corrupting commands with embedded quotes ([#465](https://github.com/Gitlawb/zero/issues/465)) ([190241b](https://github.com/Gitlawb/zero/commit/190241bd593f43211b766e0b13c8e89802d4bb37))
-* **tools:** flag piped POSIX utilities before running on Windows ([#412](https://github.com/Gitlawb/zero/issues/412)) ([5658a36](https://github.com/Gitlawb/zero/commit/5658a366274fc59a9d5336b06a21019c9c25cbf1))
-* **tools:** make grep and glob respect run cancellation ([#464](https://github.com/Gitlawb/zero/issues/464)) ([ba6c026](https://github.com/Gitlawb/zero/commit/ba6c0264697b7d7ed479f6e782fba9700a481e3d))
-* **tools:** require permission before web_search requests ([#382](https://github.com/Gitlawb/zero/issues/382)) ([960db96](https://github.com/Gitlawb/zero/commit/960db9660e4e31dc588fe8f7d6f116ff5e225566))
-* **tui:** compose help overlay through the viewport overlay pipeline ([#421](https://github.com/Gitlawb/zero/issues/421)) ([5b2b4de](https://github.com/Gitlawb/zero/commit/5b2b4dea1aaf9e0f68baa25e97e83296fb17b1a2))
-* **tui:** keep the profile name on /model switch so the stored key resolves ([#441](https://github.com/Gitlawb/zero/issues/441)) ([9134148](https://github.com/Gitlawb/zero/commit/9134148f4df3e4e556fba6c2f8babfdf6fcfeee1)), closes [#440](https://github.com/Gitlawb/zero/issues/440)
-* **tui:** resolve every permission request so the agent can't deadlock ([#397](https://github.com/Gitlawb/zero/issues/397)) ([952788f](https://github.com/Gitlawb/zero/commit/952788f72d32957659fe004521fcc8372b9ba9b4))
-* **tui:** show an M suffix for million-scale token counts ([#457](https://github.com/Gitlawb/zero/issues/457)) ([0562e3b](https://github.com/Gitlawb/zero/commit/0562e3bef7df2328610a48a1e81632a8da4aec64))
-* **tui:** title /model rows by model name, not the catalog description ([#395](https://github.com/Gitlawb/zero/issues/395)) ([cdf9d83](https://github.com/Gitlawb/zero/commit/cdf9d839ae57a729f292f36f7c5b0c67b41b288d))
-
-
-### Performance Improvements
-
-* cache TUI model registry ([#496](https://github.com/Gitlawb/zero/issues/496)) ([e7d88b4](https://github.com/Gitlawb/zero/commit/e7d88b4b518049733da25a8447c00144bd1da518))
-* universal tool-output ceiling with spill + async post-edit diagnostics ([#518](https://github.com/Gitlawb/zero/issues/518)) ([95ccd5b](https://github.com/Gitlawb/zero/commit/95ccd5bc327f6fb464ff0239f7229de789f578dc))
-
-## 0.1.0 (2026-07-02)
-
-
-### Features
-
-* publish zero to npm via release-please ([#367](https://github.com/Gitlawb/zero/issues/367)) ([8eccc26](https://github.com/Gitlawb/zero/commit/8eccc2669887bc38d35bc16a315c888e4d9ec43a))
-* **tui:** FILES sidebar panel with click-to-select and file drill-in ([#365](https://github.com/Gitlawb/zero/issues/365)) ([142c548](https://github.com/Gitlawb/zero/commit/142c548c89a8652ce300e64ddf1228ee36df7606))
-
-
-### Bug Fixes
-
-* **auth:** propagate credentials to every provider-build surface and pin children to the live provider ([#366](https://github.com/Gitlawb/zero/issues/366)) ([6e0a665](https://github.com/Gitlawb/zero/commit/6e0a665118fe0e09c4b07d482dd18f86045acd2b))
-
-## [Unreleased]
-
-### Added
-- `SECURITY.md` with a private vulnerability-reporting path, `CODE_OF_CONDUCT.md`, this changelog, and
-  GitHub issue/PR templates.
-- Interactive `/theme` picker: bare `/theme` opens a popup that live-previews each palette as you move
-  and applies on select (Esc reverts).
-- Ten built-in color themes alongside the `dark`/`light` built-ins — `dracula`, `nord`, `gruvbox`,
-  `tokyo-night`, `catppuccin`, `one-dark`, `solarized-dark`, `rose-pine`, `everforest`, and
-  `solarized-light` — selectable via `/theme <name>`, `--theme <name>`, or `ZERO_THEME`. Every palette
-  is contrast-audited to WCAG AA. The built-in light theme was reworked for legibility.
-- `--theme <name>` flag for the TUI, accepting `auto` or any registered theme (previously only the
-  `ZERO_THEME` env var existed).
-- "Accessibility / Appearance" section in the README documenting `NO_COLOR`, `ZERO_THEME`, `/theme`,
-  and `ZERO_NO_FADE`.
-
-### Changed
-- Provider connectivity health checks now allow loopback hosts for explicitly user-configured local
-  providers (Ollama / LM Studio), so the keyless local-model path verifies instead of failing with
-  "localhost hosts are blocked". The SSRF guard for fetched/remote URLs is unchanged.
-- Auth (401/403) errors now show a curated, actionable message pointing at `zero auth` / setup; the
-  raw upstream body is shown only under a verbose/debug flag.
-- No-provider / missing-key errors now point at `zero setup` and `zero auth`, and distinguish a
-  missing key from a rejected key.
-- `zero doctor` no longer reports "Overall: pass" when no provider credential is configured, and
-  formats the missing-language-server list for humans (no raw Go `map[...]`).
-- Raised the `faint`/`faintest` theme tokens (and the light-theme accent) to meet WCAG AA contrast for
-  the content they carry.
-- `NO_COLOR` is now honored for any non-empty value, per the no-color.org spec.
-
-### Removed
-- The inert `/input-style` slash command (it had no backend).
-
-### Fixed
-- README/`go.mod` Go-version mismatch and other stale public-release docs claims.

CODE_OF_CONDUCT.md

@@ -1,58 +0,0 @@
-# Contributor Covenant Code of Conduct
-
-## Our Pledge
-
-We as members, contributors, and leaders pledge to make participation in our community a
-harassment-free experience for everyone, regardless of age, body size, visible or invisible
-disability, ethnicity, sex characteristics, gender identity and expression, level of experience,
-education, socio-economic status, nationality, personal appearance, race, religion, or sexual
-identity and orientation.
-
-We pledge to act and interact in ways that contribute to an open, welcoming, diverse, inclusive, and
-healthy community.
-
-## Our Standards
-
-Examples of behavior that contributes to a positive environment:
-
-- Demonstrating empathy and kindness toward other people.
-- Being respectful of differing opinions, viewpoints, and experiences.
-- Giving and gracefully accepting constructive feedback.
-- Accepting responsibility and apologizing to those affected by our mistakes.
-- Focusing on what is best for the overall community.
-
-Examples of unacceptable behavior:
-
-- The use of sexualized language or imagery, and sexual attention or advances of any kind.
-- Trolling, insulting or derogatory comments, and personal or political attacks.
-- Public or private harassment.
-- Publishing others' private information, such as a physical or email address, without their
-  explicit permission.
-- Other conduct which could reasonably be considered inappropriate in a professional setting.
-
-## Enforcement Responsibilities
-
-Project maintainers are responsible for clarifying and enforcing our standards of acceptable behavior
-and will take appropriate and fair corrective action in response to any behavior that they deem
-inappropriate, threatening, offensive, or harmful.
-
-## Scope
-
-This Code of Conduct applies within all community spaces, and also applies when an individual is
-officially representing the community in public spaces.
-
-## Enforcement
-
-Instances of abusive, harassing, or otherwise unacceptable behavior may be reported privately to the
-project maintainers through the repository's **Security → Report a vulnerability** channel (which
-delivers a confidential report to maintainers), or by contacting a maintainer directly. All
-complaints will be reviewed and investigated promptly and fairly.
-
-Maintainers are obligated to respect the privacy and security of the reporter of any incident.
-
-## Attribution
-
-This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 2.1, available at
-<https://www.contributor-covenant.org/version/2/1/code_of_conduct.html>.
-
-[homepage]: https://www.contributor-covenant.org

LICENSE

@@ -1,21 +0,0 @@
-MIT License
-
-Copyright (c) 2026 Gitlawb
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.

Makefile

@@ -1,44 +0,0 @@
-# Zero build/test/lint targets. AGENTS.md says "Build with `make`" and "Run `make
-# lint` before opening a PR" — these targets back those instructions.
-.DEFAULT_GOAL := build
-.PHONY: build build-all test test-race vet fmt fmt-check lint tidy clean help
-
-# Build the main CLI binary into ./zero.
-build:
-	go build -o zero ./cmd/zero
-
-# Build every command in cmd/.
-build-all:
-	go build ./...
-
-# Run the full test suite with the race detector (matches CI expectations).
-test:
-	go test ./... -race -count=1
-
-# Faster, no race detector.
-test-quick:
-	go test ./...
-
-vet:
-	go vet ./...
-
-fmt:
-	gofmt -w $(shell git ls-files '*.go')
-
-# Fail if any tracked Go file is not gofmt-clean.
-fmt-check:
-	@out="$$(gofmt -l $$(git ls-files '*.go'))"; \
-	if [ -n "$$out" ]; then echo "gofmt needed:"; echo "$$out"; exit 1; fi
-
-# Lint = formatting check + vet (no extra tooling required).
-lint: fmt-check vet
-
-tidy:
-	go mod tidy
-
-clean:
-	rm -f zero
-	go clean ./...
-
-help:
-	@echo "Targets: build (default), build-all, test, test-quick, vet, fmt, fmt-check, lint, tidy, clean"

README.md

@@ -1,398 +1,217 @@
-<p align="center">
-  <img src="docs/assets/zero-logo.png" alt="Zero" width="385">
-</p>
+```
+ ███████████ ██████████ ███████████      ███████
+▒█▒▒▒▒▒▒███ ▒▒███▒▒▒▒▒█▒▒███▒▒▒▒▒███   ███▒▒▒▒▒███
+▒     ███▒   ▒███  █ ▒  ▒███    ▒███  ███     ▒▒███
+     ███     ▒██████    ▒██████████  ▒███      ▒███
+    ███      ▒███▒▒█    ▒███▒▒▒▒▒███ ▒███      ▒███
+  ████     █ ▒███ ▒   █ ▒███    ▒███ ▒▒███     ███
+ ███████████ ██████████ █████   █████ ▒▒▒███████▒
+▒▒▒▒▒▒▒▒▒▒▒ ▒▒▒▒▒▒▒▒▒▒ ▒▒▒▒▒   ▒▒▒▒▒    ▒▒▒▒▒▒▒
+```
 
-<p align="center"><strong>A terminal coding agent you own.</strong></p>
+# Zero
 
-<p align="center">
-  <a href="LICENSE"><img alt="license" src="https://img.shields.io/badge/license-MIT-blue"></a>
-  <img alt="Go 1.26.5+" src="https://img.shields.io/badge/Go-1.26.5+-00ADD8?logo=go&logoColor=white">
-  <img alt="25+ providers" src="https://img.shields.io/badge/providers-25+-34E2EA">
-  <br>
-  <strong>English</strong> | <a href="README_ZH.md">中文</a>
-</p>
+**A clean, terminal-first AI coding agent you fully own — multi-provider, scriptable, and safe by default.**
 
-Zero is an AI coding agent for your local terminal. It can inspect a repository,
-edit files, run commands, use browser/terminal helpers, and keep durable local
-sessions while you choose the model and the permission level.
+![core](https://img.shields.io/badge/core-Go--native-00ADD8?logo=go&logoColor=white)
+![distribution](https://img.shields.io/badge/distribution-npm%20wrapper-cb3837?logo=npm&logoColor=white)
+![transition](https://img.shields.io/badge/Go%20runtime-default-14151a?logo=go&logoColor=white)
+![status](https://img.shields.io/badge/status-active%20development-67e8f9)
+
+Zero is a coding agent that lives in your terminal. It runs an agentic tool loop —
+reading, editing, searching, and running commands in your repo — against **whatever
+model you choose**. Zero has a **Go-native core** with an npm distribution wrapper;
+the Go runtime is the default app path.
+
+> Zero treats the **model as a swappable, per-task choice** — no single-vendor lock-in —
+> and never mutates your system without a permission decision.
+
+---
+
+## Highlights
+
+- 🔌 **Multi-provider** — OpenAI-compatible, Anthropic, and Gemini behind one interface, with a model registry (capabilities, context limits, cost). Bring your own key and endpoint.
+- 🖥️ **Premium TUI** — Bubble Tea/Lip Gloss powers the Go-native terminal interface.
+- 🤖 **Headless & scriptable** — `zero exec` with clean `text` / `json` / `stream-json` I/O and meaningful exit codes for CI and automation.
+- 🧰 **Real tools** — read / write / edit files, `apply_patch`, `grep`, `glob`, `bash`, directory listing, and a live plan/todo.
+- 🛡️ **Safe by default** — mutating tools are permission-gated; `--skip-permissions-unsafe` is an explicit, clearly-labeled opt-out.
+- 💾 **Durable sessions** — local, append-only session event store with full-text `search`.
+- 🩺 **Operable** — built-in `doctor`, `config` inspection, secret redaction everywhere, and `update --check`.
+
+## Quick start
+
+> Requires [Go](https://go.dev/) for local development and release tooling.
+> The npm wrapper uses Node when Zero is installed as an npm package.
 
 ```bash
-zero
-zero exec "fix the failing test in ./pkg"
-zero exec --output-format stream-json < turns.jsonl
+go run ./cmd/zero    # launch the interactive TUI
 ```
 
-## Why Zero
-
-- **Use the model you want.** Bring OpenAI, Anthropic, Gemini, Groq, OpenRouter,
-  DeepSeek, Mistral, xAI, Qwen, Kimi, GitHub Models, Ollama, LM Studio, or any
-  OpenAI-/Anthropic-compatible endpoint.
-- **Stay in control.** File writes, shell commands, network access, and
-  out-of-workspace writes go through Zero's permission and sandbox policy.
-- **Works in the terminal.** The TUI has model/provider pickers, image input,
-  slash commands, live plan/tool rendering, scrollback, themes, and resume/fork
-  support.
-- **Works without the TUI.** `zero exec` is scriptable, supports text/JSON/
-  stream-JSON I/O, isolated worktrees, spec-first runs, and meaningful exit
-  codes for CI.
-- **Keeps context local.** Sessions are stored on disk, searchable, resumable,
-  and never uploaded as telemetry by Zero.
-- **Extensible when you need it.** Use MCP servers, skills, plugins, hooks, and
-  specialist subagents from the same CLI.
-
-## Install
-
-### npm
-
-```bash
-npm install -g @gitlawb/zero
-zero
-```
-
-The npm package is a small wrapper whose platform build (Linux and macOS on
-x64/arm64, Windows on x64 — including the browser/terminal control helpers)
-installs as an optional dependency straight from the npm registry — no install
-scripts, no downloads outside npm. Bun, pnpm, and yarn work the same way with
-no trust or approval steps. Installs that skip optional dependencies
-(`--omit=optional`) still work: the wrapper fetches the binary from the
-matching GitHub Release whenever it is missing. Windows on ARM runs the x64
-build under emulation. See [docs/NPM_PACKAGING.md](docs/NPM_PACKAGING.md) for
-how the package is put together.
-
-### Install scripts
-
-Linux/macOS:
-
-```bash
-curl -fsSL https://raw.githubusercontent.com/Gitlawb/zero/main/scripts/install.sh | bash
-```
-
-Windows PowerShell:
-
-```powershell
-irm https://raw.githubusercontent.com/Gitlawb/zero/main/scripts/install.ps1 | iex
-```
-
-### From source
-
-Source builds require Go 1.26.5+.
-
-```bash
-git clone https://github.com/Gitlawb/zero.git
-cd zero
-go run ./cmd/zero
-```
-
-Release installers and the npm wrapper require published GitHub Release assets.
-If you are testing before the first public release, build from source:
-
-```bash
-go build -o zero ./cmd/zero
-```
-
-On Linux, build the sandbox helper too if you want native sandboxing:
-
-```bash
-go build -o zero-linux-sandbox ./cmd/zero-linux-sandbox
-go build -o zero-seccomp ./cmd/zero-seccomp   # optional compatibility wrapper
-```
-
-Put `zero` and `zero-linux-sandbox` in the same directory on `PATH`
-(`~/.local/bin` is a good default). macOS does not need an extra helper binary.
-Windows source builds can use the main `zero.exe` as their sandbox helper; release
-archives still ship standalone Windows helper executables.
-
-More install details: [docs/INSTALL.md](docs/INSTALL.md).
-
-## First Run
-
-Start the TUI:
-
-```bash
-zero
-```
-
-The setup wizard helps you pick a provider and model. You can also configure
-providers from the command line:
-
-```bash
-zero setup
-zero providers list
-zero models list
-zero doctor
-```
-
-For API providers, set the matching environment variable before setup or enter
-the key in the wizard:
+Point Zero at a model — either set environment variables:
 
 ```bash
 export OPENAI_API_KEY=sk-...
-export ANTHROPIC_API_KEY=...
-export GEMINI_API_KEY=...
-export LONGCAT_API_KEY=...
-export MINIMAX_API_KEY=...
-export MINIMAXI_API_KEY=...
+# optional: any OpenAI-compatible endpoint / model
+export OPENAI_BASE_URL=https://api.openai.com/v1
+export OPENAI_MODEL=gpt-4.1
 ```
 
-To configure Meituan LongCat (LongCat-2.0) directly, run:
+…or save reusable provider profiles in `~/.config/zero/config.json` and manage them with
+`zero providers`. Run `zero doctor` anytime to verify your setup.
+
+## Usage
+
+### Interactive (TUI)
 
 ```bash
-zero providers setup longcat --set-active
+go run ./cmd/zero    # or: zero after installation
 ```
 
-MiniMax presets use the Anthropic-compatible endpoints for the global and China
-regions:
+Inside the TUI: type to chat and press **Enter** to send. `/` opens command suggestions
+(**Tab** accepts the first). When the prompt is empty, the arrow keys, **PgUp/PgDn**, and
+**Home/End** scroll the transcript. **Ctrl+C** exits.
+
+### Headless (`exec`)
 
 ```bash
-zero providers add minimax --set-active
-zero providers add minimaxi-cn --set-active
-```
+# one-shot
+zero exec "explain internal/agent/loop.go and suggest one improvement"
 
-To use the OpenAI-compatible endpoints instead, add a custom compatible profile
-for the required region:
+# from a file, with a specific model, as JSON for scripts
+zero exec --file task.md --model claude-sonnet-4.5 --output-format json
 
-```bash
-zero providers add custom-openai-compatible \
-  --name minimax-openai \
-  --model MiniMax-M3 \
-  --base-url https://api.minimax.io/v1 \
-  --api-key-env MINIMAX_API_KEY \
-  --set-active
-
-zero providers add custom-openai-compatible \
-  --name minimax-cn-openai \
-  --model MiniMax-M3 \
-  --base-url https://api.minimaxi.com/v1 \
-  --api-key-env MINIMAXI_API_KEY \
-  --set-active
-```
-
-For local models, run Ollama or LM Studio and then use `zero setup` or
-`zero providers detect`.
-
-## Daily Use
-
-### Interactive TUI
-
-```bash
-zero
-```
-
-Useful controls:
-
-| Control | Action |
-|---|---|
-| `Enter` | send the prompt |
-| `/` | open slash-command suggestions |
-| `Shift+Tab` | cycle permission mode |
-| `Ctrl+B` | show/hide the sidebar |
-| `Ctrl+C` | cancel or exit |
-
-Common slash commands:
-
-| Command | Purpose |
-|---|---|
-| `/model`, `/provider` | switch the active model/provider |
-| `/spec`, `/plan` | draft and review a plan before building |
-| `/image` | attach an image for vision-capable models |
-| `/resume`, `/rewind` | continue or roll back local sessions |
-| `/loop` | repeat a prompt or custom `/command` on an interval (`/loop 5m /babysit-prs`) or self-paced |
-| `/compact`, `/context` | manage context usage |
-| `/permissions`, `/tools` | inspect available tools and policy |
-| `/add-dir` | allow an extra write directory for this session |
-| `/theme`, `/doctor`, `/config` | adjust appearance and inspect setup |
-
-### Headless `exec`
-
-```bash
-zero exec "explain internal/agent/loop.go"
-zero exec --model claude-sonnet-4.5 "refactor the config loader"
-zero exec --use-spec "add rate limiting to the API client"
-zero exec --worktree "try the migration in an isolated worktree"
-zero exec --resume
-zero exec --fork <session-id> "try the other approach"
-```
-
-Programmatic use:
-
-```bash
+# multi-turn / programmatic over stdio
 zero exec --input-format stream-json --output-format stream-json < turns.jsonl
 ```
 
-The stream-JSON contract is documented in
-[docs/STREAM_JSON_PROTOCOL.md](docs/STREAM_JSON_PROTOCOL.md).
+`exec` flags: `-f, --file` · `-m, --model` · `-C, --cwd` · `-i, --input-format <text|stream-json>` ·
+`-o, --output-format <text|json|stream-json>` · `--skip-permissions-unsafe`.
+stdout carries **only** program output; logs go to stderr. See
+[`docs/STREAM_JSON_PROTOCOL.md`](docs/STREAM_JSON_PROTOCOL.md).
 
-## Safety Model
-
-Zero is designed to make side effects visible.
-
-- Workspace reads are allowed by default.
-- File writes are limited to the workspace unless you grant another directory.
-- Shell commands, network access, destructive commands, and elevated actions are
-  permission-gated.
-- `--add-dir <path>` and `/add-dir <path>` grant additional write roots without
-  giving the agent the whole filesystem.
-- Unsafe/autonomous modes are explicit opt-ins.
-- Secrets are redacted from tool output and logs where Zero controls the surface.
-
-Example:
+### Other commands
 
 ```bash
-zero --add-dir ../docs-site
-zero exec --add-dir ../shared "update both repos"
+zero providers list|current                 # inspect resolved provider profiles
+zero models list [--provider anthropic]     # inspect the model registry
+zero search "<query>" [--json --session <id> --type <event>]   # search local sessions
+zero doctor [--connectivity] [--json]        # health checks
+zero config [--json]                          # inspect resolved configuration
+zero sandbox policy [--json]                 # inspect sandbox backend support
+zero serve --mcp [-C <path>]                  # expose Zero read-only tools over MCP stdio
+zero update --check [--json --target windows-x64] # check for a newer release
 ```
 
-Sandbox behavior can be inspected with:
+## Providers & models
 
-```bash
-zero sandbox policy
-zero sandbox grants list
-```
+Selectable per task and per session. The model registry knows each model's provider,
+capabilities, context window, and cost.
 
-## Web And Local Control
-
-Zero includes local file/search/edit/shell tools, `web_fetch` for public URLs,
-and MCP support for additional tools.
-
-For local dev servers, use shell commands such as `curl` through `exec_command`
-so the normal sandbox and permission policy applies. Long-running commands stay
-attached to a background terminal session and can be listed or stopped from the
-TUI.
-
-The npm package also includes browser and terminal helper packages used by local
-browser/terminal tools. Source builds can use the same helpers when they are on
-`PATH` or configured in Zero's local-control settings.
-
-## Common Commands
-
-```text
-zero                  interactive TUI
-zero exec             one-shot or scripted agent run
-zero setup            first-run provider setup
-zero auth             OAuth/login helpers for supported providers
-zero models           model registry and capabilities
-zero providers        provider profiles and detection
-zero doctor           setup, key, and connectivity checks
-zero context          context-budget report
-zero repo-map         deterministic repository map
-zero repo-info        local repository summary
-zero search | find    search local session history
-zero sessions         inspect, resume, fork, and rewind sessions
-zero spec             manage spec-mode drafts
-zero specialist       manage specialist subagents
-zero skills           manage markdown instruction skills
-zero plugins          manage plugins
-zero hooks            manage lifecycle hooks
-zero mcp              manage MCP servers and tools
-zero serve --mcp      expose Zero tools over MCP stdio
-zero sandbox          inspect sandbox policy and grants
-zero worktrees        prepare isolated git worktrees
-zero verify           detect and run local verification checks
-zero changes          inspect and commit local git changes
-zero usage            token usage and estimated cost
-zero cron             scheduled agent jobs
-zero update           check for newer releases
-```
-
-## Extending Zero
-
-### Project and personal instructions
-
-Zero appends project-specific guidance to the system prompt from the first
-`AGENTS.md`, `ZERO.md`, or `.zero/AGENTS.md` file found in each directory from
-the git root down to your current working directory (checked in that order
-per directory). Files are injected general-to-specific, capped at 8 KiB per
-file and 32 KiB total.
-
-A personal `ZERO.md` under `config.UserConfigDir()/zero/ZERO.md`
-(`$XDG_CONFIG_HOME/zero/ZERO.md` or `~/.config/zero/ZERO.md` on Linux/macOS,
-`%AppData%\Roaming\zero\ZERO.md` on Windows) applies across every workspace, ahead of any project guidelines.
-
-### Plugins
-
-Plugins are discovered from `~/.config/zero/plugins/<name>/plugin.json` (user
-scope — `$XDG_CONFIG_HOME` or `~/.config` on every OS, independent of the
-`config.UserConfigDir()` path used above) and `<cwd>/.zero/plugins/<name>/plugin.json`
-(project scope — resolved from the current working directory, not the repo
-root), and managed with `zero plugins`. A manifest can declare:
-
-- `tools` — custom tools (`command`, `args`, `inputSchema`, and a
-  `permission` of `prompt` or `deny`; `allow` is honored only when manifest tool
-  auto-approval is enabled)
-- `hooks` — commands run on `beforeTool`, `afterTool`, `sessionStart`, or
-  `sessionEnd`
-- `prompts` and `skills` — additional prompt/skill files
-
-MCP servers (`zero mcp`) and standalone markdown skills (`zero skills`) use
-the same extension points and can also be wired up outside of a plugin
-manifest.
-
-## Appearance And Accessibility
-
-| Control | Effect |
+| Provider | Example models |
 |---|---|
-| `NO_COLOR=<anything>` | disables color output |
-| `ZERO_THEME=<name>` | selects the startup theme (`auto`, `dark`, `light`, or a color theme like `dracula`, `nord`, `gruvbox`, `tokyo-night`, `catppuccin`, `one-dark`, `solarized-dark`, `rose-pine`, `everforest`, `solarized-light`) |
-| `--theme <name>` | selects the TUI theme from the CLI (same names) |
-| `/theme` | opens the theme picker inside the TUI (live preview; `/theme <name>` switches directly) |
-| `ZERO_NO_FADE=1` | disables streaming fade animation |
+| OpenAI-compatible | `gpt-4.1`, `gpt-4.1-mini`, `gpt-4o`, `gpt-4o-mini` |
+| Anthropic | `claude-opus-4.1`, `claude-sonnet-4.5`, `claude-haiku-4.5` |
+| Google Gemini | `gemini-2.5-pro`, `gemini-2.5-flash`, `gemini-2.5-flash-lite` |
 
-Meaning does not rely on color alone; diffs, permissions, and statuses also use
-text or glyph markers.
+Any OpenAI-compatible endpoint works with just a base URL, key, and model — so local
+runtimes (Ollama, gateways, etc.) plug in the same way.
+
+## Tools
+
+| Tool | Purpose | Side effect |
+|---|---|---|
+| `read_file` · `list_directory` · `grep` · `glob` | explore & search | read |
+| `update_plan` | maintain a live task plan | plan |
+| `write_file` · `edit_file` · `apply_patch` | create & modify files | write (gated) |
+| `bash` | run shell commands | shell (gated) |
+
+Write/shell tools route through the permission policy before any side effect.
+
+## Architecture
+
+```
+ Terminal (Bubble Tea target)   headless `exec`      (future) editor ext
+            └───────────────────┬───────────────────────┘
+                    Go-native core target (loop, events, tools)
+   ┌──────────┬───────────┬──────────┬──────────┬───────────┬──────────┐
+ providers   tools     sessions    usage     redaction    doctor /   stream-json
+ + registry  registry  + search   + cost                  config
+```
+
+- **Go-native target**: `cmd/zero` is the default production runtime; the npm package delegates to the built Go binary.
+- **Surface-agnostic core**: the agent loop streams text + tool calls, executes tools, and emits a typed event stream consumed identically by every surface.
+- **Edges are interfaces**: `Provider`, `Tool`, `SessionStore`, and the permission policy are swappable.
+- **Model is data**: capabilities, cost, and routing live in the registry — never hard-coded.
+
+## Project layout
+
+```
+cmd/
+  zero/                  # production Go CLI entrypoint
+  zero-pr-review/        # deterministic PR review helper
+internal/
+  agent/                 # Go agent loop and tool-call orchestration
+  cli/                   # interactive/headless command surface
+  config/                # layered configuration and provider profiles
+  providers/             # OpenAI-compatible, Anthropic, Gemini adapters
+  tools/                 # read/write/edit/bash/grep/glob/apply_patch tools
+  tui/                   # Bubble Tea terminal interface
+  sessions/ search/      # append-only sessions and local search
+  usage/ modelregistry/  # token usage and model metadata/costs
+  doctor/ verify/        # diagnostics and self-verification
+  streamjson/            # headless stream-json protocol
+  mcp/ plugins/ hooks/   # extension-facing runtime surfaces
+bin/                     # npm wrapper entrypoint
+scripts/                 # installers
+docs/                    # PRD + protocol/install/perf docs
+```
 
 ## Development
 
 ```bash
-go test ./...
-go run ./cmd/zero-release build
-go run ./cmd/zero-release smoke
-go run ./cmd/zero-perf-bench
+go test ./...                     # run the test suite
+go run ./cmd/zero-release build   # compile the release-facing Go binary
+go run ./cmd/zero-release smoke   # verify the release-facing Go binary
+go run ./cmd/zero-perf-bench      # performance benchmarks (see docs/PERFORMANCE.md)
 ```
 
-### Code Quality and Security Checks
-
-Before committing any changes, ensure all Go code quality and security checks pass. Pinned `go run` commands matching CI constraints can be used directly without prior installation:
-
-1. **Formatting**: Run `go fmt ./...` (or `make fmt`).
-2. **Vetting**: Run `go vet ./...` (or `make vet`).
-3. **Linting**: Run `go run github.com/golangci/golangci-lint/v2/cmd/[email protected] run --enable-only unused,ineffassign,staticcheck ./...`.
-4. **Vulnerability Scan**: Run `go run golang.org/x/vuln/cmd/[email protected] ./...`.
-
-If you prefer to install these tools globally on your path, you can run:
-
-```bash
-# Install golangci-lint
-go install github.com/golangci/golangci-lint/v2/cmd/[email protected]
-
-# Install govulncheck
-go install golang.org/x/vuln/cmd/[email protected]
-```
-
-### Cross-Compile Examples
+Cross-compile by passing Go targets to the builder:
 
 ```bash
 go run ./cmd/zero-release build --goos linux --goarch amd64
 go run ./cmd/zero-release build --goos windows --goarch amd64 --output dist/zero.exe
 ```
 
+### Install from a release
+
+```bash
+# Linux / macOS
+scripts/install.sh
+
+# Windows
+powershell -ExecutionPolicy Bypass -File scripts/install.ps1
+```
+
+See [`docs/INSTALL.md`](docs/INSTALL.md) for version, repository, and install-path overrides,
+and [`docs/UPDATE.md`](docs/UPDATE.md) for the update flow.
+
 ## Documentation
 
-- [Install](docs/INSTALL.md)
-- [Update flow](docs/UPDATE.md)
-- [Stream-JSON protocol](docs/STREAM_JSON_PROTOCOL.md)
-- [Specialists](docs/SPECIALISTS.md)
-- [GitHub Action](docs/GITHUB_ACTION.md)
-- [Benchmarks](docs/BENCHMARK.md)
-- [Performance](docs/PERFORMANCE.md)
-- [Agent evals](docs/AGENT_EVALS.md)
+- [Product Requirements (PRD)](docs/PRD.md) — vision, goals, full feature spec, roadmap
+- [Stream-JSON protocol](docs/STREAM_JSON_PROTOCOL.md) — headless I/O contract
+- [Headless exec PRD](docs/M1_HEADLESS_EXEC_PRD.md)
+- [npm wrapper smoke checklist](docs/NPM_WRAPPER_SMOKE.md)
+- [Performance](docs/PERFORMANCE.md) · [Install](docs/INSTALL.md) · [Update](docs/UPDATE.md)
 
 ## Contributing
 
-Contributions are welcome. Read [CONTRIBUTING.md](CONTRIBUTING.md), run the
-relevant tests, and open a focused pull request.
-
-Security reports should follow [SECURITY.md](SECURITY.md).
+Contributions are welcome — see [CONTRIBUTING.md](CONTRIBUTING.md). Please run
+`go test ./...` and the relevant build or smoke command before opening a PR.
 
 ## License
 
-Zero is released under the [MIT License](LICENSE).
+License is being finalized; a `LICENSE` file will be added before a public release.
+
+---
+
+<sub>Go-native core with an npm distribution wrapper.</sub>

README_ZH.md

@@ -1,279 +0,0 @@
-<p align="center">
-  <img src="docs/assets/zero-logo.png" alt="Zero" width="385">
-</p>
-
-<p align="center"><strong>一个独属于你的终端编码智能体。</strong></p>
-
-<p align="center">
-  <a href="LICENSE"><img alt="license" src="https://img.shields.io/badge/license-MIT-blue"></a>
-  <img alt="Go 1.26.5+" src="https://img.shields.io/badge/Go-1.26.5+-00ADD8?logo=go&logoColor=white">
-  <img alt="25+ providers" src="https://img.shields.io/badge/providers-25+-34E2EA">
-  <br>
-  <a href="README.md">English</a> | <strong>中文</strong>
-</p>
-
-Zero 是一个用于本地终端的 AI 编码智能体。它可以检查仓库、编辑文件、运行命令、使用浏览器/终端辅助工具，并在你选择模型和权限级别的同时保持持久的本地会话。
-
-```bash
-zero
-zero exec "修复 ./pkg 中失败的测试"
-zero exec --output-format stream-json < turns.jsonl
-```
-
-## 为什么选择 Zero
-
-- **使用你想要的模型。** 支持 OpenAI、Anthropic、Gemini、Groq、OpenRouter、DeepSeek、Mistral、xAI、Qwen、Kimi、GitHub Models、Ollama、LM Studio，或任何 OpenAI/Anthropic 兼容端点。
-- **保持控制权。** 文件写入、Shell 命令、网络访问和工作区外写入都经过 Zero 的权限和沙箱策略。
-- **在终端中工作。** TUI 具有模型/提供商选择器、图片输入、斜杠命令、实时计划/工具渲染、回滚滚动、主题以及恢复/分叉支持。
-- **无 TUI 也能工作。** `zero exec` 可脚本化，支持文本/JSON/stream-JSON I/O、隔离的工作树、规范优先运行，以及用于 CI 的有意义的退出码。
-- **保持上下文本地化。** 会话存储在磁盘上，可搜索、可恢复，且 Zero 不会作为遥测数据上传。
-- **可扩展。** 使用 MCP 服务器、技能、插件、钩子和来自同一 CLI 的专业子智能体。
-
-## 安装
-
-### npm
-
-```bash
-npm install -g @gitlawb/zero
-zero
-```
-
-npm 包是一个小型包装器，其平台构建（Linux 和 macOS 的 x64/arm64、Windows 的 x64，包含浏览器/终端控制辅助工具）作为可选依赖直接从 npm registry 安装——没有安装脚本，也不会从 npm 之外下载任何内容。Bun、pnpm 和 yarn 的行为完全一致，无需任何信任或批准步骤。跳过可选依赖的安装（`--omit=optional`）也能工作：只要二进制文件缺失，包装器就会从对应的 GitHub Release 下载。Windows on ARM 通过模拟运行 x64 构建。详见 [docs/NPM_PACKAGING.md](docs/NPM_PACKAGING.md)。
-
-### 安装脚本
-
-Linux/macOS：
-
-```bash
-curl -fsSL https://raw.githubusercontent.com/Gitlawb/zero/main/scripts/install.sh | bash
-```
-
-Windows PowerShell：
-
-```powershell
-irm https://raw.githubusercontent.com/Gitlawb/zero/main/scripts/install.ps1 | iex
-```
-
-### 从源码构建
-
-源码构建需要 Go 1.26.5+。
-
-```bash
-git clone https://github.com/Gitlawb/zero.git
-cd zero
-go run ./cmd/zero
-```
-
-发布安装器和 npm 包需要已发布的 GitHub Release 资源。如果你在首次公开发布之前进行测试，请从源码构建：
-
-```bash
-go build -o zero ./cmd/zero
-```
-
-在 Linux 上，如果你需要原生沙箱，还需要构建沙箱辅助程序：
-
-```bash
-go build -o zero-linux-sandbox ./cmd/zero-linux-sandbox
-go build -o zero-seccomp ./cmd/zero-seccomp   # 可选的兼容性包装器
-```
-
-将 `zero` 和 `zero-linux-sandbox` 放在 `PATH` 上的同一目录中（`~/.local/bin` 是一个好的默认选择）。macOS 不需要额外的辅助二进制文件。Windows 源码构建可以使用主 `zero.exe` 作为沙箱辅助程序；发布包仍然附带独立的 Windows 辅助可执行文件。
-
-更多安装细节：[docs/INSTALL.md](docs/INSTALL.md)。
-
-## 首次运行
-
-启动 TUI：
-
-```bash
-zero
-```
-
-设置向导帮助你选择提供商和模型。你也可以从命令行配置提供商：
-
-```bash
-zero setup
-zero providers list
-zero models list
-zero doctor
-```
-
-对于 API 提供商，在设置之前设置匹配的环境变量或在向导中输入密钥：
-
-```bash
-export OPENAI_API_KEY=sk-...
-export ANTHROPIC_API_KEY=...
-export GEMINI_API_KEY=...
-export LONGCAT_API_KEY=...
-```
-
-要直接配置美团 LongCat（LongCat-2.0），运行：
-
-```bash
-zero providers setup longcat --set-active
-```
-
-对于本地模型，运行 Ollama 或 LM Studio，然后使用 `zero setup` 或 `zero providers detect`。
-
-## 日常使用
-
-### 交互式 TUI
-
-```bash
-zero
-```
-
-常用控制：
-
-| 控制 | 操作 |
-|---|---|
-| `Enter` | 发送提示 |
-| `/` | 打开斜杠命令建议 |
-| `Shift+Tab` | 切换权限模式 |
-| `Ctrl+B` | 显示/隐藏侧边栏 |
-| `Ctrl+C` | 取消或退出 |
-
-常用斜杠命令：
-
-| 命令 | 用途 |
-|---|---|
-| `/model`、`/provider` | 切换活动模型/提供商 |
-| `/spec`、`/plan` | 在构建之前起草和审查计划 |
-| `/image` | 为视觉模型附加图片 |
-| `/resume`、`/rewind` | 继续或回滚本地会话 |
-| `/compact`、`/context` | 管理上下文使用 |
-| `/permissions`、`/tools` | 检查可用工具和策略 |
-| `/add-dir` | 为此会话授予额外的写入目录 |
-| `/theme`、`/doctor`、`/config` | 调整外观和检查设置 |
-
-### 无头 `exec` 模式
-
-```bash
-zero exec "解释 internal/agent/loop.go"
-zero exec --model claude-sonnet-4.5 "重构配置加载器"
-zero exec --use-spec "为 API 客户端添加速率限制"
-zero exec --worktree "在隔离的工作树中尝试迁移"
-zero exec --resume
-zero exec --fork <session-id> "尝试另一种方法"
-```
-
-编程使用：
-
-```bash
-zero exec --input-format stream-json --output-format stream-json < turns.jsonl
-```
-
-Stream-JSON 协议文档在 [docs/STREAM_JSON_PROTOCOL.md](docs/STREAM_JSON_PROTOCOL.md)。
-
-## 安全模型
-
-Zero 旨在使副作用可见。
-
-- 工作区读取默认允许。
-- 文件写入限制在工作区内，除非你授予其他目录。
-- Shell 命令、网络访问、破坏性命令和提权操作需要权限授权。
-- `--add-dir <path>` 和 `/add-dir <path>` 授予额外的写入根目录，而不会给智能体整个文件系统。
-- 不安全/自主模式是显式选择加入。
-- 在 Zero 控制的界面上，密钥会从工具输出和日志中被脱敏。
-
-示例：
-
-```bash
-zero --add-dir ../docs-site
-zero exec --add-dir ../shared "更新两个仓库"
-```
-
-沙箱行为可以通过以下方式检查：
-
-```bash
-zero sandbox policy
-zero sandbox grants list
-```
-
-## Web 和本地控制
-
-Zero 包含本地文件/搜索/编辑/Shell 工具、用于公共 URL 的 `web_fetch`，以及用于额外工具的 MCP 支持。
-
-对于本地开发服务器，使用 Shell 命令（如通过 `exec_command` 的 `curl`），这样正常的沙箱和权限策略就会生效。长时间运行的命令保持附加到后台终端会话，可以从 TUI 中列出或停止。
-
-npm 包还包含本地浏览器/终端工具使用的浏览器和终端辅助包。源码构建可以在它们位于 `PATH` 上或在 Zero 的本地控制设置中配置时使用相同的辅助工具。
-
-## 常用命令
-
-```text
-zero                  交互式 TUI
-zero exec             一次性或脚本化智能体运行
-zero setup            首次运行提供商设置
-zero auth             支持提供商的 OAuth/登录辅助
-zero models           模型注册表和能力
-zero providers        提供商配置和检测
-zero doctor           设置、密钥和连接检查
-zero context          上下文预算报告
-zero repo-map         确定性仓库映射
-zero repo-info        本地仓库摘要
-zero search | find    搜索本地会话历史
-zero sessions         检查、恢复、分叉和回滚会话
-zero spec             管理规范模式草稿
-zero specialist       管理专业子智能体
-zero skills           管理 Markdown 指令技能
-zero plugins          管理插件
-zero hooks            管理生命周期钩子
-zero mcp              管理 MCP 服务器和工具
-zero serve --mcp      通过 MCP stdio 暴露 Zero 工具
-zero sandbox          检查沙箱策略和授权
-zero worktrees        准备隔离的 Git 工作树
-zero verify           检测和运行本地验证检查
-zero changes          检查和提交本地 Git 变更
-zero usage            Token 使用量和估算成本
-zero cron             定时智能体任务
-zero update           检查更新版本
-```
-
-## 外观和无障碍
-
-| 控制 | 效果 |
-|---|---|
-| `NO_COLOR=<任意值>` | 禁用颜色输出 |
-| `ZERO_THEME=<名称>` | 选择启动主题（`auto`、`dark`、`light`，或颜色主题如 `dracula`、`nord`、`gruvbox`、`tokyo-night`、`catppuccin`、`one-dark`、`solarized-dark`、`rose-pine`、`everforest`、`solarized-light`） |
-| `--theme <名称>` | 从 CLI 选择 TUI 主题（相同名称） |
-| `/theme` | 在 TUI 中打开主题选择器（实时预览；`/theme <名称>` 直接切换） |
-| `ZERO_NO_FADE=1` | 禁用流式淡入动画 |
-
-含义不仅仅依赖于颜色；差异、权限和状态也使用文本或标记符号。
-
-## 开发
-
-```bash
-go test ./...
-go run ./cmd/zero-release build
-go run ./cmd/zero-release smoke
-go run ./cmd/zero-perf-bench
-```
-
-交叉编译示例：
-
-```bash
-go run ./cmd/zero-release build --goos linux --goarch amd64
-go run ./cmd/zero-release build --goos windows --goarch amd64 --output dist/zero.exe
-```
-
-## 文档
-
-- [安装](docs/INSTALL.md)
-- [更新流程](docs/UPDATE.md)
-- [Stream-JSON 协议](docs/STREAM_JSON_PROTOCOL.md)
-- [专家](docs/SPECIALISTS.md)
-- [GitHub Action](docs/GITHUB_ACTION.md)
-- [基准测试](docs/BENCHMARK.md)
-- [性能](docs/PERFORMANCE.md)
-- [智能体评估](docs/AGENT_EVALS.md)
-
-## 贡献
-
-欢迎贡献。阅读 [CONTRIBUTING.md](CONTRIBUTING.md)，运行相关测试，然后提交一个聚焦的拉取请求。
-
-安全报告应遵循 [SECURITY.md](SECURITY.md)。
-
-## 许可证
-
-Zero 基于 [MIT 许可证](LICENSE) 发布。

SECURITY.md

@@ -1,36 +0,0 @@
-# Security Policy
-
-Zero is a terminal coding agent: it reads and edits files in your repository and can run shell
-commands and call external model/tool endpoints. We take security reports seriously.
-
-## Reporting a vulnerability
-
-**Please do not open a public issue for security vulnerabilities.**
-
-Report privately through GitHub's private vulnerability reporting:
-
-1. Go to the repository's **Security** tab → **Report a vulnerability** (GitHub Security Advisories).
-2. Describe the issue, the affected version/commit, reproduction steps, and the impact.
-
-If you cannot use GitHub Security Advisories, open a regular issue that contains **only** a request to
-be contacted for a security report (no details), and a maintainer will arrange a private channel.
-
-Please include, where possible:
-
-- The version, branch, or commit affected.
-- A minimal reproduction and the impact (e.g. sandbox escape, secret exposure, RCE).
-- Any logs or output, with secrets redacted.
-
-## What to expect
-
-- We aim to acknowledge a report within a few business days.
-- We will work with you on a fix and a coordinated disclosure timeline, and credit you in the release
-  notes unless you prefer to remain anonymous.
-
-## Scope notes
-
-- Findings that require a malicious local user who already controls the machine running Zero, or that
-  depend on a user explicitly disabling the sandbox (`--skip-permissions-unsafe`), are generally
-  out of scope — but please report anything you are unsure about.
-- Zero sends no telemetry. Network calls go to the model/tool providers you configure (and, for
-  keyless web search, to the documented Firecrawl endpoint — see the README).

action.yml

@@ -1,296 +0,0 @@
-name: ZERO
-description: >-
-  Run the ZERO coding agent headlessly in a GitHub workflow. ZERO is
-  model- and provider-agnostic: you supply the provider and API key as inputs.
-author: Gitlawb
-
-branding:
-  icon: terminal
-  color: green
-
-inputs:
-  prompt:
-    description: The instruction for ZERO to execute. Mutually exclusive with prompt-file.
-    required: false
-    default: ""
-  prompt-file:
-    description: Path (relative to working-directory) to a file whose contents are the prompt.
-    required: false
-    default: ""
-  provider:
-    description: >-
-      Provider id to activate (for example openai, anthropic, gemini, ollama, or
-      any configured OpenAI-/Anthropic-compatible endpoint). Optional when the
-      repository already carries a .zero/config.json with an active provider.
-    required: false
-    default: ""
-  api-key:
-    description: >-
-      The provider API key. Pass it from a repository or organization secret;
-      it is exported only for the ZERO step and never written to the log.
-    required: false
-    default: ""
-  api-key-env:
-    description: >-
-      Name of the environment variable the chosen provider reads its key from
-      (for example OPENAI_API_KEY or ANTHROPIC_API_KEY). When set together with
-      api-key, that variable is exported only for the ZERO step. Leave the action
-      provider-agnostic by passing the env name your provider expects rather
-      than hardcoding one here.
-    required: false
-    default: ""
-  model:
-    description: Model id to use for the run. Defaults to the resolved provider's default.
-    required: false
-    default: ""
-  mode:
-    description: Run mode passed to `zero exec --mode` (for example smart, deep, or fast).
-    required: false
-    default: ""
-  auto:
-    description: >-
-      Autonomy ceiling passed to `zero exec --auto` (low, medium, or high).
-      Defaults to the conservative `low` so an unattended CI run cannot take
-      high-impact actions implicitly.
-    required: false
-    default: low
-  self-correct:
-    description: >-
-      When true, allow ZERO to request a stronger model mid-run if it hits a
-      wall (maps to `zero exec --allow-escalation`).
-    required: false
-    default: "false"
-  add-dir:
-    description: >-
-      Newline- or comma-separated extra write roots granted to the run (each
-      passed as `zero exec --add-dir`). The checked-out repository is always
-      writable; everything else stays read-only unless listed here.
-    required: false
-    default: ""
-  worktree:
-    description: When true, run inside an isolated git worktree (`zero exec --worktree`).
-    required: false
-    default: "false"
-  output-format:
-    description: >-
-      Output format for `zero exec` (text, json, or stream-json). Defaults to
-      stream-json, which is captured to a file and exposed as the output-file
-      output and uploaded by the caller if desired.
-    required: false
-    default: stream-json
-  post-to:
-    description: >-
-      Optional destination for a run summary after ZERO finishes: `pr-comment`
-      posts to the triggering pull request using github-token, `slack` posts to
-      slack-webhook-url, or `none` (default) posts nowhere.
-    required: false
-    default: none
-  slack-webhook-url:
-    description: >-
-      Slack incoming-webhook (or generic webhook) URL used when post-to is
-      slack. Pass it from a secret; it is never written to the log.
-    required: false
-    default: ""
-  github-token:
-    description: >-
-      Token used to post a PR comment when post-to is pr-comment. Defaults to
-      the workflow's GITHUB_TOKEN; the workflow must grant pull-requests: write.
-    required: false
-    default: ${{ github.token }}
-  working-directory:
-    description: Directory to run ZERO in. Defaults to the workspace root.
-    required: false
-    default: ${{ github.workspace }}
-  zero-version:
-    description: >-
-      ZERO release version/tag to install (for example v1.2.3 or latest).
-      Defaults to the ref this action was resolved at, falling back to latest.
-    required: false
-    default: ""
-  zero-repo:
-    description: Repository to install the ZERO release from (owner/repo).
-    required: false
-    default: Gitlawb/zero
-
-outputs:
-  exit-code:
-    description: The exit code ZERO returned (0 success, 2 usage, 3 provider, non-zero otherwise).
-    value: ${{ steps.run.outputs.exit-code }}
-  output-file:
-    description: Path to the captured ZERO stdout (the raw output-format stream).
-    value: ${{ steps.run.outputs.output-file }}
-  summary:
-    description: A short human-readable summary line parsed from the run, when available.
-    value: ${{ steps.run.outputs.summary }}
-
-runs:
-  using: composite
-  steps:
-    - name: Install ZERO
-      shell: bash
-      env:
-        ZERO_REPO: ${{ inputs.zero-repo }}
-        ZERO_VERSION_INPUT: ${{ inputs.zero-version }}
-        ACTION_REF: ${{ github.action_ref }}
-        ACTION_PATH: ${{ github.action_path }}
-        RUNNER_TEMP: ${{ runner.temp }}
-      run: |
-        set -euo pipefail
-        if [ "${RUNNER_OS}" = "Windows" ]; then
-          echo "::error::The ZERO action currently supports Linux and macOS runners only." >&2
-          exit 1
-        fi
-        # Resolve the version to install: explicit input wins; otherwise the ref
-        # this action was checked out at (a tag); otherwise the latest release.
-        version="${ZERO_VERSION_INPUT}"
-        if [ -z "${version}" ]; then
-          version="${ACTION_REF}"
-        fi
-        # A commit-SHA pin (github.action_ref is the 40-hex SHA) is not a release
-        # tag, so fall back to latest rather than fetching releases/download/v<sha>.
-        case "${version}" in
-          ""|refs/*|*/*) version="latest" ;;
-        esac
-        if printf '%s' "${version}" | grep -qiE '^[0-9a-f]{40}$'; then
-          version="latest"
-        fi
-        install_dir="${RUNNER_TEMP}/zero-bin"
-        mkdir -p "${install_dir}"
-        # Reuse ZERO's own checksum-verifying installer, shipped with the action.
-        ZERO_REPO="${ZERO_REPO}" ZERO_VERSION="${version}" ZERO_INSTALL_DIR="${install_dir}" \
-          bash "${ACTION_PATH}/scripts/install.sh"
-        echo "${install_dir}" >> "${GITHUB_PATH}"
-
-    - name: Run ZERO
-      id: run
-      shell: bash
-      working-directory: ${{ inputs.working-directory }}
-      env:
-        # Secrets are passed via the environment and never echoed. `set +x` and
-        # the absence of any `echo` of these values keep them out of the log.
-        ZERO_INPUT_API_KEY: ${{ inputs.api-key }}
-        ZERO_INPUT_API_KEY_ENV: ${{ inputs.api-key-env }}
-        ZERO_INPUT_PROVIDER: ${{ inputs.provider }}
-        ZERO_INPUT_PROMPT: ${{ inputs.prompt }}
-        ZERO_INPUT_PROMPT_FILE: ${{ inputs.prompt-file }}
-        ZERO_INPUT_MODEL: ${{ inputs.model }}
-        ZERO_INPUT_MODE: ${{ inputs.mode }}
-        ZERO_INPUT_AUTO: ${{ inputs.auto }}
-        ZERO_INPUT_SELF_CORRECT: ${{ inputs.self-correct }}
-        ZERO_INPUT_ADD_DIR: ${{ inputs.add-dir }}
-        ZERO_INPUT_WORKTREE: ${{ inputs.worktree }}
-        ZERO_INPUT_OUTPUT_FORMAT: ${{ inputs.output-format }}
-        ACTION_PATH: ${{ github.action_path }}
-        RUNNER_TEMP: ${{ runner.temp }}
-      run: |
-        set -euo pipefail
-
-        # Export the provider key under the env name the provider expects. This
-        # keeps the action provider-agnostic: no provider->env mapping is baked
-        # in. The value is never printed.
-        if [ -n "${ZERO_INPUT_API_KEY}" ] && [ -n "${ZERO_INPUT_API_KEY_ENV}" ]; then
-          if ! printf '%s' "${ZERO_INPUT_API_KEY_ENV}" | grep -qE '^[A-Za-z_][A-Za-z0-9_]*$'; then
-            echo "::error::api-key-env must be a valid environment variable name." >&2
-            exit 2
-          fi
-          # Export in-process for this Run ZERO step so zero can read it; do
-          # not persist provider credentials to later job steps.
-          export "${ZERO_INPUT_API_KEY_ENV}=${ZERO_INPUT_API_KEY}"
-        fi
-        if [ -n "${ZERO_INPUT_PROVIDER}" ]; then
-          export ZERO_PROVIDER="${ZERO_INPUT_PROVIDER}"
-        fi
-
-        args=(exec)
-
-        if [ -n "${ZERO_INPUT_PROMPT_FILE}" ] && [ -n "${ZERO_INPUT_PROMPT}" ]; then
-          echo "::error::The 'prompt' and 'prompt-file' inputs are mutually exclusive." >&2
-          exit 2
-        fi
-        if [ -n "${ZERO_INPUT_PROMPT_FILE}" ]; then
-          args+=(--file "${ZERO_INPUT_PROMPT_FILE}")
-        elif [ -n "${ZERO_INPUT_PROMPT}" ]; then
-          args+=(--prompt "${ZERO_INPUT_PROMPT}")
-        else
-          echo "::error::Provide either the 'prompt' or 'prompt-file' input." >&2
-          exit 2
-        fi
-
-        [ -n "${ZERO_INPUT_MODEL}" ] && args+=(--model "${ZERO_INPUT_MODEL}")
-        [ -n "${ZERO_INPUT_MODE}" ] && args+=(--mode "${ZERO_INPUT_MODE}")
-        [ -n "${ZERO_INPUT_AUTO}" ] && args+=(--auto "${ZERO_INPUT_AUTO}")
-        [ "${ZERO_INPUT_SELF_CORRECT}" = "true" ] && args+=(--allow-escalation)
-        [ "${ZERO_INPUT_WORKTREE}" = "true" ] && args+=(--worktree)
-        args+=(--output-format "${ZERO_INPUT_OUTPUT_FORMAT}")
-        # No --notify in CI (no terminal); the optional Slack post is handled
-        # separately below so it works regardless of output format.
-        args+=(--no-notify)
-
-        # Split add-dir on newlines and commas; each non-empty entry is a root.
-        if [ -n "${ZERO_INPUT_ADD_DIR}" ]; then
-          while IFS= read -r dir; do
-            # Trim only surrounding whitespace so paths with internal spaces survive.
-            dir="$(printf '%s' "${dir}" | sed 's/^[[:space:]]*//; s/[[:space:]]*$//')"
-            [ -n "${dir}" ] && args+=(--add-dir "${dir}")
-          done < <(printf '%s' "${ZERO_INPUT_ADD_DIR}" | tr ',' '\n')
-        fi
-
-        output_file="${RUNNER_TEMP}/zero-output.txt"
-
-        # Run ZERO. Capture stdout to a file AND echo it to the log. Surface the
-        # real exit code as the step's status. The sandbox is active by default;
-        # this action never enables unsafe permission-skipping.
-        set +e
-        zero "${args[@]}" | tee "${output_file}"
-        code=${PIPESTATUS[0]}
-        set -e
-
-        echo "exit-code=${code}" >> "${GITHUB_OUTPUT}"
-        echo "output-file=${output_file}" >> "${GITHUB_OUTPUT}"
-
-        # Best-effort one-line summary. Structured output ends with lifecycle
-        # metadata, so parse final/error events instead of tailing stdout.
-        summary="$(node "${ACTION_PATH}/scripts/action-summary.mjs" "${ZERO_INPUT_OUTPUT_FORMAT}" "${output_file}" || true)"
-        {
-          echo "summary<<ZERO_EOF"
-          printf '%s\n' "${summary}"
-          echo "ZERO_EOF"
-        } >> "${GITHUB_OUTPUT}"
-
-        exit "${code}"
-
-    - name: Post run summary to pull request
-      if: ${{ always() && inputs.post-to == 'pr-comment' && github.event.pull_request.number != '' }}
-      shell: bash
-      env:
-        GH_TOKEN: ${{ inputs.github-token }}
-        PR_NUMBER: ${{ github.event.pull_request.number }}
-        ZERO_EXIT: ${{ steps.run.outputs.exit-code }}
-        ZERO_SUMMARY: ${{ steps.run.outputs.summary }}
-      run: |
-        set -euo pipefail
-        status="succeeded"
-        [ "${ZERO_EXIT}" != "0" ] && status="failed (exit ${ZERO_EXIT})"
-        body="$(printf 'ZERO run %s.\n\n%s' "${status}" "${ZERO_SUMMARY}")"
-        gh api "repos/${GITHUB_REPOSITORY}/issues/${PR_NUMBER}/comments" \
-          -f body="${body}" >/dev/null
-
-    - name: Post run summary to Slack
-      if: ${{ always() && inputs.post-to == 'slack' && inputs.slack-webhook-url != '' }}
-      shell: bash
-      env:
-        SLACK_WEBHOOK_URL: ${{ inputs.slack-webhook-url }}
-        ZERO_EXIT: ${{ steps.run.outputs.exit-code }}
-        ZERO_SUMMARY: ${{ steps.run.outputs.summary }}
-        RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-      run: |
-        set -euo pipefail
-        status="succeeded"
-        [ "${ZERO_EXIT}" != "0" ] && status="failed (exit ${ZERO_EXIT})"
-        text="ZERO run ${status} for ${GITHUB_REPOSITORY}@${GITHUB_REF_NAME}. ${ZERO_SUMMARY} ${RUN_URL}"
-        # Slack incoming webhooks accept {"text": "..."}; jq escapes the body and
-        # the URL is never echoed. A delivery failure is a warning, not a failure.
-        payload="$(jq -n --arg text "${text}" '{text: $text}')"
-        curl --fail --silent --show-error -X POST -H 'Content-Type: application/json' \
-          -d "${payload}" "${SLACK_WEBHOOK_URL}" >/dev/null || \
-          echo "::warning::Slack webhook delivery failed (run not affected)."

bin/zero.js

@@ -2,7 +2,6 @@
 
 import { spawnSync } from 'node:child_process';
 import { existsSync } from 'node:fs';
-import { createRequire } from 'node:module';
 import { dirname, join } from 'node:path';
 import { fileURLToPath } from 'node:url';
 
@@ -10,139 +9,18 @@ function zeroBinaryName(platform = process.platform) {
   return platform === 'win32' ? 'zero.exe' : 'zero';
 }
 
-function helperShimNames(name, platform = process.platform) {
-  if (platform === 'win32') {
-    return [`${name}.cmd`, `${name}.exe`, name];
-  }
-  return [name];
-}
-
-function commandForShim(path, platform = process.platform) {
-  if (platform === 'win32' && path.toLowerCase().endsWith('.cmd')) {
-    return {
-      command: process.env.ComSpec || 'cmd.exe',
-      prefixArgs: ['/d', '/s', '/c', `"${path.replace(/"/g, '""')}"`],
-    };
-  }
-  return { command: path, prefixArgs: [] };
-}
-
-function resolveHelper(packageRoot, name) {
-  const binDir = join(packageRoot, 'node_modules', '.bin');
-  for (const shimName of helperShimNames(name)) {
-    const candidate = join(binDir, shimName);
-    if (!existsSync(candidate)) continue;
-    return {
-      ...commandForShim(candidate),
-      pathPrepend: [binDir],
-    };
-  }
-  return null;
-}
-
-// Legacy path: helpers installed as npm dependencies of the wrapper. Current
-// installs vendor the helpers inside the platform package instead (a helpers/
-// directory next to the binary, found by the Go side on its own), so this
-// manifest is usually empty — it only matters for a downloaded-binary install
-// whose wrapper still carries helper dependencies.
-function localControlHelperManifest(packageRoot) {
-  const helpers = {};
-  for (const name of ['agent-browser', 'tuistory']) {
-    const helper = resolveHelper(packageRoot, name);
-    if (helper) helpers[name] = helper;
-  }
-  if (Object.keys(helpers).length === 0) return '';
-  return JSON.stringify({ version: 1, helpers });
-}
-
-// The platform payload is a version of @gitlawb/zero itself, installed under
-// an npm: alias (see docs/NPM_PACKAGING.md). The alias name is derived from
-// process.platform/process.arch, so an unsupported platform simply fails to
-// resolve and we fall through to the downloader.
-function platformPackageBinary() {
-  const aliasedName = `@gitlawb/zero-${process.platform}-${process.arch}`;
-  let manifestPath;
-  try {
-    manifestPath = createRequire(import.meta.url).resolve(`${aliasedName}/package.json`);
-  } catch {
-    return null;
-  }
-  const candidate = join(dirname(manifestPath), zeroBinaryName());
-  return existsSync(candidate) ? candidate : null;
-}
-
-// Mirrors the release matrix (internal/release/release.go): the platforms a
-// GitHub Release asset exists for. android maps to the linux asset; there is
-// no windows-arm64 asset (Windows on ARM runs the x64 build under emulation).
-function releaseAssetAvailable() {
-  const platform = { linux: 'linux', android: 'linux', darwin: 'macos', win32: 'windows' }[
-    process.platform
-  ];
-  const arch = process.arch === 'x64' || process.arch === 'arm64' ? process.arch : null;
-  if (!platform || !arch) return false;
-  return !(platform === 'windows' && arch === 'arm64');
-}
-
-// Fallback for installs without a platform package (--omit=optional, package
-// managers that skip optionalDependencies): fetch the binary next to the
-// wrapper with the same checksum-verified downloader that used to run as
-// postinstall. Deliberately NOT cached on failure — a transient network error
-// self-heals on the next run — so this retries on every invocation until a
-// binary is in place. Platforms with no release asset skip the attempt.
-function downloadMissingBinary(packageRoot) {
-  const downloadScript = join(packageRoot, 'scripts', 'postinstall.mjs');
-  if (!existsSync(downloadScript) || !releaseAssetAvailable()) return;
-  console.error(
-    '[zero] no platform package installed — fetching the native binary from the GitHub Release (retried on each run until it succeeds).',
-  );
-  spawnSync(process.execPath, [downloadScript], {
-    stdio: ['ignore', 'inherit', 'inherit'],
-  });
-}
-
 const packageRoot = dirname(dirname(fileURLToPath(import.meta.url)));
+const nativePath = join(packageRoot, zeroBinaryName());
 
-function resolveNativeBinary() {
-  const fromPlatformPackage = platformPackageBinary();
-  if (fromPlatformPackage) return fromPlatformPackage;
-  const downloaded = join(packageRoot, zeroBinaryName());
-  if (existsSync(downloaded)) return downloaded;
-  downloadMissingBinary(packageRoot);
-  return existsSync(downloaded) ? downloaded : null;
-}
-
-const nativePath = resolveNativeBinary();
-
-if (!nativePath) {
+if (!existsSync(nativePath)) {
   console.error(
-    '[zero] No native binary is available for this install.\n' +
-      'Normally npm installs it as an optional dependency of @gitlawb/zero\n' +
-      `(@gitlawb/zero-${process.platform}-${process.arch}), and the wrapper can\n` +
-      'also download it from the GitHub Release when it is missing.\n' +
-      '\n' +
-      'Things to try:\n' +
-      '  - reinstall without omitting optional dependencies:\n' +
-      '      npm install -g @gitlawb/zero\n' +
-      '  - run the downloader manually (needs write access to the package dir):\n' +
-      `      node "${join(packageRoot, 'scripts', 'postinstall.mjs')}"\n` +
-      '\n' +
-      'If this platform has no prebuilt binary, build from source:\n' +
-      'https://github.com/Gitlawb/zero (go run ./cmd/zero, requires Go 1.26+).',
+    '[zero] No native binary found next to the npm wrapper. Reinstall the zero package or run `go run ./cmd/zero-release build` from the repository.'
   );
   process.exit(1);
 }
 
-const env = { ...process.env };
-const localControlHelpers = localControlHelperManifest(packageRoot);
-if (localControlHelpers) {
-  env.ZERO_LOCAL_CONTROL_HELPERS = localControlHelpers;
-} else {
-  delete env.ZERO_LOCAL_CONTROL_HELPERS;
-}
-
 const child = spawnSync(nativePath, process.argv.slice(2), {
   stdio: 'inherit',
-  env,
 });
 
 if (child.error) {

cmd/zero-linux-sandbox/main.go

@@ -1,13 +0,0 @@
-//go:build linux
-
-package main
-
-import (
-	"os"
-
-	"github.com/Gitlawb/zero/internal/sandbox"
-)
-
-func main() {
-	os.Exit(sandbox.RunLinuxSandboxHelper(os.Args[1:], os.Stderr))
-}

cmd/zero-perf-bench/main.go

@@ -33,12 +33,6 @@ func main() {
 }
 
 func run(args []string, getenv func(string) string, stdout io.Writer, stderr io.Writer) int {
-	// The `tasks` subcommand is the reproducible Terminal-Bench-style task harness;
-	// the default (no subcommand) path stays the cold-start/RSS perf benchmark so
-	// existing invocations are unchanged.
-	if len(args) > 0 && args[0] == "tasks" {
-		return runTasksCommand(args[1:], getenv, stdout, stderr)
-	}
 	options, err := parseArgs(args, getenv)
 	if err != nil {
 		_, _ = fmt.Fprintln(stderr, err.Error())
@@ -205,7 +199,6 @@ func parseArgs(args []string, getenv func(string) string) (cliOptions, error) {
 func helpText() string {
 	return strings.Join([]string{
 		"Usage: zero-perf-bench [options]",
-		"       zero-perf-bench tasks [options]   (Terminal-Bench-style task harness; see `tasks --help`)",
 		"",
 		"Options:",
 		"  --iterations <n>             Measured samples to collect (default: 5)",

cmd/zero-perf-bench/tasks.go

@@ -1,218 +0,0 @@
-package main
-
-import (
-	"bytes"
-	"context"
-	"fmt"
-	"io"
-	"os"
-	"path/filepath"
-	"strings"
-
-	"github.com/Gitlawb/zero/internal/perfbench"
-)
-
-// taskOptions configures the `zero-perf-bench tasks` subcommand: the reproducible
-// Terminal-Bench-style harness that runs ZERO headlessly against a task set and
-// records a publishable result (model + commit + self-correct flag + date).
-type taskOptions struct {
-	SuitePath   string
-	Model       string
-	Mode        string
-	SelfCorrect bool
-	Binary      string
-	Version     string
-	Commit      string
-	Output      string
-	JSON        bool
-	DryRun      bool
-	Help        bool
-}
-
-func runTasksCommand(args []string, getenv func(string) string, stdout io.Writer, stderr io.Writer) int {
-	options, err := parseTaskArgs(args, getenv)
-	if err != nil {
-		_, _ = fmt.Fprintln(stderr, err.Error())
-		return 2
-	}
-	if options.Help {
-		_, _ = fmt.Fprint(stdout, taskHelpText())
-		return 0
-	}
-
-	set, err := perfbench.LoadTaskSet(options.SuitePath)
-	if err != nil {
-		_, _ = fmt.Fprintln(stderr, "[zero] Task benchmark failed: "+err.Error())
-		return 1
-	}
-
-	runner := perfbench.SkipRunner
-	if !options.DryRun {
-		binary := strings.TrimSpace(options.Binary)
-		if binary == "" {
-			_, _ = fmt.Fprintln(stderr, "[zero] Task benchmark failed: --binary is required unless --dry-run is set")
-			return 2
-		}
-		runner = perfbench.NewExecRunner(binary)
-	}
-
-	result, err := perfbench.RunTasks(context.Background(), set, perfbench.TaskConfig{
-		Model:       options.Model,
-		Mode:        options.Mode,
-		SelfCorrect: options.SelfCorrect,
-		Version:     options.Version,
-		Commit:      options.Commit,
-		Runner:      runner,
-	})
-	if err != nil {
-		_, _ = fmt.Fprintln(stderr, "[zero] Task benchmark failed: "+err.Error())
-		return 1
-	}
-
-	if options.Output != "" {
-		if err := writeTaskReport(options.Output, result); err != nil {
-			_, _ = fmt.Fprintln(stderr, "[zero] Task benchmark failed: "+err.Error())
-			return 1
-		}
-	}
-	if options.JSON {
-		if err := perfbench.WriteTaskJSON(stdout, result); err != nil {
-			_, _ = fmt.Fprintln(stderr, "[zero] Task benchmark failed: "+err.Error())
-			return 1
-		}
-		return 0
-	}
-	_, _ = fmt.Fprintln(stdout, perfbench.FormatTaskSummary(result))
-	return 0
-}
-
-func parseTaskArgs(args []string, getenv func(string) string) (taskOptions, error) {
-	options := taskOptions{
-		Version: strings.TrimSpace(getenv("ZERO_BENCH_VERSION")),
-		Commit:  strings.TrimSpace(getenv("ZERO_BENCH_COMMIT")),
-	}
-	for index := 0; index < len(args); index++ {
-		arg := args[index]
-		flag, inlineValue := splitFlagValue(arg)
-		switch flag {
-		case "--suite":
-			value, next, err := readOptionValue(args, inlineValue, index, flag)
-			if err != nil {
-				return options, err
-			}
-			options.SuitePath = value
-			index = next
-		case "--model":
-			value, next, err := readOptionValue(args, inlineValue, index, flag)
-			if err != nil {
-				return options, err
-			}
-			options.Model = value
-			index = next
-		case "--mode":
-			value, next, err := readOptionValue(args, inlineValue, index, flag)
-			if err != nil {
-				return options, err
-			}
-			options.Mode = value
-			index = next
-		case "--binary":
-			value, next, err := readOptionValue(args, inlineValue, index, flag)
-			if err != nil {
-				return options, err
-			}
-			options.Binary = value
-			index = next
-		case "--version":
-			value, next, err := readOptionValue(args, inlineValue, index, flag)
-			if err != nil {
-				return options, err
-			}
-			options.Version = value
-			index = next
-		case "--commit":
-			value, next, err := readOptionValue(args, inlineValue, index, flag)
-			if err != nil {
-				return options, err
-			}
-			options.Commit = value
-			index = next
-		case "--output":
-			value, next, err := readOptionValue(args, inlineValue, index, flag)
-			if err != nil {
-				return options, err
-			}
-			options.Output = value
-			index = next
-		case "--self-correct":
-			if strings.Contains(arg, "=") {
-				return options, fmt.Errorf("%s does not accept a value", flag)
-			}
-			options.SelfCorrect = true
-		case "--json":
-			if strings.Contains(arg, "=") {
-				return options, fmt.Errorf("%s does not accept a value", flag)
-			}
-			options.JSON = true
-		case "--dry-run":
-			if strings.Contains(arg, "=") {
-				return options, fmt.Errorf("%s does not accept a value", flag)
-			}
-			options.DryRun = true
-		case "-h", "--help":
-			if strings.Contains(arg, "=") {
-				return options, fmt.Errorf("%s does not accept a value", flag)
-			}
-			options.Help = true
-		default:
-			return options, fmt.Errorf("unknown option: %s", arg)
-		}
-	}
-	if options.Help {
-		return options, nil
-	}
-	if strings.TrimSpace(options.SuitePath) == "" {
-		return options, fmt.Errorf("--suite is required")
-	}
-	if strings.TrimSpace(options.Model) == "" {
-		return options, fmt.Errorf("--model is required")
-	}
-	return options, nil
-}
-
-func writeTaskReport(path string, result perfbench.TaskRunResult) error {
-	if dir := filepath.Dir(path); dir != "" {
-		if err := os.MkdirAll(dir, 0o755); err != nil {
-			return err
-		}
-	}
-	var buffer bytes.Buffer
-	if err := perfbench.WriteTaskJSON(&buffer, result); err != nil {
-		return err
-	}
-	return os.WriteFile(path, buffer.Bytes(), 0o644)
-}
-
-func taskHelpText() string {
-	return strings.Join([]string{
-		"Usage: zero-perf-bench tasks [options]",
-		"",
-		"Runs ZERO headlessly against a Terminal-Bench-style task set and records a",
-		"reproducible, publishable result (model + commit + self-correct flag + date).",
-		"",
-		"Options:",
-		"  --suite <path>      Task set JSON file (required)",
-		"  --model <model>     Model to run (required); recorded with the result",
-		"  --mode <name>       Exec mode preset to apply",
-		"  --self-correct      Enable the post-edit verify-and-correct loop",
-		"  --binary <path>     Path to the `zero` binary (required unless --dry-run)",
-		"  --version <v>       Record the ZERO version (default: $ZERO_BENCH_VERSION)",
-		"  --commit <sha>      Record the ZERO commit (default: $ZERO_BENCH_COMMIT)",
-		"  --output <path>     Write the JSON result to path",
-		"  --json              Print only the JSON result",
-		"  --dry-run           Record every task as skipped without invoking the agent",
-		"  -h, --help          Show this help",
-		"",
-		"Run twice — with and without --self-correct — to publish the delta.",
-	}, "\n") + "\n"
-}

cmd/zero-perf-bench/tasks_test.go

@@ -1,170 +0,0 @@
-package main
-
-import (
-	"bytes"
-	"encoding/json"
-	"os"
-	"path/filepath"
-	"strings"
-	"testing"
-)
-
-func writeTaskSuite(t *testing.T, body string) string {
-	t.Helper()
-	dir := t.TempDir()
-	path := filepath.Join(dir, "suite.json")
-	if err := os.WriteFile(path, []byte(body), 0o600); err != nil {
-		t.Fatalf("write suite: %v", err)
-	}
-	return path
-}
-
-func TestParseTaskArgsRequiresSuiteAndModel(t *testing.T) {
-	if _, err := parseTaskArgs([]string{"--model", "m"}, emptyEnv); err == nil || !strings.Contains(err.Error(), "--suite") {
-		t.Fatalf("expected missing-suite error, got %v", err)
-	}
-	if _, err := parseTaskArgs([]string{"--suite", "/x.json"}, emptyEnv); err == nil || !strings.Contains(err.Error(), "--model") {
-		t.Fatalf("expected missing-model error, got %v", err)
-	}
-}
-
-func TestParseTaskArgsReadsFlagsAndEnv(t *testing.T) {
-	env := func(key string) string {
-		return map[string]string{
-			"ZERO_BENCH_COMMIT":  "deadbeef",
-			"ZERO_BENCH_VERSION": "9.9.9",
-		}[key]
-	}
-	options, err := parseTaskArgs([]string{
-		"--suite", "/tmp/suite.json",
-		"--model", "test-model",
-		"--mode", "build",
-		"--self-correct",
-		"--binary", "/usr/local/bin/zero",
-		"--output", "dist/bench.json",
-		"--json",
-	}, env)
-	if err != nil {
-		t.Fatalf("parseTaskArgs error: %v", err)
-	}
-	if options.SuitePath != "/tmp/suite.json" || options.Model != "test-model" || options.Mode != "build" {
-		t.Fatalf("options = %#v", options)
-	}
-	if !options.SelfCorrect || !options.JSON {
-		t.Fatalf("flags not parsed: %#v", options)
-	}
-	if options.Binary != "/usr/local/bin/zero" || options.Output != "dist/bench.json" {
-		t.Fatalf("binary/output = %q/%q", options.Binary, options.Output)
-	}
-	if options.Commit != "deadbeef" || options.Version != "9.9.9" {
-		t.Fatalf("commit/version from env = %q/%q", options.Commit, options.Version)
-	}
-}
-
-func TestParseTaskArgsFlagsOverrideEnv(t *testing.T) {
-	env := func(key string) string {
-		return map[string]string{"ZERO_BENCH_COMMIT": "fromenv"}[key]
-	}
-	options, err := parseTaskArgs([]string{"--suite", "/s.json", "--model", "m", "--commit", "fromflag"}, env)
-	if err != nil {
-		t.Fatalf("parseTaskArgs error: %v", err)
-	}
-	if options.Commit != "fromflag" {
-		t.Fatalf("commit = %q, want fromflag (flag overrides env)", options.Commit)
-	}
-}
-
-func TestRunTasksCommandWritesRecordWithoutBinary(t *testing.T) {
-	// With --dry-run no agent is invoked: every task is recorded as skipped. This
-	// exercises the full record path (model + commit + self-correct flag) without
-	// needing a real zero binary or model.
-	suite := writeTaskSuite(t, `{
-		"id": "terminal-bench-mini",
-		"name": "Terminal-Bench (mini)",
-		"tasks": [
-			{"id": "t1", "name": "fix", "prompt": "do the thing"},
-			{"id": "t2", "name": "flag", "prompt": "add a flag"}
-		]
-	}`)
-	outPath := filepath.Join(t.TempDir(), "bench.json")
-
-	var stdout, stderr bytes.Buffer
-	code := runTasksCommand([]string{
-		"--suite", suite,
-		"--model", "test-model",
-		"--self-correct",
-		"--commit", "abc1234",
-		"--version", "1.2.3",
-		"--dry-run",
-		"--output", outPath,
-	}, emptyEnv, &stdout, &stderr)
-	if code != 0 {
-		t.Fatalf("runTasksCommand code = %d, stderr = %q", code, stderr.String())
-	}
-
-	data, err := os.ReadFile(outPath)
-	if err != nil {
-		t.Fatalf("read output: %v", err)
-	}
-	var record struct {
-		Model       string `json:"model"`
-		Commit      string `json:"commit"`
-		SelfCorrect bool   `json:"selfCorrect"`
-		Suite       string `json:"suite"`
-		Attempted   int    `json:"tasksAttempted"`
-	}
-	if err := json.Unmarshal(data, &record); err != nil {
-		t.Fatalf("decode record: %v\n%s", err, data)
-	}
-	if record.Model != "test-model" || record.Commit != "abc1234" || !record.SelfCorrect {
-		t.Fatalf("record = %#v", record)
-	}
-	if record.Suite != "terminal-bench-mini" || record.Attempted != 2 {
-		t.Fatalf("record suite/attempted = %q/%d", record.Suite, record.Attempted)
-	}
-	if !strings.Contains(stdout.String(), "self-correct: on") {
-		t.Fatalf("summary missing self-correct line:\n%s", stdout.String())
-	}
-}
-
-func TestRunTasksCommandRejectsMissingSuiteFile(t *testing.T) {
-	var stdout, stderr bytes.Buffer
-	code := runTasksCommand([]string{"--suite", "/no/such/suite.json", "--model", "m", "--dry-run"}, emptyEnv, &stdout, &stderr)
-	if code == 0 {
-		t.Fatalf("expected non-zero exit for missing suite file")
-	}
-	if stderr.Len() == 0 {
-		t.Fatalf("expected an error message on stderr")
-	}
-}
-
-func TestSampleTaskSuiteIsValidAndRunnableDry(t *testing.T) {
-	// The published example suite (referenced by docs/BENCHMARK.md) must load and
-	// dry-run cleanly so the documented command never bit-rots.
-	outPath := filepath.Join(t.TempDir(), "bench.json")
-	var stdout, stderr bytes.Buffer
-	code := runTasksCommand([]string{
-		"--suite", filepath.Join("testdata", "terminal-bench-sample.json"),
-		"--model", "example-model",
-		"--dry-run",
-		"--output", outPath,
-	}, emptyEnv, &stdout, &stderr)
-	if code != 0 {
-		t.Fatalf("sample suite dry run code = %d, stderr = %q", code, stderr.String())
-	}
-	if !strings.Contains(stdout.String(), "terminal-bench-sample") {
-		t.Fatalf("summary should name the suite:\n%s", stdout.String())
-	}
-}
-
-func TestTasksSubcommandRoutedFromMain(t *testing.T) {
-	var stdout, stderr bytes.Buffer
-	// "tasks --help" must route to the task harness help, not the perf-bench flags.
-	code := run([]string{"tasks", "--help"}, emptyEnv, &stdout, &stderr)
-	if code != 0 {
-		t.Fatalf("run tasks --help code = %d stderr = %q", code, stderr.String())
-	}
-	if !strings.Contains(stdout.String(), "zero-perf-bench tasks") {
-		t.Fatalf("tasks help missing usage:\n%s", stdout.String())
-	}
-}

cmd/zero-perf-bench/testdata/terminal-bench-sample.json

@@ -1,20 +0,0 @@
-{
-  "id": "terminal-bench-sample",
-  "name": "Terminal-Bench (sample)",
-  "tasks": [
-    {
-      "id": "hello-fix",
-      "name": "make the failing test pass",
-      "prompt": "The test in ./hello fails. Fix the implementation so `go test ./...` passes.",
-      "workspaceFixture": "./hello",
-      "verificationCommand": ["go", "test", "./..."]
-    },
-    {
-      "id": "add-json-flag",
-      "name": "add a --json flag",
-      "prompt": "Add a --json flag to the CLI that prints the result as JSON, and a test for it.",
-      "workspaceFixture": "./cli",
-      "verificationCommand": ["go", "test", "./..."]
-    }
-  ]
-}

cmd/zero-seccomp/main.go

@@ -1,35 +0,0 @@
-//go:build linux
-
-// Command zero-seccomp is retained as a compatibility wrapper for existing Linux
-// installs and scripts. The main sandbox path now applies the same optional
-// Unix-socket filter inside zero-linux-sandbox when sandbox.blockUnixSockets is
-// enabled.
-package main
-
-import (
-	"fmt"
-	"os"
-	"os/exec"
-	"syscall"
-
-	"github.com/Gitlawb/zero/internal/sandbox"
-)
-
-func main() {
-	if len(os.Args) < 2 {
-		fmt.Fprintln(os.Stderr, "usage: zero-seccomp <command> [args...]")
-		os.Exit(2)
-	}
-	if err := sandbox.ApplyUnixSocketBlock(); err != nil {
-		fmt.Fprintln(os.Stderr, "zero-seccomp: warning: "+err.Error()+"; running without the Unix-socket filter")
-	}
-	binary, err := exec.LookPath(os.Args[1])
-	if err != nil {
-		fmt.Fprintln(os.Stderr, "zero-seccomp: "+err.Error())
-		os.Exit(127)
-	}
-	if err := syscall.Exec(binary, os.Args[1:], os.Environ()); err != nil {
-		fmt.Fprintln(os.Stderr, "zero-seccomp: exec failed: "+err.Error())
-		os.Exit(126)
-	}
-}

cmd/zero-seccomp/main_unsupported.go

@@ -1,13 +0,0 @@
-//go:build !linux
-
-package main
-
-import (
-	"fmt"
-	"os"
-)
-
-func main() {
-	fmt.Fprintln(os.Stderr, "zero-seccomp is only supported on Linux")
-	os.Exit(2)
-}

cmd/zero-windows-command-runner/main.go

@@ -1,11 +0,0 @@
-package main
-
-import (
-	"os"
-
-	"github.com/Gitlawb/zero/internal/sandbox"
-)
-
-func main() {
-	os.Exit(sandbox.RunWindowsSandboxCommandRunner(os.Args[1:], os.Stderr))
-}

cmd/zero-windows-sandbox-setup/main.go

@@ -1,11 +0,0 @@
-package main
-
-import (
-	"os"
-
-	"github.com/Gitlawb/zero/internal/sandbox"
-)
-
-func main() {
-	os.Exit(sandbox.RunWindowsSandboxSetup(os.Args[1:], os.Stderr))
-}

docs/AGENT_EVALS.md

@@ -1,279 +0,0 @@
-# Offline Agent Evals
-
-Zero agent evals are maintainer fixtures for checking coding-agent behavior
-without calling a live model. They describe a task, the files the agent is
-expected to change, the commands that should verify the result, and the scoring
-rules an offline harness can apply to a captured run.
-
-These fixtures are intentionally local-first. They do not prove provider quality
-or live model execution by themselves; they give tests and CLI workflows a
-stable sample suite to validate, run against copied workspaces, and score from
-saved outputs. The eval harness is local and offline-testable. It only makes
-live model calls when the supplied agent command does.
-
-## Suite Format
-
-Sample suites live under `internal/agenteval/testdata/`. Tiny fixture
-workspaces live under `internal/agenteval/testdata/fixtures/`.
-
-Each suite JSON file contains:
-
-- `id`: stable suite identifier for filters and reports.
-- `name` and `description`: maintainer-facing suite metadata.
-- `tasks`: coding-agent tasks with prompts, file expectations, verification
-  commands, and offline scoring inputs.
-
-Task fields used by the sample suite:
-
-- `id`: stable task identifier for filters and reports.
-- `name` and `description`: short task metadata.
-- `tags`: stable category labels such as `docs`, `go`, or `wrapper`.
-- `difficulty`: a coarse task size such as `easy`, `medium`, or `hard`.
-- `prompt`: the user request to give an agent.
-- `workspaceFixture`: the fixture workspace to copy before running the task.
-- `expectedChangedFiles`: files that should change for a complete solution.
-- `forbiddenChangedFiles`: files that must not change during the task.
-- `requiredTraceEvents`: JSONL agent events that must appear in benchmark
-  stdout.
-- `contextChecks`: required and forbidden files checked in the materialized
-  workspace before verification commands run.
-- `verificationCommands`: commands a maintainer or harness can run after the
-  agent output is applied.
-
-The scoring contract matches command results by `verificationCommands[].id`,
-compares changed files against `expectedChangedFiles`, rejects any
-`forbiddenChangedFiles`, checks the materialized workspace with `contextChecks`,
-and can require agent trace events during benchmark runs. The loader rejects
-unknown JSON fields so suite changes fail fast.
-
-Example richer task rubric:
-
-```json
-{
-  "tags": ["docs", "jsonl"],
-  "difficulty": "easy",
-  "forbiddenChangedFiles": ["go.mod", "package.json"],
-  "requiredTraceEvents": ["tool:read_file", "verify:go-test"],
-  "contextChecks": {
-    "requiredFiles": ["docs/STREAM_JSON_PROTOCOL.md"],
-    "forbiddenFiles": ["node_modules/cache.txt"]
-  }
-}
-```
-
-## Modes
-
-### Validate Mode
-
-`zero eval` defaults to `validate` mode. In validate mode, the command performs
-schema and contract checks only: it parses the suite, rejects invalid task
-definitions, and reports the number of tasks and checks. It does not copy
-fixtures, invoke an agent, score a workspace, or execute verification commands.
-
-```bash
-go run ./cmd/zero eval --suite internal/agenteval/testdata/sample_suite.json
-```
-
-Use JSON output when another local tool needs the validation summary:
-
-```bash
-go run ./cmd/zero eval --suite internal/agenteval/testdata/sample_suite.json --json
-```
-
-### Run Mode
-
-`zero eval run` scores one already-mutated Git workspace. It does not copy
-fixtures or invoke an agent; point it at a Git worktree where a fixture has
-already been copied, initialized, and attempted by an agent or deterministic
-local script. The runner executes each `verificationCommands` entry, collects
-changed files with `git status --porcelain`, and emits the task-success report
-contract below. `--workspace` is required: it must point at the prepared fixture
-worktree, never the current directory, so the suite's verification commands
-(`go test`, `git`, ...) don't run against your real repo.
-
-```bash
-go run ./cmd/zero eval run \
-  --suite internal/agenteval/testdata/sample_suite.json \
-  --task document-stream-json-verify-events \
-  --workspace /tmp/zero-eval-workspace
-```
-
-Persist the report for comparison between prompt or model changes:
-
-```bash
-go run ./cmd/zero eval run \
-  --suite internal/agenteval/testdata/sample_suite.json \
-  --task document-stream-json-verify-events \
-  --workspace /tmp/zero-eval-workspace \
-  --report-dir /tmp/zero-eval-report \
-  --json
-```
-
-### Bench Mode
-
-`zero eval bench` runs the full benchmark harness for one task or a suite. Bench
-mode copies each task fixture into `--work-root`, initializes a clean Git
-baseline, runs the supplied `--agent-command` in that workspace, then scores the
-result with the same scorer used by run mode.
-
-Agent commands are passed as argv, without shell interpolation. The harness
-expands these placeholders in each argument:
-
-- `{workspace}`: copied task workspace path.
-- `{prompt}`: task prompt from the suite.
-- `{task_id}`: selected task ID.
-- `{model}`: current model ID for model-matrix benchmark runs.
-
-Use `--model <id>` more than once, or `--models a,b,c`, to run the same task
-matrix across several models. When no model is supplied, the harness preserves
-the previous single-run behavior and `{model}` expands to an empty string.
-
-Example using a real local agent command:
-
-```bash
-go run ./cmd/zero eval bench \
-  --suite internal/agenteval/testdata/sample_suite.json \
-  --task document-stream-json-verify-events \
-  --work-root /tmp/zero-evals \
-  --agent-command zero exec --cwd {workspace} {prompt}
-```
-
-Example with model selection:
-
-```bash
-go run ./cmd/zero eval bench \
-  --suite internal/agenteval/testdata/sample_suite.json \
-  --task document-stream-json-verify-events \
-  --work-root /tmp/zero-evals \
-  --model gpt-5 \
-  --agent-command zero exec --model {model} --cwd {workspace} {prompt}
-```
-
-Include `{task_id}` when the agent wrapper needs stable per-task logging,
-branching, or fixture-specific behavior:
-
-```bash
-go run ./cmd/zero eval bench \
-  --suite internal/agenteval/testdata/sample_suite.json \
-  --work-root /tmp/zero-evals \
-  --agent-command zero-agent-wrapper --task {task_id} --workspace {workspace} --prompt {prompt}
-```
-
-The same wrapper can emit JSONL trace events to stdout for future trace scoring:
-
-```json
-{"type":"tool","name":"read_file"}
-{"event":"verify","name":"go-test"}
-```
-
-Those events are required by adding keys such as `tool:read_file` and
-`verify:go-test` to `requiredTraceEvents`.
-
-For deterministic offline testing, point `--agent-command` at a local script
-that edits the copied workspace without calling a model:
-
-```bash
-go run ./cmd/zero eval bench \
-  --suite internal/agenteval/testdata/sample_suite.json \
-  --task document-stream-json-verify-events \
-  --work-root /tmp/zero-evals \
-  --agent-command ./scripts/fake-agent --workspace {workspace} --task {task_id} --prompt {prompt}
-```
-
-Bound a benchmark run with `--timeout` (a Go duration) so a wedged or
-interactive agent cannot block the harness forever. The timeout applies per
-task and cancels materialization, the agent process, and scoring:
-
-```bash
-go run ./cmd/zero eval bench \
-  --suite internal/agenteval/testdata/sample_suite.json \
-  --task document-stream-json-verify-events \
-  --work-root /tmp/zero-evals \
-  --timeout 5m \
-  --agent-command ./scripts/fake-agent --workspace {workspace} --prompt {prompt}
-```
-
-Use `--report-dir` to persist the CLI report artifact. The file is always named
-`agent-eval-report.json`; with bench mode it records the suite status, task
-counts, pass/fail totals, failures, and the nested benchmark report with each
-task/model run. Use `--keep-workspaces` when you also need to inspect the
-materialized workspaces after the run:
-
-```bash
-go run ./cmd/zero eval bench \
-  --suite internal/agenteval/testdata/sample_suite.json \
-  --task add-npm-wrapper-argv-helper \
-  --work-root /tmp/zero-evals \
-  --keep-workspaces \
-  --report-dir /tmp/zero-eval-report \
-  --json \
-  --agent-command ./scripts/fake-agent --workspace {workspace} --task {task_id} --prompt {prompt}
-```
-
-**Scoring caveat:** changed-file scoring inspects the workspace with
-`git status --porcelain` against the baseline commit. An agent that *commits*
-its own changes (or otherwise leaves a clean working tree) defeats this check —
-the committed edits no longer appear as changed files, so `expectedChangedFiles`
-will not match. Agents under bench should leave their edits uncommitted.
-
-Run the package tests when changing the suite schema or scorer:
-
-```bash
-go test ./internal/agenteval
-```
-
-For a faster manual fixture check:
-
-```bash
-go test ./internal/verify ./internal/selfverify
-```
-
-Or parse the JSON directly with any strict JSON parser. For example:
-
-```bash
-python -m json.tool internal/agenteval/testdata/sample_suite.json
-```
-
-The `internal/agenteval` tests load every JSON file under
-`internal/agenteval/testdata/` and reject missing task IDs, empty verification
-commands, and malformed changed-file expectations.
-
-## Report JSON
-
-Scored reports use contract `zero.agenteval.report.v1`.
-
-- `suiteId` and `taskId`: identify the suite and selected task.
-- `status`: overall `pass`, `fail`, `blocked`, or `error`.
-- `ok`: true only when every result passes.
-- `summary`: total result counts by status.
-- `changedFiles`: normalized files collected from the workspace.
-- `results`: one result per verification command, plus configured
-  `changed_files`, `forbidden_changed_files`, `context_checks`, and
-  `trace_events` checks.
-- `error`: task-selection or report-level error, when present.
-
-Command results include the command ID, display name, command argv, status,
-exit code, stdout, stderr, and an optional message. File-based results include
-expected, actual, missing, and unexpected files. Trace results include expected,
-actual, and missing event keys.
-
-## Score Interpretation
-
-Scores are offline quality signals, not pass/fail release gates by default. The
-statuses below are produced when a harness supplies captured command results and
-changed files.
-
-- `pass`: every verification command exited successfully, changed files matched
-  `expectedChangedFiles`, forbidden files stayed untouched, configured context
-  checks passed, and required trace events were present.
-- `fail`: at least one command failed, changed files were missing or
-  unexpected, a forbidden file changed, a context file check failed, or a
-  required trace event was missing.
-- `blocked`: the harness could not run the task or collect the expected inputs.
-- `error`: the suite, task ID, command ID, or captured input could not be
-  interpreted.
-
-Real task-success measurement comes from the combination of prompt, fixture,
-verification commands, and changed-file expectations. Prefer comparing results
-between runs of the same suite revision. Do not compare results across suites
-unless the task mix and scoring contract are unchanged.

docs/BENCHMARK.md

@@ -1,130 +0,0 @@
-# Zero Task Benchmark
-
-This is the **task** benchmark: how often ZERO completes a real coding task
-end-to-end, headless, unattended. It is separate from
-[`docs/PERFORMANCE.md`](PERFORMANCE.md), which measures process startup and
-memory, not task success.
-
-The methodology is the point, not the digit. A task-success score is **largely
-model-bounded** — most of the number comes from whichever model you bring. So we
-record the model with every result and we publish the score **with and without
-the self-correct loop**, because the delta between those two runs is the part
-ZERO actually contributes: the agent noticing its own broken edit and fixing it
-before it hands the task back.
-
-## What is recorded
-
-Every run produces a self-describing JSON record so a published number is
-reproducible and auditable from the record alone:
-
-| Field            | Meaning                                                       |
-| ---------------- | ------------------------------------------------------------- |
-| `suite`          | Task set id (which tasks produced the number)                 |
-| `model`          | The model that ran — the score is model-bounded, so this is required |
-| `mode`           | Exec mode preset, if any                                      |
-| `selfCorrect`    | Whether the post-edit verify-and-correct loop was enabled     |
-| `version`        | ZERO version                                                  |
-| `commit`         | ZERO commit the run was built from                            |
-| `date`           | UTC timestamp of the run                                      |
-| `tasksAttempted` | Tasks attempted                                               |
-| `tasksPassed`    | Tasks whose verification passed                               |
-| `passRate`       | `tasksPassed / tasksAttempted`                                |
-| `tasks`          | Per-task pass/fail/error with detail                          |
-
-## Integration point: headless `zero exec`
-
-The harness drives ZERO through its headless surface — the same path CI uses:
-
-```bash
-zero exec --output-format stream-json --model <model> [--self-correct] "<task prompt>"
-```
-
-Per task, the harness reads the terminal `run_end` event's exit code from the
-stream-json output to decide pass/fail. When a task carries a
-`verificationCommand` (e.g. `go test ./...`), that command's exit status is
-authoritative — mirroring Terminal-Bench's external-verifier model: the task is
-"passed" only when the project's own checks pass after the agent finishes.
-
-## Task set format
-
-A task set is a JSON manifest. A runnable sample lives at
-[`cmd/zero-perf-bench/testdata/terminal-bench-sample.json`](../cmd/zero-perf-bench/testdata/terminal-bench-sample.json):
-
-```json
-{
-  "id": "terminal-bench-sample",
-  "name": "Terminal-Bench (sample)",
-  "tasks": [
-    {
-      "id": "hello-fix",
-      "name": "make the failing test pass",
-      "prompt": "The test in ./hello fails. Fix the implementation so `go test ./...` passes.",
-      "workspaceFixture": "./hello",
-      "verificationCommand": ["go", "test", "./..."]
-    }
-  ]
-}
-```
-
-## The exact command
-
-Build the binary, then run the task harness **twice** against the same task set
-and the same model — once without self-correct, once with — and stamp the version
-and commit so the records are reproducible:
-
-```bash
-# build the production binary
-go run ./cmd/zero-release build
-
-VERSION=$(git describe --tags --always)
-COMMIT=$(git rev-parse --short HEAD)
-SUITE=cmd/zero-perf-bench/testdata/terminal-bench-sample.json
-MODEL=<your-model>
-
-# baseline: self-correct OFF
-go run ./cmd/zero-perf-bench tasks \
-  --suite "$SUITE" --binary ./zero --model "$MODEL" \
-  --version "$VERSION" --commit "$COMMIT" \
-  --output dist/bench/tasks-baseline.json
-
-# self-correct ON (auto-fix needs --auto medium or high; see note below)
-go run ./cmd/zero-perf-bench tasks \
-  --suite "$SUITE" --binary ./zero --model "$MODEL" --self-correct \
-  --version "$VERSION" --commit "$COMMIT" \
-  --output dist/bench/tasks-selfcorrect.json
-```
-
-`--version`/`--commit` also read from `ZERO_BENCH_VERSION` / `ZERO_BENCH_COMMIT`
-when the flags are omitted, so CI can stamp them once in the environment.
-
-Use `--dry-run` to exercise the record path without invoking a model (every task
-is recorded as skipped) — useful for validating a task set before a real run.
-
-> **Self-correct and autonomy.** `--self-correct` runs the verify-and-correct
-> loop after each mutating edit. Whether a detected failure is **auto-fixed** or
-> only **reported** is gated by the run's autonomy: pass `--auto medium` (or
-> `high`) for the loop to drive corrective rounds. At the default low autonomy it
-> reports failures without auto-fixing, so the with/without delta is measured at
-> `--auto medium` or higher.
-
-## Published result
-
-Fill in after a clean run on a fixed machine. Keep both records; the delta is the
-headline.
-
-| Run                     | Model     | Self-correct | Pass rate | Commit |
-| ----------------------- | --------- | ------------ | --------- | ------ |
-| Baseline                | _TBD_     | off          | _TBD_     | _TBD_  |
-| With self-correct       | _TBD_     | on           | _TBD_     | _TBD_  |
-| **Self-correct delta**  |           |              | **_TBD_** |        |
-
-Report the model alongside the number every time. A score without its model is
-not a claim about ZERO — it is a claim about the model. The honest signal is the
-delta: how much the self-correct loop moved the same model on the same tasks.
-
-## Reproducing a published number
-
-1. Check out the `commit` from the record.
-2. `go run ./cmd/zero-release build`.
-3. Run the two commands above with the recorded `model` and `suite`.
-4. Compare `passRate` in the new records against the published ones.

docs/EXTENDING.md

@@ -1,327 +0,0 @@
-# Extending Zero
-
-Zero is an open-source terminal coding agent. Out of the box it does the obvious things — read, edit, run, search — but the design point of the project is that **every surface is configurable**. This document is the user-facing guide for that configuration.
-
-If you only want to *use* Zero, the [README](../README.md) is enough. This page is for the other three jobs:
-
-1. Tell the agent about *your* project (drop an `AGENTS.md` in your repo).
-2. Add new specialist sub-agents.
-3. Wire Zero into the rest of your toolchain (MCP, skills, hooks, plugins).
-
-## 1. Drop a project `AGENTS.md`
-
-When Zero starts in a directory, it looks for project-level instructions and injects them into the system prompt. The lookup walks from your current working directory **up to the nearest git root** and reads the first matching file at each level — general rules at the repo root, more specific rules in sub-trees. Files are labeled with their directory in the prompt (e.g. `## Project guidelines (services/api/AGENTS.md)`).
-
-Accepted file names, in priority order at each level:
-
-| Path | Notes |
-| --- | --- |
-| `./AGENTS.md` | The classic spot — committed to your repo, shared with the team. |
-| `./ZERO.md` | Brand-specific alias. Same format, lower priority. |
-| `./.zero/AGENTS.md` | Project-local, hidden, gitignored. Personal notes that stay out of git. |
-
-Matching is **case-insensitive** on the basename, so `AGENTS.md`, `Agents.md`, and `agents.md` resolve to the same file on Windows and macOS. The git-tracked filename in this repo is `AGENTS.md` — keep that on case-sensitive filesystems (Linux, the WSL filesystem, or a CI runner) to match what the loader looks for.
-
-Both files use the same format. YAML frontmatter is optional; the markdown body is loaded as instructions for the agent. Zero reads the file once at session start, so changes take effect on the next `zero` launch — not mid-session.
-
-```markdown
-# Project conventions for <your project>
-
-- Build with `make`, not `go build` directly.
-- Tests live next to the source file (`foo_test.go` next to `foo.go`).
-- Run `make lint` before opening a PR.
-- Never edit files under `third_party/` — those are vendored.
-```
-
-Tips:
-
-- Keep each file under ~8 KiB. Zero caps the **total** across all matched files at 32 KiB; everything past the cap is dropped.
-- Re-state rules in the imperative voice: "Run `make lint`", not "you should consider running the linter".
-- Don't put secrets, model IDs, or environment-specific paths in `AGENTS.md`. Use `config.json` for those.
-- In a monorepo, drop a narrower `AGENTS.md` in each sub-tree (e.g. `services/api/AGENTS.md`). Zero picks those up automatically when you launch from inside the sub-tree.
-- A YAML frontmatter block (`---\n...\n---`) at the top is preserved verbatim in the injected prompt but is not parsed for `globs:` or `alwaysApply:` scoping today — keep the body self-contained.
-
-### Personal guidelines, across every project
-
-For preferences that follow *you*, not a specific repo (tone, tooling habits, workflow), drop a `ZERO.md` in your user config directory: `~/.config/zero/ZERO.md` on Linux/macOS, `%AppData%\zero\ZERO.md` on Windows — the same directory as `config.json` and your personal specialists. Same format and 8 KiB cap as the project files above, and the same case-insensitive basename match.
-
-This file is injected as its own `## User guidelines` section, before the project's `AGENTS.md`/`ZERO.md`, and is labeled as personal preference in the prompt: project guidelines are the later, more specific instruction and take precedence over it when the two conflict.
-
-## 2. Custom specialists
-
-Specialists are Zero's sub-agents. Three scopes, in priority order:
-
-| Scope | Path | Shared? |
-| --- | --- | --- |
-| Built-in | compiled into Zero | yes — `worker`, `explorer`, `code-review` |
-| User | `~/.config/zero/specialists/*.md` | no — your machine only |
-| Project | `./.zero/specialists/*.md` | yes — the repo team |
-
-Project overrides user overrides built-in when names collide.
-
-A specialist is a markdown manifest with frontmatter and a system prompt:
-
-```markdown
----
-description: Reviews API changes for breaking-change risk and missing tests.
-tools: read-only,plan
----
-
-You review API changes. For every changed hunk in `internal/api/` or any file
-that ends in `_api.go`:
-
-1. Confirm the public signature is backward-compatible, or note the breaking
-   change explicitly with the migration path.
-2. Confirm a corresponding test exists in `internal/api/*_test.go` and that
-   the new behaviour is exercised.
-3. Flag any new exported symbol without a doc comment.
-
-Reply with one JSON object per finding: `{"file", "line", "severity", "message", "fix"}`.
-```
-
-CLI management (the prompt is passed inline via `--prompt`):
-
-```bash
-zero specialist list
-zero specialist show api-reviewer
-zero specialist create api-reviewer \
-    --project \
-    --description "Reviews API changes" \
-    --tools read-only,plan \
-    --prompt "$(cat api-reviewer.md)"
-zero specialist edit api-reviewer --project
-zero specialist delete api-reviewer --project
-zero specialist path                       # prints the resolved specialists directory
-```
-
-The full format spec (frontmatter fields, tool scopes, prompt conventions) is in [`docs/SPECIALISTS.md`](SPECIALISTS.md).
-
-> **Roadmap.** An in-UI specialist manager (create / edit / delete / preview) is on the backlog. Today you use the `zero specialist` CLI subcommands above.
-
-## 3. Skills
-
-Skills are markdown instruction packs the agent can pull in on demand. Each skill is a directory containing a `SKILL.md`. Standalone project skill directories are not supported in this version (shared project-wide skills must go in `AGENTS.md` or as a hook). However, project plugins (section 6) may bundle skills, which are merged into the active run.
-
-Discovery root: `$ZERO_SKILLS_DIR` → `$XDG_DATA_HOME/zero/skills` → `~/.local/share/zero/skills/`. A missing directory is fine — Zero just reports "no skills".
-
-```text
-~/.local/share/zero/skills/
-  run-benchmarks/
-    SKILL.md
-  write-changelog/
-    SKILL.md
-```
-
-`SKILL.md` format:
-
-```markdown
----
-description: Run the project's benchmark suite and summarize the deltas.
----
-
-# Run benchmarks
-
-1. `make bench` — captures the wall-clock and RSS before and after.
-2. `benchstat before.txt after.txt` — diffs the two.
-3. Report any regression > 5% with the function name and the previous value.
-```
-
-Only `name` and `description` are recognized in the frontmatter today. The `name` defaults to the directory name. Within a single skills root, duplicate names are resolved by lexicographic directory order. During an agent run, Zero loads the default skills directory before plugin skill roots; earlier roots win name collisions silently. Plugin-declared skills (section 6) are merged into the active agent run at plugin activation time, so bundled skills appear in the available skills list and can be loaded with the `skill` tool.
-
-The `skill` core tool lets the agent load any discovered skill by name.
-
-## 4. Hooks
-
-Hooks fire shell commands on lifecycle events. Configure them in JSON:
-
-- User: `~/.config/zero/hooks.json`
-- Project: `./.zero/hooks.json`
-
-```json
-{
-  "enabled": true,
-  "hooks": [
-    {
-      "id": "block-rm-rf",
-      "event": "beforeTool",
-      "matcher": "bash",
-      "command": "/usr/local/bin/zero-hook-block-rmrf.sh",
-      "enabled": true
-    },
-    {
-      "id": "log-session",
-      "event": "sessionStart",
-      "command": "/usr/local/bin/zero-hook-log.sh",
-      "enabled": true
-    }
-  ]
-}
-```
-
-The `args` array (when present) is passed verbatim to `exec.CommandContext`. The actual hook payload — event name, matcher, tool call id, tool name, tool input, tool output, status — is delivered to the command as **JSON on stdin**, not via `${...}` substitution. A typical handler reads stdin and decides what to do:
-
-```bash
-#!/usr/bin/env bash
-# /usr/local/bin/zero-hook-block-rmrf.sh
-set -euo pipefail
-payload="$(cat)"
-if printf '%s' "$payload" | grep -q '"input":"[^"]*rm[[:space:]]+-rf'; then
-  echo "refusing rm -rf" >&2
-  exit 1
-fi
-```
-
-Events the agent emits (in dispatch order):
-
-| Event | Fires when | Matcher allowed? |
-| --- | --- | --- |
-| `beforeTool` | A tool is about to run | yes (tool name) |
-| `afterTool` | A tool just returned | yes (tool name) |
-| `sessionStart` | A session begins | no |
-| `sessionEnd` | A session ends | no |
-| `specialistStart` | A sub-agent is spawned | yes (specialist name) |
-| `specialistStop` | A sub-agent ends | yes (specialist name) |
-
-A hook's exit code decides what happens next: `0` continues, non-zero blocks the tool call (`beforeTool`) or surfaces an error (`afterTool`). Hook execution is recorded in the audit log; the audit is reachable from the agent's view of past actions, not from a dedicated `zero doctor` check.
-
-> **Roadmap.** An in-UI hooks manager is on the backlog. Today you edit the JSON directly.
-
-## 5. MCP — Model Context Protocol
-
-Zero is both an **MCP client** (it can call external MCP servers) and an **MCP server** (other agents can call its tools).
-
-### As a client — configure MCP servers in `config.json`
-
-```json
-{
-  "mcp": {
-    "servers": {
-      "docs": {
-        "type": "stdio",
-        "command": "docs-mcp",
-        "args": ["--port", "7777"]
-      },
-      "github": {
-        "type": "http",
-        "url": "https://api.example.com/mcp",
-        "headers": { "Authorization": "Bearer YOUR_TOKEN_HERE" }
-      }
-    }
-  }
-}
-```
-
-Manage via CLI:
-
-```bash
-zero mcp add docs --type stdio -- docs-mcp --port 7777
-zero mcp add github --type http --url https://api.example.com/mcp \
-    --header "Authorization=Bearer YOUR_TOKEN_HERE"
-zero mcp list
-zero mcp check docs
-zero mcp remove github
-zero mcp oauth login github
-```
-
-Servers are merged from user and project configs (project wins on conflicts). Token-bearing values in `config.json` are sent verbatim — there is no `${env:...}` expansion — so prefer one of:
-
-- A wrapper script that sources the secret and execs the real command.
-- A `--header` value produced by command substitution (`"Authorization=Bearer $(print-token)"`) in a private shell config that you keep out of git.
-- A secret manager that injects the env var your MCP server reads on its own (the `command` and `args` then run inside that environment).
-
-Project MCP config remains supported for shared team servers, but a project layer
-that changes an existing server's target (`url`, `command`, or `args`) does not
-inherit user credentials. Clear or replace `headers`, `env`, or `oauth` explicitly,
-or use a new server name. OAuth tokens are bound to the resolved server identity,
-so changing a server target can require `zero mcp oauth login <server>` again.
-OAuth server names ending in `.<32 hex chars>` are reserved for token storage.
-
-### As a server — expose Zero's tools to another agent
-
-```bash
-zero serve --mcp
-```
-
-The server speaks MCP over stdio. Configure it from the receiving side as a `stdio` server whose command is `zero serve --mcp`.
-
-## 6. Plugins
-
-A plugin is a self-contained directory that bundles tools, hooks, and skills for one capability. Plugins live at:
-
-- User: `~/.config/zero/plugins/<id>/`
-- Project: `./.zero/plugins/<id>/`
-
-Each plugin has a `plugin.json` manifest:
-
-```json
-{
-  "id": "github-pr-review",
-  "name": "GitHub PR Review",
-  "description": "Adds review tools for GitHub PRs.",
-  "version": "1.0.0",
-  "tools": [
-    { "name": "list_prs", "command": "./tools/list_prs.sh" }
-  ],
-  "hooks": [
-    { "name": "pre-merge-check", "event": "beforeTool", "command": "./hooks/pre-merge.sh" }
-  ],
-  "skills": [
-    { "path": "./skills/review-checklist/SKILL.md" }
-  ]
-}
-```
-
-Install and manage:
-
-```bash
-zero plugins add ./github-pr-review      # copy into ~/.config/zero/plugins/ or ./.zero/plugins/
-zero plugins list
-zero plugins remove github-pr-review    # alias: rm
-```
-
-A plugin is enabled by being present in the plugins directory and disabled by removing it (or by the user setting `"enabled": false` in its `plugin.json`). Plugins are not enabled or disabled by a CLI subcommand today.
-
-Plugin commands run with the plugin directory as their working directory. Use relative paths; the loader resolves them at activation time.
-
-> **Roadmap.** An in-UI plugins manager (browse, install, enable / disable) is on the backlog. Today you use the `zero plugins` CLI subcommands above.
-
-## 7. Configuration locations
-
-Five configuration sources, in precedence order (later sources override earlier ones):
-
-| Source | Path / Key | Notes |
-| --- | --- | --- |
-| Built-in defaults | compiled in | Lowest priority. |
-| User config | `~/.config/zero/config.json` | Your machine. Never committed. |
-| Project config | `./.zero/config.json` | The repo. Committed (or not, your call). |
-| Environment | `ZERO_*` | Provider commands, secrets, skills dir override. |
-| CLI flags | `--model`, `--mode`, ... | Highest priority, per-invocation. |
-
-The user config holds things that should follow the user across projects (default provider, default model, theme). The project config holds things the team agreed on (provider catalog, sandbox policies, model restrictions).
-
-The sandbox `additionalWriteRoots` key is **ignored in project config** by design — a checked-out repo cannot widen its own sandbox. Set it in the user config or pass `--add-dir` per-invocation.
-
-## 8. End-to-end example
-
-A team that wants every contributor's Zero to behave the same way commits:
-
-- `AGENTS.md` — project conventions, build commands, do-not-edit lists.
-- `.zero/config.json` — provider catalog, default model, allowed tools.
-- `.zero/specialists/api-reviewer.md` — the team's PR-review specialist.
-- `.zero/hooks.json` — block `rm -rf` on `beforeTool`.
-- `.zero/plugins/internal-tooling/` — a plugin that adds the team's internal CLI tools to the agent's toolset.
-
-Each contributor adds only:
-
-- `~/.config/zero/config.json` — their personal API keys, theme, default mode.
-- `~/.config/zero/ZERO.md` — personal preferences that follow them across every project (see section 1).
-- `~/.local/share/zero/skills/` — personal skills they keep across projects.
-
-That's it. Run `zero` from the repo root and the agent has the team's full instruction set, every contributor's personal setup, and nothing else.
-
-## 9. Reference
-
-- [README](../README.md) — install, quickstart, command reference.
-- [docs/SPECIALISTS.md](SPECIALISTS.md) — full specialist manifest spec.
-- [docs/STREAM_JSON_PROTOCOL.md](STREAM_JSON_PROTOCOL.md) — `zero exec` I/O contract.
-- [docs/INSTALL.md](INSTALL.md) — install from source or release.

docs/GITHUB_ACTION.md

@@ -1,203 +0,0 @@
-# ZERO GitHub Action
-
-Run ZERO headlessly inside a GitHub workflow. The action is a thin wrapper around
-`zero exec`: it installs a pinned ZERO release, runs your prompt in the checked-out
-repository, surfaces ZERO's exit code as the step status, captures the output as a
-file, and can optionally post a summary to the triggering pull request or to Slack.
-
-ZERO is model- and provider-agnostic, and so is this action: **you** choose the
-provider and supply the API key. Nothing is hardcoded to any single provider.
-
-## Quick start
-
-```yaml
-# .github/workflows/zero.yml
-name: ZERO
-on:
-  workflow_dispatch:
-    inputs:
-      task:
-        description: What should ZERO do?
-        required: true
-
-permissions:
-  contents: write
-
-jobs:
-  run:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-      - uses: Gitlawb/zero@v1
-        with:
-          prompt: ${{ inputs.task }}
-          provider: openai
-          api-key-env: OPENAI_API_KEY
-          api-key: ${{ secrets.OPENAI_API_KEY }}
-          model: gpt-4.1
-```
-
-> The `provider`, `api-key-env`, and `api-key` trio is how you stay
-> provider-neutral: `api-key-env` is the environment variable name the chosen
-> provider reads its key from (for example `ANTHROPIC_API_KEY`, `GEMINI_API_KEY`,
-> `OPENROUTER_API_KEY`), and `api-key` is the secret value. The action exports
-> that variable only for the ZERO step and never prints it. If your repository
-> already commits a `.zero/config.json` with an active provider, you can omit
-> `provider`.
-
-## Inputs
-
-| Input | Required | Default | Description |
-| --- | --- | --- | --- |
-| `prompt` | one of `prompt`/`prompt-file` | `""` | The instruction for ZERO to execute. |
-| `prompt-file` | one of `prompt`/`prompt-file` | `""` | Path (relative to `working-directory`) to a file whose contents are the prompt. |
-| `provider` | no | `""` | Provider id to activate (e.g. `openai`, `anthropic`, `gemini`, `ollama`, or any compatible endpoint). |
-| `api-key` | no | `""` | The provider API key. Pass from a secret; exported only for the ZERO step and never logged. |
-| `api-key-env` | no | `""` | Env var name the provider reads its key from (e.g. `OPENAI_API_KEY`). Exported only for the ZERO step when used with `api-key`. |
-| `model` | no | `""` | Model id. Defaults to the resolved provider's default. |
-| `mode` | no | `""` | Run mode (`zero exec --mode`), e.g. `smart`, `deep`, `fast`. |
-| `auto` | no | `low` | Autonomy ceiling (`zero exec --auto`): `low`, `medium`, or `high`. Conservative by default. |
-| `self-correct` | no | `false` | Allow mid-run model escalation (`zero exec --allow-escalation`). |
-| `add-dir` | no | `""` | Newline-/comma-separated extra write roots (`zero exec --add-dir`). |
-| `worktree` | no | `false` | Run in an isolated git worktree (`zero exec --worktree`). |
-| `output-format` | no | `stream-json` | `text`, `json`, or `stream-json`. Captured to a file. |
-| `post-to` | no | `none` | `pr-comment`, `slack`, or `none`. Where to post a summary after the run. |
-| `slack-webhook-url` | no | `""` | Slack incoming-webhook (or generic webhook) URL for `post-to: slack`. Pass from a secret. |
-| `github-token` | no | `${{ github.token }}` | Token used to post a PR comment. Requires `pull-requests: write`. |
-| `working-directory` | no | `${{ github.workspace }}` | Directory to run ZERO in. |
-| `zero-version` | no | (action ref → `latest`) | ZERO release version/tag to install, e.g. `v1.2.3` or `latest`. |
-| `zero-repo` | no | `Gitlawb/zero` | Repository to install the ZERO release from. |
-
-## Outputs
-
-| Output | Description |
-| --- | --- |
-| `exit-code` | ZERO's exit code (`0` success, `2` usage, `3` provider, non-zero otherwise). |
-| `output-file` | Path to the captured ZERO stdout (the raw `output-format` stream). |
-| `summary` | A short, single-line summary parsed from the run, when available. |
-
-The step **fails when ZERO returns a non-zero exit code**, so a failed run fails
-the job by default. Use `continue-on-error: true` on the step (and read
-`exit-code`) if you want to handle failures yourself.
-
-## Examples
-
-### Run ZERO on every issue labeled `zero`
-
-```yaml
-name: ZERO issue triage
-on:
-  issues:
-    types: [labeled]
-
-permissions:
-  contents: write
-  issues: write
-  pull-requests: write
-
-jobs:
-  triage:
-    if: github.event.label.name == 'zero'
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-      - uses: Gitlawb/zero@v1
-        with:
-          prompt: |
-            Investigate this issue and propose a fix.
-
-            Title: ${{ github.event.issue.title }}
-
-            ${{ github.event.issue.body }}
-          provider: anthropic
-          api-key-env: ANTHROPIC_API_KEY
-          api-key: ${{ secrets.ANTHROPIC_API_KEY }}
-          auto: low
-          post-to: slack
-          slack-webhook-url: ${{ secrets.ZERO_SLACK_WEBHOOK_URL }}
-```
-
-### Nightly dependency-upgrade PR
-
-```yaml
-name: ZERO nightly deps
-on:
-  schedule:
-    - cron: "0 6 * * 1" # Mondays 06:00 UTC
-
-permissions:
-  contents: write
-  pull-requests: write
-
-jobs:
-  deps:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-      - uses: Gitlawb/zero@v1
-        id: zero
-        with:
-          prompt-file: .github/zero/upgrade-deps.md
-          provider: openai
-          api-key-env: OPENAI_API_KEY
-          api-key: ${{ secrets.OPENAI_API_KEY }}
-          worktree: true
-          auto: medium
-      - name: Open pull request
-        run: |
-          git checkout -b zero/deps-$(date +%Y%m%d)
-          git commit -am "chore(deps): nightly upgrade via ZERO" || exit 0
-          git push -u origin HEAD
-          gh pr create --fill --label dependencies
-        env:
-          GH_TOKEN: ${{ github.token }}
-      - name: Upload ZERO output
-        if: always()
-        uses: actions/upload-artifact@v4
-        with:
-          name: zero-output
-          path: ${{ steps.zero.outputs.output-file }}
-          if-no-files-found: warn
-```
-
-## Security notes
-
-- **Secrets are passed as Action secrets and never logged.** The action exports
-  `api-key` (under `api-key-env`) only for the ZERO step. `slack-webhook-url`
-  is passed only to the Slack post step. They are never echoed or persisted to
-  later workflow steps, and ZERO redacts secret-shaped strings from its own
-  output.
-- **The sandbox is always active.** This action never passes
-  `--skip-permissions-unsafe`, so writes stay inside the checked-out repository
-  (plus any roots you grant with `add-dir`). Unsafe mode is never enabled
-  implicitly.
-- **Autonomy defaults to `low`.** Raise it deliberately (`auto: medium`/`high`)
-  only for tasks you trust to run unattended.
-- **Least-privilege tokens.** Grant only the permissions the workflow needs
-  (`contents: write` to edit files, `pull-requests: write` to comment). The
-  default `GITHUB_TOKEN` is scoped to the repository.
-- **Pin the action.** Reference a tag (`Gitlawb/zero@v1`) or a commit SHA so a
-  workflow run uses a known ZERO version. `zero-version` lets you pin the
-  installed binary independently of the action ref.
-- **Linux and macOS runners** are supported; Windows runners are rejected with a
-  clear error.
-
-## Slack / webhook notifier
-
-The action's `post-to: slack` step sends a one-line summary to a Slack incoming
-webhook after the run. ZERO also has a built-in webhook notifier sink
-(`internal/notify`) that an unattended run can use to report
-"finished / needs input / verify failed after N retries" to Slack or any generic
-webhook:
-
-- Configure the destination with the `ZERO_SLACK_WEBHOOK_URL` environment variable
-  (or settings). A blank URL disables the sink.
-- The sink POSTs a JSON body `{ "text", "type", "message", "summary?", "links?" }`.
-  The `text` field is what Slack renders; the structured fields carry the
-  machine-readable detail.
-- **Fail-soft:** a non-2xx response or a transport error is logged (redacted) and
-  swallowed — a webhook problem never crashes the run.
-- **Redaction:** the message, summary, links, and the webhook URL itself are run
-  through ZERO's redaction before being sent or logged, so tokens never leak.
-- **Egress/proxy:** the notifier uses the default HTTP transport, which honors
-  `HTTP_PROXY`/`HTTPS_PROXY` when a proxy is configured.

docs/HOW_ZERO_WORKS.md

@@ -1,787 +0,0 @@
-# How Zero Works
-
-This document explains Zero's architecture at a high level: the layers, module
-boundaries, data flow, and runtime concepts behind the interactive TUI and the
-headless `zero exec` mode.
-
-It is written as a progressive map: each high-level section introduces the
-concepts you need for orientation, then points to the later section that explains
-the mechanics in more detail.
-
-## Big Picture
-
-Zero is a Go CLI application with two main user-facing surfaces:
-
-- **Interactive TUI** (`zero`): a Bubble Tea application that renders chat,
-  tool cards, permission prompts, model/provider pickers, and session controls.
-- **Headless exec** (`zero exec ...`): a scriptable command runner that uses the
-  same agent loop and tools, but writes text, JSON, or stream-JSON events.
-
-Both surfaces converge on the same core agent runtime:
-
-```mermaid
-flowchart TD
-    User[User / automation] --> CLI[cmd/zero -> internal/cli]
-
-    CLI --> TUI[Interactive TUI<br/>internal/tui]
-    CLI --> Exec[Headless exec<br/>internal/cli/exec.go]
-    CLI --> ACP[ACP / editor bridge<br/>internal/acp]
-
-    TUI --> Agent[Agent loop<br/>internal/agent]
-    Exec --> Agent
-    ACP --> Agent
-
-    Agent --> Provider[Provider runtime<br/>internal/zeroruntime.Provider]
-    Provider --> APIs[OpenAI / Anthropic / Gemini / compatible APIs]
-
-    Agent --> Registry[Tool registry<br/>internal/tools.Registry]
-    Registry --> CoreTools[Core tools<br/>read/edit/shell/search/plan]
-    Registry --> MCP[MCP tools<br/>internal/mcp]
-    Registry --> Specialists[Specialist tools<br/>internal/specialist]
-    Registry --> Plugins[Plugin tools<br/>internal/plugins]
-    Registry --> LocalControl[Browser / terminal / desktop helpers]
-
-    Agent --> Sandbox[Permission + sandbox engine<br/>internal/sandbox]
-    Agent --> Hooks[Hooks dispatcher<br/>internal/hooks]
-    Agent --> Sessions[Session event store<br/>internal/sessions]
-```
-
-The important split is:
-
-- `internal/tui` is the **presentation layer**. For the mechanics, see
-  [Presentation Layer: Bubble Tea TUI](#presentation-layer-bubble-tea-tui) and
-  [How the TUI and Agent Loop Connect](#how-the-tui-and-agent-loop-connect).
-- `internal/agent` is the **reasoning/tool loop**. For the per-turn mechanics,
-  see [Agent Loop](#agent-loop), [Tool Execution Lifecycle](#tool-execution-lifecycle),
-  and [Context Budget and Compaction](#context-budget-and-compaction).
-- `internal/tools`, `internal/sandbox`, and provider packages are the **runtime
-  capability layer**. For the details, see [Runtime Provider Layer](#runtime-provider-layer),
-  [Tools and Capabilities](#tools-and-capabilities), and
-  [Permission and Sandbox Model](#permission-and-sandbox-model).
-- `internal/sessions` is the **durable event log** used for resume, fork,
-  rewind, sub-agents, and history. See [Session Storage and History](#session-storage-and-history).
-
-## Architecture Spine
-
-The easiest way to understand Zero is not as a set of independent packages, but
-as one spine with several mechanisms attached to it:
-
-```mermaid
-flowchart LR
-    Surface[Surface<br/>TUI / exec / ACP] --> Run[Run request<br/>prompt + runtime options]
-    Run --> Agent[Agent loop<br/>turns, tools, recovery]
-    Agent --> Capabilities[Capabilities<br/>providers + tools + sandbox]
-    Capabilities --> Events[Runtime events<br/>text, tool calls, permissions, usage]
-    Events --> Surface
-    Events --> Sessions[Session log<br/>resume, fork, rewind, history]
-
-    Config[Config / guidelines / skills] --> Run
-    Extensions[MCP / specialists / plugins / hooks] --> Capabilities
-```
-
-The **TUI + agent loop** is the main story because it explains the normal user
-experience: type a prompt, watch the model stream, approve tool use, see file
-changes, and receive a final answer. Other mechanisms are still important, but
-most of them attach to the same spine:
-
-| Mechanism | Where it attaches | What it changes |
-| --- | --- | --- |
-| Config, provider profiles, model registry | Before a run starts | Which provider/model is used and what defaults apply. |
-| Project/user guidelines, skills, specialists context | Prompt construction | What instructions and available capabilities the model sees. |
-| Tool registry, MCP, plugins, local-control helpers | Capability layer | Which actions the model may request. |
-| Permission policy, sandbox, command-prefix grants | Tool execution | Whether a requested action can run, needs approval, or is denied. |
-| Hooks | Tool lifecycle | Optional before/after observers or blockers around tool calls. |
-| Sessions, checkpoints, compaction metadata | Event persistence/context | What can be resumed, forked, rewound, or summarized for later turns. |
-| Stream-JSON | Surface layer | How the same runtime events are emitted for automation instead of rendered in the TUI. |
-
-So this document focuses first on the TUI/agent-loop bridge, then explains the
-supporting mechanisms in terms of where they connect to that bridge.
-
-Use this as the reading path:
-
-| If you see this concept in a high-level flow | Read this detail section |
-| --- | --- |
-| Surface submits a run | [Startup Flow](#startup-flow), then [How the TUI and Agent Loop Connect](#how-the-tui-and-agent-loop-connect) |
-| Agent builds context and advances turns | [Agent Loop](#agent-loop) |
-| “Maybe compact context” | [Context Budget and Compaction](#context-budget-and-compaction) |
-| Model requests a tool | [Tool Execution Lifecycle](#tool-execution-lifecycle), then [Tools and Capabilities](#tools-and-capabilities) |
-| Tool needs approval or sandbox access | [Permission and Sandbox Model](#permission-and-sandbox-model) |
-| Events appear in the UI or stream-JSON | [How the TUI and Agent Loop Connect](#how-the-tui-and-agent-loop-connect), [Stream-JSON Protocol](#stream-json-protocol) |
-| Resume, fork, rewind, or sub-agent history | [Session Storage and History](#session-storage-and-history) |
-
-## Main Packages
-
-The sections above describe the architecture in concepts. If you want to connect
-those concepts to implementation details, use this table as the code map before
-reading the deeper flows below.
-
-| Package | Role |
-| --- | --- |
-| `cmd/zero` | Tiny binary entrypoint; calls `internal/cli.Run`. |
-| `internal/cli` | Command parsing, config resolution, provider creation, registry setup, TUI/exec/ACP launch paths. |
-| `internal/tui` | Bubble Tea model/update/view code, transcript rendering, composer, prompts, slash commands, setup wizard. |
-| `internal/agent` | Core loop that sends messages to the model, streams output, executes tools, handles permissions, compaction, retries, and final answers. |
-| `internal/zeroruntime` | Provider-neutral message, tool-call, stream, usage, and image types. Providers implement `Provider.StreamCompletion`. |
-| `internal/providers` | Factory that turns a resolved provider profile into an OpenAI, Anthropic, Gemini, Codex, or compatible provider adapter. |
-| `internal/tools` | Tool interface, registry, core read/write/shell/network tools, local-control wrappers, tool safety metadata, redaction, output ceilings. |
-| `internal/sandbox` | Permission policy, path/network checks, grant store, command-prefix approvals, and platform sandbox backends. |
-| `internal/sessions` | Local append-only session storage: `metadata.json` + `events.jsonl`. |
-| `internal/config` | User/project config resolution, active provider, preferences, sandbox/tool settings. |
-| `internal/mcp` | MCP server config, client runtime, permission store, and MCP tool registration. |
-| `internal/specialist` / `internal/swarm` | Sub-agent manifests and team/member orchestration exposed as tools. |
-| `internal/localcontrol` / `internal/browser` | Runtime helpers used by config-gated local-control tool wrappers. |
-| `internal/plugins` / `internal/skills` / `internal/hooks` | Extension surfaces loaded into the agent context, tool registry, or tool lifecycle. |
-| `internal/streamjson` | Machine-readable protocol for `zero exec --input-format/--output-format stream-json`. |
-
-## Startup Flow
-
-### Interactive TUI (`zero`)
-
-```mermaid
-sequenceDiagram
-    participant Main as cmd/zero
-    participant CLI as internal/cli
-    participant Config as internal/config
-    participant TUI as internal/tui
-    participant Agent as internal/agent
-
-    Main->>CLI: Run(args)
-    CLI->>Config: resolve config for workspace
-    CLI->>CLI: build provider
-    CLI->>CLI: build tool registry
-    CLI->>CLI: register core, config-gated local-control, specialists, MCP, plugins
-    CLI->>CLI: create sandbox engine + session store
-    CLI->>TUI: run Bubble Tea with Options
-    TUI->>TUI: Init / Update / View event loop
-    TUI->>Agent: agent.Run on submitted prompt
-```
-
-The TUI owns terminal interaction. It tracks composer state, transcript rows,
-scrolling, permission dialogs, model/provider pickers, setup wizard state, and
-visual updates. When the user submits a prompt, the TUI builds `agent.Options`
-with callbacks like `OnText`, `OnToolCall`, `OnToolResult`, `OnPermission`,
-`OnAskUser`, and `OnUsage`. Those callbacks convert runtime events into visible
-transcript rows and session events.
-
-### Headless exec (`zero exec`)
-
-```mermaid
-flowchart TD
-    A[Parse exec flags] --> B[Resolve config + provider]
-    B --> C[Build tool registry]
-    C --> D[Resolve prompt / stream-json input / images]
-    D --> E[Prepare or resume session]
-    E --> F[Create stream-json/text/json writer]
-    F --> G[agent.Run]
-    G --> H[Write final output and exit code]
-```
-
-`zero exec` uses the same `agent.Run` loop as the TUI, but without interactive
-terminal state. It can persist sessions, resume/fork existing sessions, run in
-stream-JSON mode for automation, and always requires a stronger completion
-signal so a no-tool assistant response is not treated as success while work is
-still pending.
-
-## Agent Loop
-
-The core loop lives in `internal/agent.Run`. Conceptually, every turn follows
-this cycle:
-
-```mermaid
-flowchart TD
-    Start[Build system prompt + user prompt] --> Expose[Partition visible tools]
-    Expose --> Compact[Maybe compact context]
-    Compact --> Request[Send CompletionRequest to provider]
-    Request --> Stream[Collect streamed text, reasoning, usage, tool calls]
-
-    Stream --> HasTools{Tool calls?}
-    HasTools -- No --> FinalCheck{Final answer acceptable?}
-    FinalCheck -- Yes --> Done[Return Result]
-    FinalCheck -- No --> Nudge[Append user nudge and continue]
-
-    HasTools -- Yes --> Execute[Execute tool calls]
-    Execute --> Permission[Permission / sandbox / hooks]
-    Permission --> ToolResult[Append tool result messages]
-    ToolResult --> Diagnostics[Post-edit diagnostics / self-correct]
-    Diagnostics --> Expose
-```
-
-Key responsibilities of the loop:
-
-1. **Build prompt state** — combines the system prompt, project/user guidelines,
-   loaded skills/specialist instructions, the current user prompt, and images. The
-   resulting message list is the model-facing context that later tool results are
-   appended to. Resumed session history is replayed by the caller before a new run,
-   not passed as a separate `agent.Run` history field.
-2. **Expose tools** — starts from the registry built by the CLI, then filters
-   tool definitions by permission mode, enabled or disabled filters, and
-   deferred-loading rules. Tool schemas also count toward context budget, so they
-   matter for compaction decisions.
-3. **Call provider** — sends a provider-neutral `zeroruntime.CompletionRequest`.
-4. **Stream events** — forwards text, reasoning, tool-call deltas, and usage to
-   the caller via callbacks.
-5. **Execute tools** — decodes tool arguments, checks filters, applies
-   permissions and sandbox policy, dispatches hooks, runs the registered tool,
-   redacts output, and appends a tool message. See
-   [Tool Execution Lifecycle](#tool-execution-lifecycle).
-6. **Recover and guard** — handles malformed tool calls, repeated failures,
-   empty turns, context pressure, stream reconnect/retry, and max-turn fallback.
-   Context pressure is handled by [Context Budget and Compaction](#context-budget-and-compaction).
-7. **Finalize** — returns a final answer or a structured incomplete/stop reason.
-
-The model never directly mutates the workspace. It asks for tool calls; Zero
-validates and executes those calls through the registry, permission policy, and
-sandbox engine.
-
-## Tool Execution Lifecycle
-
-A tool call is part of the model conversation, not a side channel that bypasses
-it. The model proposes an action, Zero decides whether that action is allowed to
-run, the tool returns a structured result, and that result becomes context for the
-next provider request.
-
-```mermaid
-sequenceDiagram
-    participant Model as LLM / provider
-    participant Agent as agent.Run
-    participant Policy as filters + permissions + sandbox
-    participant Hooks as hooks
-    participant Tool as tool implementation
-    participant Surface as TUI / exec callbacks
-    participant Transcript as model conversation + session events
-
-    Agent->>Model: request with messages + available tool schemas
-    Model-->>Agent: assistant text and/or tool_call(name, args)
-    Agent-->>Surface: stream text / tool-call preview
-    Agent->>Agent: collect and decode tool arguments
-    Agent->>Policy: check tool filters, permission mode, sandbox policy
-
-    alt denied before execution
-        Policy-->>Agent: denied result
-        Agent-->>Surface: tool failed/denied event
-        Agent->>Transcript: append tool result explaining denial
-    else needs user approval
-        Policy-->>Surface: permission request
-        Surface-->>Policy: allow / deny / remember decision
-        Policy-->>Agent: decision
-    end
-
-    alt approved to execute
-        Agent->>Hooks: beforeTool hooks
-        Hooks-->>Agent: allow / block / annotate
-        Agent->>Tool: execute(ctx, args)
-        Tool-->>Agent: output, error, metadata
-        Agent->>Hooks: afterTool hooks
-        Agent->>Agent: redact, truncate, classify success/failure
-        Agent-->>Surface: OnToolResult callback
-        Agent->>Transcript: append tool result message
-    end
-
-    Agent->>Model: next request includes prior assistant tool_call + tool result
-    Model-->>Agent: continue with more tools or final answer
-```
-
-Conceptually, one requested tool goes through these phases:
-
-1. **Proposal** — the provider streams assistant content and may emit one or more
-   tool calls. Zero can show the partial tool arguments in the TUI while they are
-   still streaming, but nothing has executed yet.
-2. **Decode and gate** — the agent validates JSON arguments, checks whether the
-   tool is enabled, applies permission mode, asks the sandbox engine about path,
-   network, shell, or command-prefix constraints, and may invoke permission
-   callbacks.
-3. **Execute** — the selected registry entry runs locally, through MCP, as a
-   specialist/sub-agent, or through a plugin/local-control helper. Hooks can run
-   before or after this step.
-4. **Normalize result** — Zero turns the outcome into an agent `ToolResult`: a
-   success, failure, denial, cancellation, timeout, or validation error with
-   output/metadata that may be redacted or truncated.
-5. **Append to context** — the result is appended as a tool-result message paired
-   with the assistant tool call. The next model request includes that result, so
-   the model can inspect it and decide what to do next.
-6. **Render and persist** — the result is surfaced to the TUI, stream-JSON
-   writer, or ACP client. TUI and exec record rich tool/session events for
-   resume/history; ACP currently persists conversational messages and streams
-   tool/permission updates to the client.
-
-Tool success and failure are both informative context. A successful `read_file`
-result gives the model file contents; a failed `edit_file` result gives the model
-an error such as “old string not found”; a denied shell command tells the model
-that the action is not currently permitted. The model is expected to use that
-feedback to continue, choose another tool, ask the user, or explain the blocker.
-
-The judgement that “information is sufficient” is made by the model, not by the
-TUI. Zero controls the loop mechanics: it keeps appending messages, enforces
-safety, retries or nudges around malformed/empty/incomplete turns, and stops on
-hard limits. But the semantic decision to call another tool versus produce a
-final answer comes from the next provider response. In other words:
-
-```mermaid
-flowchart TD
-    Result[Tool result appended to conversation] --> Next[Next provider request]
-    Next --> Decide{Model response}
-    Decide -- More information/action needed --> MoreTools[Request another tool]
-    MoreTools --> Result
-    Decide -- User input needed --> AskUser[Call ask_user or explain question]
-    AskUser --> Result
-    Decide -- Enough information --> Final[Final assistant answer]
-    Decide -- Invalid / empty / unsafe --> Guard[Agent guardrail, retry, denial, or stop reason]
-    Guard --> Next
-```
-
-This is why tool output is usually more than UI decoration. It is part of the
-working memory for the next model turn, subject to context-window pressure and
-compaction. The TUI may summarize or prettify it for the user, but the agent loop
-also stores the machine-facing result that the model sees.
-
-## Context Budget and Compaction
-
-The `Maybe compact context` step in the agent-loop diagram is Zero's context-window
-pressure valve. It is not a separate memory engine; it rewrites the current
-model-facing conversation so the next provider request fits within the configured
-context window.
-
-Compaction is enabled only when the run knows a positive `ContextWindow`. When it
-is enabled, the loop does this near the top of each turn:
-
-```mermaid
-flowchart TD
-    Tools[Partition visible tools] --> Estimate[Estimate messages + tool-definition tokens]
-    Estimate --> Under{Under threshold?}
-    Under -- Yes --> Keep[Keep context as-is]
-    Under -- No --> Grown{Grown past low-water mark?}
-    Grown -- No --> Keep
-    Grown -- Yes --> Prune[Prune stale large tool outputs]
-    Prune --> Recheck{Back under threshold?}
-    Recheck -- Yes --> KeepPruned[Use pruned context]
-    Recheck -- No --> Summarize[Summarize oldest middle]
-    Summarize --> Replace[system messages + summary + recent suffix]
-```
-
-The trigger is intentionally approximate:
-
-- Zero estimates request size using message text, images, tool-call arguments, and
-  the tool definitions advertised on that turn.
-- The proactive threshold is roughly **two-thirds of the context window**.
-- Provider usage from completed turns is used to calibrate the rough estimate over
-  time, so the budget gauge and compaction decision track the provider more
-  closely after a few turns.
-- A low-water mark prevents the loop from re-summarizing the same already-compacted
-  history on every turn.
-
-When compaction actually runs, it tries to preserve the parts most likely to be
-needed by the model:
-
-- leading system messages stay verbatim;
-- the most recent turns stay verbatim (`CompactionPreserveLast`, defaulting to the
-  agent's normal preserve count);
-- the preserved suffix is widened to a provider-valid turn boundary so tool-result
-  messages are not replayed without their corresponding assistant tool call;
-- the older middle is summarized into one user-role message labeled
-  `[Summary of earlier conversation]`;
-- structured state that should not be paraphrased away, such as active plan/loaded
-  skill state, is preserved alongside the prose summary.
-
-There are two compaction paths:
-
-1. **Proactive compaction** runs before the provider call when the estimated
-   request is over threshold.
-2. **Reactive compaction** runs after a provider returns a context-limit style
-   error. Zero compacts once and retries that same turn; if the provider still
-   rejects the request, the error is surfaced rather than looping forever.
-
-Compaction failures are conservative. If the summarizer fails during proactive
-compaction, Zero keeps the original messages and tries to continue. It does not
-drop conversation history just to make room.
-
-## Presentation Layer: Bubble Tea TUI
-
-`internal/tui` is a Bubble Tea application. The usual Bubble Tea shape applies:
-
-```mermaid
-flowchart LR
-    Init[model.Init] --> Update[model.Update]
-    Update --> View[model.View]
-    View --> Terminal[Terminal frame]
-    Terminal --> Msg[tea.Msg: key, mouse, window, agent event]
-    Msg --> Update
-```
-
-The TUI is intentionally not the agent brain. It is responsible for:
-
-- collecting user input from the composer;
-- rendering assistant text, reasoning, tool cards, permissions, plans, usage,
-  files, and sidebars;
-- handling slash commands and modal surfaces;
-- converting agent callbacks into Bubble Tea messages;
-- persisting visible session events;
-- allowing live provider/model/permission changes.
-
-When a prompt is submitted, the TUI starts a background command that invokes
-`agent.Run`. Streaming callbacks send messages back into the Bubble Tea update
-loop, so rendering remains responsive while the agent is working.
-
-## How the TUI and Agent Loop Connect
-
-A common source of confusion is that Zero has two event loops running at once:
-
-1. **Bubble Tea loop** — owns terminal input, in-memory UI state, and frames.
-2. **Agent loop** — owns model turns, tool execution, permissions, retries, and
-   stop conditions.
-
-They are connected by `tea.Cmd` and `agent.Options` callbacks. The TUI never
-runs provider/tool logic directly inside `Update`; instead, submit handling gates
-and dispatches a prompt, the TUI starts a background command, and that command
-calls `agent.Run` with callbacks that send runtime events back to the Bubble Tea
-program. Function names in this section are illustrative code anchors, not API
-boundaries.
-
-```mermaid
-sequenceDiagram
-    participant User
-    participant TUI as Bubble Tea model<br/>internal/tui
-    participant Cmd as tea.Cmd goroutine<br/>runAgentWithOptions
-    participant Agent as agent.Run<br/>internal/agent
-    participant Provider as Provider.StreamCompletion
-    participant Tools as Tool registry + sandbox
-    participant Store as Session store
-
-    User->>TUI: type prompt / press Enter
-    TUI->>TUI: validate/gate submit and append visible user row
-    TUI->>Cmd: launch tea.Cmd with runID + cancellable context
-    Cmd->>Agent: agent.Run(prompt, provider, options)
-    Agent->>Provider: stream completion request
-    Provider-->>Agent: text / reasoning / tool-call deltas
-    Agent-->>Cmd: OnText / OnReasoning / OnToolCallStart callbacks
-    Cmd-->>TUI: runtimeMessageSink(tea.Msg)
-    TUI->>TUI: Update appends/updates transcript rows
-    TUI-->>User: View renders new frame
-
-    Agent->>Tools: execute collected tool call
-    Tools->>Tools: permission, sandbox, hooks, redaction
-    Tools-->>Agent: tool result
-    Agent-->>Cmd: OnToolCall / OnToolResult / OnPermission / OnUsage
-    Cmd->>Store: batch session events for resume/history
-    Cmd-->>TUI: final agentResponseMsg
-    TUI->>TUI: mark run complete, flush rows/events, unlock composer
-```
-
-In the TUI path, the same event can have three representations:
-
-| Runtime thing | Agent representation | TUI representation | Session representation |
-| --- | --- | --- | --- |
-| Assistant text | streamed provider text forwarded through `OnText` | incremental assistant row text | assistant message event |
-| Reasoning preview | `OnReasoning` callback | collapsible/preview reasoning row | usually not the same as final assistant text |
-| Tool call | `agent.ToolCall` from collected provider stream | tool card, specialist card, or plan panel update | `EventToolCall` |
-| Tool result | `agent.ToolResult` | result card/status, changed files, plan/specialist updates | `EventToolResult` |
-| Permission prompt/audit | `PermissionRequest` / `PermissionEvent` | modal prompt or permission row | permission request/decision events |
-| Usage/context | provider usage + measured context | context gauge / usage metadata | usage event |
-
-### One interactive turn, end to end
-
-For a normal prompt in the interactive TUI:
-
-1. Submit handling validates composer state and command gates, then dispatches
-   the prompt to the launch path.
-2. The launch path appends immediate UI rows, creates a cancellable run context,
-   and starts `runAgent` as a `tea.Cmd`; run bookkeeping assigns the `runID` used
-   to correlate later events.
-3. `runAgentWithOptions` copies the current UI-controlled runtime choices into
-   `agent.Options`: provider/model labels, permission mode, cwd, session ID,
-   images, registry, response style, context window, and self-correction hooks.
-4. The command wraps agent callbacks. For example:
-   - `OnText` sends incremental assistant text into the TUI message sink.
-   - `OnToolCallStart` / `OnToolCallDelta` show live tool-argument streaming.
-   - `OnToolCall` creates visible tool rows and appends session events.
-   - `OnToolResult` creates result rows, updates plan/specialist side panels, and
-     records changed files.
-   - `OnPermissionRequest` sends a prompt to the TUI and blocks the agent loop
-     until the user answers or the run context is cancelled.
-   - `OnAskUser` does the same for model-initiated clarification questions.
-5. `agent.Run` builds provider-neutral messages and tool definitions, streams a
-   provider response, executes any requested tools through the registry/sandbox,
-   appends tool results back into the model conversation, and repeats until it has
-   a final answer or a stop condition.
-6. When `agent.Run` returns, the command emits `agentResponseMsg`. The Bubble Tea
-   loop marks the run complete, flushes queued rows/session events, updates usage
-   and file state, and re-enables the composer.
-
-The important architectural point: **the TUI is stateful and reactive, but the
-agent loop is the source of runtime truth for model/tool progress**. Presentation
-state is derived from callback events and persisted session events; workspace
-changes happen only through tools executed by the agent runtime.
-
-### Feature flows use the same bridge
-
-For one interactive turn, the supporting systems do not run beside the TUI/agent
-loop; they run inside or around it:
-
-```mermaid
-flowchart TD
-    Submit[TUI submit] --> Options[Build agent.Options]
-    Options --> Context[Assemble context<br/>config, guidelines, history, skills]
-    Context --> ProviderCall[Provider stream]
-    ProviderCall --> StreamEvents[Text/reasoning/tool-call events]
-    StreamEvents --> Render[TUI render callbacks]
-
-    ProviderCall --> ToolCall{Tool call?}
-    ToolCall -- No --> Final[Final response]
-    ToolCall -- Yes --> ToolGate[Tool gate<br/>filters, permissions, sandbox]
-    ToolGate --> HooksBefore[beforeTool hooks]
-    HooksBefore --> Execute[Tool / MCP / specialist / plugin]
-    Execute --> HooksAfter[afterTool hooks]
-    HooksAfter --> ToolResult[Tool result message]
-    ToolResult --> Persist[Session events]
-    ToolResult --> Context
-
-    Final --> Persist
-    Persist --> Resume[Resume / fork / rewind / parent metadata]
-```
-
-Most interactive features are variations of that same bridge rather than separate
-architectures:
-
-- **Slash commands** either mutate TUI state directly (`/model`, `/permission`,
-  `/help`) or start short `tea.Cmd` jobs that later return a message to `Update`
-  (`/doctor`, `/search`, `/compact`, session commands).
-- **Permissions** sit between a model-requested tool call and actual execution:
-  the agent blocks on `OnPermissionRequest`, the TUI surfaces a modal and sends
-  the decision back through a channel, and the session log records the
-  request/decision events.
-- **Plans, specialists/sub-agents, and ask-user prompts** all follow the same
-  runtime pattern: the model requests a capability, the agent records the
-  call/result in its conversation, and the TUI mirrors the visible state as plan
-  panels, specialist cards, or question forms. Specialist session metadata can
-  carry parent/root and sub-agent/spec fields, with richer linkage depending on
-  the specialist or swarm path.
-- **Streaming tool writes** start as provider tool-call deltas before execution;
-  the TUI shows the arguments live, then replaces/augments them with the actual
-  `ToolResult` after the tool runs.
-- **Compaction** is an agent-loop context-management step; the TUI may show that
-  it happened, but the purpose is to keep the next provider request within budget.
-- **Stream-JSON** is another surface over the same callbacks: `zero exec`
-  serializes the events as JSON protocol events instead of Bubble Tea views.
-
-So a feature-specific deep dive should usually explain: what user event starts it,
-which `tea.Msg`/callback carries it, which agent/tool event is authoritative, and
-which session events make it resumable.
-
-## Runtime Provider Layer
-
-Providers are hidden behind this interface in `internal/zeroruntime`:
-
-```go
-type Provider interface {
-    StreamCompletion(ctx context.Context, request CompletionRequest) (<-chan StreamEvent, error)
-}
-```
-
-The agent loop only knows about provider-neutral messages, tool definitions,
-stream events, reasoning blocks, images, and token usage. Provider adapters in
-`internal/providers/*` translate those neutral types to provider-specific APIs:
-
-- OpenAI and OpenAI-compatible APIs;
-- Anthropic and Anthropic-compatible APIs;
-- Gemini;
-- Codex / ChatGPT OAuth-backed runtime.
-
-`internal/providers.New` resolves the configured model through the model
-registry/catalog, normalizes API model IDs and base URLs, attaches API keys or
-OAuth resolvers, and returns the concrete adapter.
-
-## Tools and Capabilities
-
-Tools implement a common interface and are registered by name in
-`internal/tools.Registry`.
-
-```mermaid
-flowchart TD
-    AgentToolCall[Tool call from model] --> Decode[Decode JSON args]
-    Decode --> Filter[Enabled/disabled + permission-mode filter]
-    Filter --> Precheck[Pre-permission rejecters]
-    Precheck --> SandboxEval[Sandbox Evaluate]
-    SandboxEval --> Prompt{Needs user approval?}
-    Prompt -- Yes --> Decision[Permission callback]
-    Prompt -- No --> Run[Registry.RunWithOptions]
-    Decision --> Run
-    Run --> Redact[Redact secrets + enforce output ceiling]
-    Redact --> Message[Return tool result to model]
-```
-
-Core tool groups include:
-
-- **Read-only tools**: file reads, directory listing, glob, grep, LSP navigation,
-  skills, `ask_user`, permission requests.
-- **Write tools**: `write_file`, `edit_file`, `apply_patch`, `update_plan`.
-- **Shell tools**: `exec_command`, `write_stdin`, legacy `bash`.
-- **Network tools**: `web_fetch` and optional web search backend.
-- **Local-control tools**: config-gated wrappers for browser, terminal, desktop,
-  and artifact helpers backed by packages such as `internal/localcontrol` and
-  `internal/browser`.
-- **Extension tools**: MCP tools, specialists/sub-agents, swarm tools, plugins.
-
-Tool definitions include safety metadata: side-effect class, permission level,
-reason, and whether prompt-gated tools are advertised in auto-like modes. The
-agent uses this metadata plus per-turn filters and deferred-loading state to
-choose which registered tools the model can see and whether a call can run
-immediately.
-
-## Permission and Sandbox Model
-
-Zero has two related but separate checks:
-
-1. **Tool permission metadata** says whether a tool is allowed, denied, or needs
-   approval.
-2. **Sandbox policy** evaluates the actual request: path scope, network access,
-   destructive shell patterns, explicit escalation, persistent/session grants,
-   and platform isolation availability.
-
-```mermaid
-flowchart TD
-    Call[Tool call] --> Safety[Tool Safety metadata]
-    Call --> Scope[Workspace/additional roots scope]
-    Safety --> Engine[Sandbox Engine Evaluate]
-    Scope --> Engine
-    Engine --> Allow[Allow]
-    Engine --> Prompt[Prompt user]
-    Engine --> Deny[Deny]
-    Prompt --> Grant[Turn/session/persistent grant]
-    Grant --> Execute[Execute tool]
-    Allow --> Execute
-    Deny --> Result[Return denied tool result]
-```
-
-The sandbox engine can:
-
-- auto-allow workspace writes for safe file-edit tools;
-- block reads/writes outside the allowed scope;
-- prompt for recoverable outside-workspace access;
-- deny or prompt for network depending on policy;
-- detect destructive shell commands;
-- run shell commands through a platform backend when available;
-- remember session or persistent grants;
-- support narrow command-prefix approvals.
-
-The user-facing TUI and `zero exec` mode both route permission requests through
-callbacks, but the TUI can show interactive modals while headless mode generally
-returns non-interactive fallback behavior unless a protocol/client handles it.
-
-## Session Storage and History
-
-Sessions live locally under Zero's data directory. A session directory contains:
-
-- `metadata.json` — session ID, title, cwd, provider/model, parent/root IDs,
-  spec/sub-agent fields, timestamps, and event count.
-- `events.jsonl` — append-only event log with messages, tool calls/results,
-  permissions, usage, checkpoints, forks, compaction, and specs. TUI and exec use
-  the richer event stream; ACP currently records conversational message history
-  while streaming tool and permission updates to the client.
-
-```mermaid
-flowchart LR
-    TUI[TUI callbacks] --> Store[internal/sessions.Store]
-    Exec[exec recorder] --> Store
-    Store --> Metadata[metadata.json]
-    Store --> Events[events.jsonl]
-    Events --> Resume[resume]
-    Events --> Fork[fork]
-    Events --> Rewind[rewind/checkpoints]
-    Events --> Children[parent/root metadata]
-```
-
-This is why sessions can be resumed, forked, rewound, and inspected without a
-remote service. Zero stores the local event history; providers only receive the
-messages needed for the current completion request.
-
-## Stream-JSON Protocol
-
-`zero exec --output-format stream-json` emits newline-delimited JSON events for
-integrations and CI. The protocol is implemented in `internal/streamjson` and is
-specified in [STREAM_JSON_PROTOCOL.md](STREAM_JSON_PROTOCOL.md).
-
-Common output event types include:
-
-- `run_start`, `text`, `reasoning`, `tool_call`, `tool_result`;
-- `permission`, `permission_request`, `permission_decision`;
-- `checkpoint`, `restore`, `usage`, `final`, `warning`, `error`, `run_end`.
-
-Input stream-JSON can provide `prompt` or `message` events and image payloads.
-The CLI parses those events, resolves a single prompt, validates/decodes images,
-and passes the result into the same agent loop used everywhere else.
-
-## Extensibility Flow
-
-Zero's extension surfaces are loaded before the agent loop starts:
-
-```mermaid
-flowchart TD
-    Config[Config resolution] --> Registry[Tool registry]
-    Registry --> Core[Core tools]
-    Registry --> MCP[MCP server tools]
-    Registry --> Specialist[Specialist / Task tools]
-    Registry --> Plugin[Plugin tools]
-
-    Config --> Prompt[System prompt context]
-    Prompt --> Guidelines[AGENTS.md / ZERO.md]
-    Prompt --> Skills[Skills list]
-    Prompt --> SpecialistContext[Specialist descriptions]
-
-    ToolLifecycle[Tool lifecycle] --> Hooks[beforeTool / afterTool hooks]
-```
-
-- **Project/user guidelines** become prompt context.
-- **Skills** are instruction packs the model can load on demand.
-- **Specialists** are sub-agents callable through the `Task` tool.
-- **MCP servers** contribute external tools.
-- **Plugins** can add tools, hooks, and skill roots. Bootstrap merges plugin skill
-  roots into the runtime skill tool and prompt-visible skill list; the base
-  `internal/skills` scanner still scans one root at a time.
-- **Hooks** can observe or block tool lifecycle events.
-
-## End-to-End Data Flow
-
-A typical interactive edit request looks like this:
-
-```mermaid
-sequenceDiagram
-    participant U as User
-    participant T as TUI
-    participant A as Agent loop
-    participant P as Provider
-    participant R as Tool registry
-    participant S as Sandbox
-    participant FS as Workspace/session files
-
-    U->>T: Type prompt and press Enter
-    T->>A: agent.Run(prompt, provider, options)
-    A->>P: CompletionRequest(messages, tool definitions)
-    P-->>A: Stream text + tool call edit_file
-    A->>S: Evaluate edit_file args
-    S-->>A: allow/prompt/deny decision
-    A->>T: permission callback if prompt needed
-    T-->>A: user decision
-    A->>R: Run edit_file
-    R->>FS: Write workspace file
-    R-->>A: Tool result + changed files
-    A->>P: Next CompletionRequest with tool result
-    P-->>A: Final answer
-    A-->>T: Result + callbacks
-    T->>FS: Append session events
-    T-->>U: Render final transcript
-```
-
-The same request in `zero exec` follows the same provider/tool/sandbox path, but
-streaming events are written to stdout/stderr instead of rendered in Bubble Tea.
-
-## Mental Model
-
-When you work on Zero, start by locating the change on the four-layer spine from
-[Architecture Spine](#architecture-spine) — Surface, Agent loop, Capabilities,
-Persistence — then jump to the package that owns that boundary:
-
-- UI behavior? Start in `internal/tui`.
-- Command-line behavior? Start in `internal/cli`.
-- Model/tool-turn behavior? Start in `internal/agent`.
-- A new capability? Register a tool in `internal/tools` or an extension surface.
-- Safety policy? Start in `internal/sandbox` and agent permission wiring.
-- Resume/history behavior? Start in `internal/sessions` and session recorders.

docs/INSTALL.md

@@ -1,55 +1,27 @@
-# Installing Zero
+# Zero Install Scripts
 
-Zero is distributed as:
+Zero release archives are published as:
 
-- an npm package, `@gitlawb/zero`
-- release archives on GitHub Releases
-- source builds with Go 1.26.5+
+- `zero-v<version>-linux-<arch>.tar.gz`
+- `zero-v<version>-macos-<arch>.tar.gz`
+- `zero-v<version>-windows-<arch>.zip`
 
-The install scripts download a platform-specific release archive and require a
-published GitHub Release for the requested version. The npm package is
-self-contained: the platform binary installs from the npm registry.
+Each archive must have a matching `.sha256` file. The install scripts download both files and verify the checksum before copying the binary.
 
-## npm
+Maintainers can verify the release directory before upload:
 
 ```bash
-npm install -g @gitlawb/zero
-zero
+go run ./cmd/zero-release package
+go run ./cmd/zero-release verify
 ```
 
-The package supports Linux and macOS on x64 and arm64, and Windows on x64
-(Windows on ARM runs the x64 build under emulation). It installs a small
-`zero` wrapper plus, as an optional dependency, a platform payload with the
-native binary and the bundled browser/terminal control helpers. There are no
-install scripts and nothing is downloaded from outside the npm registry, so
-the install is silent and works identically under npm, Bun, pnpm, and yarn —
-no trust or approval steps. See [NPM_PACKAGING.md](NPM_PACKAGING.md) for the
-package architecture.
+`verify:release` requires every archive in `dist/release` to have a same-directory `.sha256` file whose contents are:
 
-Requirements:
-
-- Node.js 18+
-- network access to npm
-
-If the install skipped optional dependencies (`npm install --omit=optional`,
-or a package manager configured to do so), the wrapper falls back to
-downloading the binary from the matching GitHub Release, with checksum
-verification. The fetch is retried on each run until a binary is in place (a
-transient network failure heals itself; an install directory the current user
-cannot write to keeps failing — rerun the install or the command below with
-sufficient permissions). To trigger the fetch manually:
-
-```bash
-node "$(npm root -g)/@gitlawb/zero/scripts/postinstall.mjs"
+```text
+<sha256>  <archive-name>
 ```
 
-## Linux And macOS Script
-
-Install the latest release:
-
-```bash
-curl -fsSL https://raw.githubusercontent.com/Gitlawb/zero/main/scripts/install.sh | bash
-```
+## Linux And macOS
 
 From a checkout:
 
@@ -61,14 +33,18 @@ Install a specific version:
 
 ```bash
 ZERO_VERSION=0.1.0 scripts/install.sh
-scripts/install.sh --version 0.1.0
 ```
 
-Install somewhere else:
+Install to a custom directory:
 
 ```bash
 ZERO_INSTALL_DIR="$HOME/bin" scripts/install.sh
-scripts/install.sh --install-dir "$HOME/bin"
+```
+
+Install from a fork or internal repository:
+
+```bash
+scripts/install.sh --repo owner/repo
 ```
 
 Defaults:
@@ -77,15 +53,9 @@ Defaults:
 - Version: latest GitHub release
 - Install path: `~/.local/bin/zero`
 
-Requirements: Bash, `curl` or `wget`, `tar`, and `shasum` or `sha256sum`.
+Requirements: `curl` or `wget`, `tar`, and `shasum` or `sha256sum`.
 
-## Windows PowerShell Script
-
-Install the latest release:
-
-```powershell
-irm https://raw.githubusercontent.com/Gitlawb/zero/main/scripts/install.ps1 | iex
-```
+## Windows
 
 From a checkout:
 
@@ -99,145 +69,30 @@ Install a specific version:
 powershell -ExecutionPolicy Bypass -File scripts/install.ps1 -Version 0.1.0
 ```
 
-Install somewhere else:
+Install to a custom directory:
 
 ```powershell
 powershell -ExecutionPolicy Bypass -File scripts/install.ps1 -InstallDir "$env:USERPROFILE\bin"
 ```
 
+Install from a fork or internal repository:
+
+```powershell
+powershell -ExecutionPolicy Bypass -File scripts/install.ps1 -Repository owner/repo
+```
+
 Defaults:
 
 - Repository: `Gitlawb/zero`
 - Version: latest GitHub release
 - Install path: `%LOCALAPPDATA%\zero\bin\zero.exe`
 
-## From Source
-
-```bash
-git clone https://github.com/Gitlawb/zero.git
-cd zero
-go run ./cmd/zero
-```
-
-Build a local binary:
-
-```bash
-go build -o zero ./cmd/zero
-```
-
-Source builds require Go 1.26.5+.
-
-### Sandbox Helpers For Source Builds
-
-Release archives include the platform sandbox helpers. If you build directly
-from source, build the helpers you need:
-
-Linux:
-
-```bash
-go build -o zero ./cmd/zero
-go build -o zero-linux-sandbox ./cmd/zero-linux-sandbox
-go build -o zero-seccomp ./cmd/zero-seccomp
-```
-
-Put `zero` and `zero-linux-sandbox` in the same directory on `PATH`, for example
-`~/.local/bin`. `zero-seccomp` is kept as a compatibility wrapper; the sandbox
-helper applies the Unix-socket filter itself when that sandbox option is enabled.
-Linux native sandboxing also requires Bubblewrap to be installed.
-
-macOS uses the system sandbox and does not need an extra helper binary.
-
-### Termux (Android)
-
-Zero can run natively on Android via [Termux](https://termux.dev/). Build with
-`GOOS=android` to avoid the `faccessat2` syscall that is blocked by Samsung's
-seccomp filter on Android:
-
-```bash
-# Install Go in Termux
-pkg install golang
-
-# Build Zero for Android
-git clone https://github.com/Gitlawb/zero.git
-cd zero
-CGO_ENABLED=0 GOOS=android GOARCH=arm64 go build -ldflags="-s -w" -o zero ./cmd/zero
-
-# Move into PATH
-mv zero ~/.local/bin/
-```
-
-> **Why `GOOS=android`?** Go 1.26+ detects `runtime.GOOS == "android"` and skips
-> the `faccessat2` syscall inside `os/exec.findExecutable`, falling back to
-> permission-bit checks. Without this flag, Android's seccomp sends SIGSYS and
-> kills the process whenever Zero looks up a binary on `PATH` (git, sh, etc.).
-
-**DNS.** Android does not expose `/etc/resolv.conf`. Go's pure-Go DNS resolver
-needs one. Use `proot` to bind-mount Termux's resolver config:
-
-```bash
-pkg install proot
-proot -b "$PREFIX/etc/resolv.conf:/etc/resolv.conf" zero
-```
-
-Create a wrapper at `~/.local/bin/zero` to avoid typing proot every time:
-
-```bash
-#!/data/data/com.termux/files/usr/bin/bash
-exec proot -b "$PREFIX/etc/resolv.conf:/etc/resolv.conf" ~/.local/bin/zero.bin "$@"
-```
-
-**Scroll.** On native Termux (not under PRoot), mouse scrolling works out of the
-box. The TUI uses Bubble Tea's `AllMotion` mouse mode by default. If you run Zero
-inside PRoot (e.g. through proot-distro), the scroll fix activates `CellMotion`
-to avoid PRoot's ptrace interference with the 1003 escape sequence.
-
-**Providers.** Zero works with any OpenAI-compatible provider on Termux. For
-example, to use OpenCode Zen's free tier:
-
-```bash
-zero providers add opencode \
-  --name opencode \
-  --model deepseek-v4-flash-free \
-  --base-url https://opencode.ai/zen/v1 \
-  --set-active
-```
-
-Windows source builds can use the main `zero.exe` as the command runner and setup
-helper through Zero's built-in self-dispatch path. If you want a release-style
-layout anyway, build the standalone helper executables next to `zero.exe`:
-
-```powershell
-go build -o zero.exe ./cmd/zero
-go build -o zero-windows-command-runner.exe ./cmd/zero-windows-command-runner
-go build -o zero-windows-sandbox-setup.exe ./cmd/zero-windows-sandbox-setup
-```
-
-## Release Archive Format
-
-Release archives are named:
-
-- `zero-v<version>-linux-<arch>.tar.gz`
-- `zero-v<version>-macos-<arch>.tar.gz`
-- `zero-v<version>-windows-<arch>.zip`
-
-Supported targets:
-
-- `linux-x64`
-- `linux-arm64`
-- `macos-x64`
-- `macos-arm64`
-- `windows-x64` (Windows on ARM runs this build under emulation)
-
-Each archive must have a matching `.sha256` file. The install scripts download
-both files, verify the checksum, and then copy the binary into the install
-directory.
-
 ## Updating
 
-Check for a newer release:
+Check whether a newer release exists:
 
 ```bash
 zero update --check
 ```
 
-Then reinstall with npm or rerun the install script for the version you want.
+For M2, updates are check-only. Re-run the installer to replace the local binary after reviewing the release.

docs/M1_HEADLESS_EXEC_PRD.md

@@ -0,0 +1,87 @@
+# M1 Headless Exec PRD
+
+Status: Draft
+Owner: Vasanth
+Milestone: M1
+Scope: `zero exec` CLI surface
+
+## Goal
+
+Ship the first production-shaped headless command for Zero:
+
+```sh
+zero exec "inspect this repo and summarize the next fix"
+```
+
+This command should be useful for local scripts, CI smoke runs, and later VS Code integration without pulling in the full future protocol, session, MCP, sandbox, or plugin roadmap.
+
+## Non-Goals
+
+- No daemon mode.
+- No JSON-RPC input protocol.
+- No MCP, plugin, marketplace, or subagent support.
+- No session resume/fork/search.
+- No new provider adapters in this slice.
+- No architecture rewrite of the agent loop.
+
+## User Stories
+
+1. As a CLI user, I can run one prompt headlessly and receive the assistant answer on stdout.
+2. As a script author, I can choose `text` or JSONL output.
+3. As a reviewer, I can see tool activity without corrupting text stdout.
+4. As a model tester, I can override the configured model for one run.
+5. As a cautious user, prompt-gated tools stay disabled unless I pass an explicit unsafe flag.
+
+## CLI Contract
+
+```sh
+zero exec [prompt...] [options]
+```
+
+Options for M1:
+
+- `-f, --file <path>`: read the prompt from a file.
+- `-m, --model <model>`: override the configured model for this run.
+- `-C, --cwd <path>`: run from another working directory.
+- `-o, --output-format <text|json>`: choose stdout format.
+- `--skip-permissions-unsafe`: grant prompt-gated tools for this run.
+
+The existing `zero -p "prompt"` shortcut remains as a compatibility path and should call the same runner.
+
+## Output Contract
+
+Text mode:
+
+- stdout contains assistant text only.
+- stderr contains tool rows, warnings, and errors.
+
+JSON mode:
+
+- stdout is newline-delimited JSON events.
+- Required event types for this slice: `run_start`, `text`, `tool_call`, `tool_result`, `final`, `warning`, `error`, `done`.
+
+## Exit Codes
+
+- `0`: success
+- `1`: unexpected crash
+- `2`: usage error
+- `3`: provider/config/runtime error
+- `4`: reserved for future tool failure policy
+- `5`: reserved for future permission-denied policy
+
+## Permission Behavior
+
+Default headless mode advertises only tools marked `permission: "allow"`.
+
+`--skip-permissions-unsafe` advertises tools that are not marked `deny` and grants prompt-gated tools through `ToolRegistry.run`. It must not bypass schema validation or direct tool execution safeguards.
+
+## Acceptance Criteria
+
+- `go test ./...` passes.
+- `go run ./cmd/zero-release build` passes.
+- `go run ./cmd/zero-release smoke` passes.
+- `zero exec --help` documents the M1 flags.
+- Usage errors do not initialize providers.
+- Provider failures return exit code `3`.
+- Normal mode does not advertise `bash`, `write_file`, `edit_file`, or `apply_patch`.
+- Unsafe mode still uses the registry execution path.

docs/NPM_PACKAGING.md

@@ -1,162 +0,0 @@
-# npm packaging
-
-How `@gitlawb/zero` is put together on npm, why it is shaped this way, and the
-rules the release pipeline must follow. Read this before touching
-`package.json`, `bin/zero.js`, `scripts/postinstall.mjs`,
-`scripts/npm/build-platform-packages.mjs`, or the npm-publish steps of
-`release-artifacts.yml`.
-
-## Goals
-
-A `npm install -g @gitlawb/zero` must be **silent and self-contained**:
-
-- No `EBADENGINE` warnings — ours or from any package in the dependency tree.
-- No install scripts anywhere in the tree, so npm's `allow-scripts` gating,
-  Bun's blocked-by-default lifecycle scripts, and pnpm's strict mode all
-  install zero without prompts, trust ceremonies, or broken binaries.
-- No network fetches outside the npm registry at install time. GitHub being
-  down or rate-limited must not break `npm install`.
-- Browser control (`agent-browser`) and terminal control (`tuistory`) work out
-  of the box.
-
-## Architecture
-
-The model is the one used by Codex, esbuild, and Biome: a tiny wrapper package
-plus per-platform payloads carrying the native binaries.
-
-```
-@gitlawb/zero                    <- wrapper: bin/zero.js + optionalDependencies
-├─ @gitlawb/zero-darwin-arm64 -> npm:@gitlawb/zero@{version}-darwin-arm64
-├─ @gitlawb/zero-darwin-x64   -> npm:@gitlawb/zero@{version}-darwin-x64
-├─ @gitlawb/zero-linux-arm64  -> npm:@gitlawb/zero@{version}-linux-arm64
-├─ @gitlawb/zero-linux-x64    -> npm:@gitlawb/zero@{version}-linux-x64
-└─ @gitlawb/zero-win32-x64    -> npm:@gitlawb/zero@{version}-win32-x64
-```
-
-- The platform "packages" are **versions of the same `@gitlawb/zero` package**,
-  published at suffixed versions (`0.4.0-linux-x64`) and referenced through
-  `npm:` aliases in `optionalDependencies`. One package name means one npm
-  trusted-publisher configuration — no new publish credentials per platform.
-  The alias suffixes use Node's `process.platform`/`process.arch` names so
-  `bin/zero.js` can derive its platform package directly.
-- Each platform version sets `os` and `cpu`, so npm installs exactly one of
-  them and skips the rest.
-- A platform payload is assembled from the platform's release archive by
-  `scripts/npm/build-platform-packages.mjs` and contains the `zero` binary,
-  the platform's sandbox helpers, and the vendored `helpers/` tree (see
-  below). It has no `bin`, no scripts, and no dependencies — it is inert on
-  its own; the wrapper execs the binary out of it.
-- The published wrapper has **no scripts and no dependencies** — the
-  assembly script strips both from the repo `package.json` and injects the
-  exact-version `optionalDependencies`. (The repo `package.json` keeps
-  `agent-browser`/`tuistory` in `dependencies` purely as the version pins for
-  the vendored helpers tree; they are never installed by consumers.)
-- There is **no windows-arm64 build** (matches the release matrix); Windows on
-  ARM runs the x64 build under emulation via the first-run fallback.
-
-### Binary resolution in `bin/zero.js`
-
-1. Resolve `@gitlawb/zero-<platform>-<arch>` and exec the `zero` binary from
-   it. This wins over any previously downloaded copy — the platform version is
-   pinned to the wrapper release.
-2. Fall back to a binary previously downloaded next to the wrapper.
-3. If neither exists (`--omit=optional`, package managers that skip optional
-   dependencies), run the **fallback downloader**: `scripts/postinstall.mjs`,
-   the exact logic that used to run as a postinstall hook (HTTPS-only, SHA-256
-   verified against the release's own checksum file, no zip-slip), invoked by
-   the wrapper itself. Failures are deliberately not cached — the fetch
-   retries on every run until a binary is in place, so a transient network
-   error self-heals. Platforms with no release asset (anything outside the
-   matrix above, including windows-arm64) skip the attempt entirely rather
-   than probing the network each run.
-4. If the download is impossible too, print build-from-source guidance.
-
-There is deliberately **no `scripts.postinstall`** in any published
-`package.json`. The downloader exists only as a first-run fallback.
-
-### Vendored helpers (`helpers/`)
-
-`agent-browser` (Apache-2.0, vercel-labs/agent-browser) and `tuistory` are
-**vendored binaries/packages inside the platform payload**, not npm
-dependencies of the wrapper:
-
-- As a dependency, agent-browser's `engines: { node: ">=24", pnpm: ">=11" }`
-  and postinstall script produce `EBADENGINE` and `allow-scripts` warnings for
-  every installer. As a vendored tree it produces none, because npm never
-  resolves it.
-- `zero-release package` already stages the helpers tree into every release
-  archive (`stageLocalControlHelpers` runs `npm ci` from the repo's
-  `package.json` pins + lockfile on the native builder). The assembly script
-  reuses that staged tree, so npm installs and `install.sh` installs get
-  identical helpers.
-- The Go binary resolves helpers from `<binary dir>/helpers/node_modules/.bin`
-  on its own (`internal/localcontrol/browser.go`, `adjacentHelper`) — no
-  wrapper involvement, no configuration.
-- **Symlink materialization:** `npm pack` silently drops symlinks, and npm's
-  own `.bin` shims are symlinks on POSIX. The assembly script rewrites every
-  `.bin` symlink into a relocatable `#!/bin/sh` exec shim and dereferences any
-  other symlink, so the vendored tree survives publishing. (Verified
-  empirically; nested `node_modules` under `helpers/` IS packed — only the
-  package-root `node_modules` is always ignored.)
-- **Binary pruning:** agent-browser ships one ~11 MB native binary per
-  platform (7 total). The assembly script keeps only the payload's own
-  platform binary (plus the musl variant on linux, which its launcher detects
-  at runtime) — about 65 MB saved per platform package.
-
-## Publishing rules (release pipeline)
-
-These are the invariants `release-artifacts.yml` must hold. Breaking the first
-one is user-visible immediately.
-
-1. **Platform versions must never become `latest`.** `0.4.0-linux-x64` is a
-   semver *prerelease* of `0.4.0`; publishing it without an explicit
-   non-`latest` dist-tag would clobber `latest` and users would install a
-   platform payload as the CLI. The workflow publishes platform versions with
-   `--tag platform` and asserts `latest` survived before the wrapper publish,
-   then asserts `latest` equals the wrapper version after it. Both lookups
-   tolerate a failed `npm view` (registry lag, or no `latest` tag yet on a
-   bootstrap publish) — a missing value must not abort the job between the
-   platform and wrapper publishes, which would strand a half-published
-   release.
-2. **Platform versions publish before the wrapper.** The wrapper's
-   `optionalDependencies` pin exact suffixed versions; publishing the wrapper
-   first would create a window where installs resolve aliases that 404.
-3. **Exact-version pinning.** The wrapper at `X.Y.Z` references platform
-   versions `X.Y.Z-<platform>-<arch>` exactly — never ranges — so a wrapper
-   and its binaries can never skew.
-4. All publishes go through the existing npm OIDC trusted-publishing flow and
-   the `npm-publish` environment gate; no tokens. Because the platform
-   payloads are versions of the same package, they reuse the single
-   trusted-publisher configuration.
-
-`zero update` keeps working unchanged: it detects an npm install by finding a
-`package.json` named `@gitlawb/zero` next to the running binary — true inside
-a platform payload too — and updates via `npm install -g @gitlawb/zero@latest`.
-
-## Runbook: bumping the vendored helpers
-
-The vendored helper versions are pinned by the repo `package.json`
-`dependencies` + `package-lock.json`; the release build vendors exactly what
-the lockfile resolves. No `^range` surprises ship: a release carries what its
-lockfile pinned at build time.
-
-- Routine bump: `npm install agent-browser@<version>` (or `tuistory@…`) in the
-  repo to update pin + lockfile, verify the browser/terminal tools end-to-end,
-  ship with the next release.
-- Security response: if a vendored helper publishes a security fix, the bump
-  is release-worthy on its own — cut a patch release of zero carrying only the
-  pin change.
-- Attribution: agent-browser is Apache-2.0; its LICENSE ships inside the
-  vendored package directory (`helpers/node_modules/agent-browser/`), which
-  satisfies redistribution attribution.
-
-## History / rationale
-
-Until v0.3.x the npm package was a wrapper with a `postinstall` downloader and
-`agent-browser`/`tuistory` as regular dependencies. That produced three
-warnings on every install (`EBADENGINE` from agent-browser's `node >=24`
-engines pin, and `allow-scripts` warnings for both postinstall scripts), broke
-silently under Bun's default script blocking, and coupled `npm install` to
-GitHub Releases availability. The platform-package model removes the warnings
-structurally — there is nothing left in the tree for a package manager to warn
-about — rather than asking users to approve or suppress them.

docs/NPM_WRAPPER_SMOKE.md

@@ -1,8 +1,8 @@
 # npm Wrapper Smoke Checklist
 
 Run this checklist when a PR changes npm distribution files such as
-`package.json`, `bin/zero.js`, `scripts/npm/build-platform-packages.mjs`,
-Go build or release tooling, or the npm `bin` wrapper.
+`package.json`, `bin/zero.js`, Go build or release tooling, or the npm `bin`
+wrapper.
 
 ## Required Checks
 
@@ -21,52 +21,11 @@ go run ./cmd/zero-release build
 go run ./cmd/zero-release smoke
 ```
 
-## End-To-End Assembly Smoke
-
-Build a real archive for the host platform and assemble the npm payloads from
-it (this is exactly what the publish job does):
-
-```bash
-go run ./cmd/zero-release package
-node scripts/npm/build-platform-packages.mjs \
-  --artifacts-dir dist/release --out-dir dist/npm --only <platform>-<arch>
-```
-
-Then simulate an installed layout and run the wrapper through each resolution
-path:
-
-```bash
-# platform-package path
-mkdir -p /tmp/zero-sim/node_modules/@gitlawb
-cp -R dist/npm/wrapper /tmp/zero-sim/node_modules/@gitlawb/zero
-cp -R dist/npm/platforms/zero-<platform>-<arch> \
-  "/tmp/zero-sim/node_modules/@gitlawb/zero-<platform>-<arch>"
-node /tmp/zero-sim/node_modules/@gitlawb/zero/bin/zero.js --version
-
-# first-run download fallback (delete the platform package first)
-rm -rf "/tmp/zero-sim/node_modules/@gitlawb/zero-<platform>-<arch>"
-node /tmp/zero-sim/node_modules/@gitlawb/zero/bin/zero.js --version
-```
-
 ## Checklist
 
-- `package.json` has the expected package name (`@gitlawb/zero`), version,
-  `bin.zero` entry, and NO `scripts` entries — the published package must be
-  free of lifecycle scripts (see NPM_PACKAGING.md).
-- `scripts/npm/build-platform-packages.mjs` emits platform payloads whose
-  `package.json` is `@gitlawb/zero@<version>-<platform>-<arch>` with matching
-  `os`/`cpu`, contains the executable binary and the vendored `helpers/` tree
-  with regular-file (non-symlink) `.bin` shims, and a wrapper payload with no
-  scripts, no dependencies, and the full five-alias `optionalDependencies`
-  matrix.
-- `scripts/postinstall.mjs` (the first-run fallback) resolves the correct
-  release asset name/URL per platform (`ZERO_INSTALL_DRY_RUN=1` prints the
-  plan), verifies the downloaded archive's SHA-256, and extracts only the
-  known binary basenames into place. `ZERO_SKIP_DOWNLOAD=1` opts out cleanly
-  (exit 0) and an unsupported platform/arch is a non-fatal skip.
-- The wrapper prefers the platform package binary over a previously downloaded
-  one, and `agent-browser`/`tuistory` run from the platform package's
-  `helpers/node_modules/.bin` shims.
+- `package.json` has the expected package name, version, and `bin.zero` entry.
+- The wrapper binary resolves through the package `bin` entry and
+  `node_modules/.bin` in a package-install smoke test.
 - The built binary exits 0 for `zero --version` or `zero --help`.
 - `zero --version` reports `zero <package.json version>`.
 - Release packaging still emits the expected archive and checksum names when

docs/PERFORMANCE.md

@@ -1,6 +1,6 @@
 # Zero Performance Benchmarks
 
-The performance harness tracks three release-facing signals:
+The M2 performance harness tracks three release-facing signals:
 
 - Cold start: process startup time for `zero --version`.
 - Binary first output: time from spawning the built `zero --version` command to

docs/PRD.md

@@ -0,0 +1,602 @@
+# Zero — Product Requirements Document (v1.0)
+
+**Status:** DRAFT v1.0
+**Target:** Compete with Claude Code + Cursor + OpenCode
+**Timeline:** 6-12 months
+**Stack:** Go native core + npm distribution wrapper
+
+---
+
+## 1. Vision
+
+**Zero is a terminal-first, editor-aware, multi-provider AI coding platform.**
+
+It combines the best of:
+- **Claude Code** — terminal-native agentic coding
+- **Cursor** — IDE integration + smart editing
+- **OpenCode** — multi-provider + headless
+
+**One Go-native agent core. Three surfaces. Zero lock-in.**
+
+---
+
+## 2. Core Surfaces
+
+### 2.1 Terminal (TUI) — Primary
+- Interactive coding agent in terminal
+- Streaming, append-style (like Claude Code)
+- Bubble Tea + Lip Gloss-based
+- Full keyboard control
+- Mouse support
+
+### 2.2 Editor (VS Code Extension)
+- Same agent core, different UI
+- Inline suggestions + chat panel
+- File editing with diff view
+- Diagnostics integration
+- LSP-aware
+
+### 2.3 Headless (CLI/CI)
+- `zero exec "prompt"` for scripts
+- `zero serve` for daemon mode
+- JSON-RPC over stdio/WebSocket
+- MCP server support
+
+---
+
+## 3. Architecture
+
+```
+┌─────────────────────────────────────────────────┐
+│              ZERO PLATFORM                        │
+├─────────────────────────────────────────────────┤
+│                                                  │
+│  ┌──────────┐  ┌──────────┐  ┌──────────┐     │
+│  │   TUI    │  │   VSCode │  │ Headless │     │
+│  │ Surface  │  │ Surface  │  │ Surface  │     │
+│  └────┬─────┘  └────┬─────┘  └────┬─────┘     │
+│       │             │             │            │
+│       └─────────────┼─────────────┘            │
+│                     │                          │
+│              ┌──────▼──────┐                   │
+│              │  Agent Core │                   │
+│              │  (ReAct)    │                   │
+│              └──────┬──────┘                   │
+│                     │                          │
+│  ┌──────┬──────┬────┴────┬──────┬──────┐      │
+│  │Tools │Perms │Sessions │Provi │Memory│      │
+│  └──────┴──────┴─────────┴──────┴──────┘      │
+│                                                  │
+└─────────────────────────────────────────────────┘
+```
+
+**Layering Rules:**
+- Agent Core never imports a surface
+- Surfaces are thin shells (I/O + rendering)
+- All state persists to local files
+- No cloud dependency
+
+---
+
+## 4. Functional Requirements (24 Features)
+
+### F1 — Agent Core (P0)
+- ReAct loop: think → tool → result → think
+- Stream text + tool calls + thinking
+- Parallel tool execution where safe
+- Bound turns (max 100)
+- Clean interrupt (Esc/Ctrl-C)
+- Event stream: `text | tool_call | tool_result | thinking | usage | plan_update | error | turn_end`
+
+### F2 — Tools (P0)
+
+**Core (8):**
+- `read_file` — read with line ranges, images
+- `write_file` — create/overwrite
+- `edit_file` — exact-string edit (uniqueness guard)
+- `bash` — run shell command
+- `grep` — ripgrep content search
+- `glob` — file pattern matching
+- `apply_patch` — multi-hunk patches
+- `update_plan` — todo list
+
+**Advanced (4):**
+- `web_fetch` / `web_search` — web access
+- `task` — spawn subagent
+- `structured_output` — JSON-Schema validated
+- `notebook_edit` — Jupyter notebooks
+
+**Tool Requirements:**
+- Go typed parameter schemas -> JSON Schema
+- Side-effect class: `read | write | shell | network | out_of_workspace`
+- Structured results: `{ status, output, truncated?, meta? }`
+- Atomic execution
+- Permission gating
+
+### F3 — Providers & Model Registry (P0)
+- OpenAI-compatible (works with OpenAI, OpenRouter, Ollama, etc.)
+- Anthropic (Claude)
+- Google Gemini
+- Azure OpenAI
+- AWS Bedrock
+- xAI (Grok)
+
+**Model Registry:**
+
+```go
+type ModelEntry struct {
+  ID              string
+  Name            string
+  Provider        string
+  APIProviders    []string
+  ContextLimits   ContextLimits
+  ReasoningEffort ReasoningEffort
+  Capabilities    ModelCapabilities
+  Cost            ModelCost
+  Tier            string
+  MatchPatterns   []string
+}
+```
+
+- Per-task model selection (`-m`)
+- Provider rotation/failover
+- Cost tracking
+- "Spec model" for planning
+- Fallback model on failure
+
+### F4 — Terminal UI (P0)
+- **Layout:** header (cwd · git · model) → tool rows → diff → response → input → footer
+- **Streaming:** append-only transcript model with live tail updates
+- **Markdown:** full CommonMark + syntax highlighting
+- **Tool rows:** collapsible with one-line summaries
+- **Input:** `/` commands, `@` files, `!` bash, `Esc` interrupt
+- **Robustness:** resize, wide-char, ANSI, non-TTY fallback
+- **Splash:** ZERO wordmark
+- **Themes:** light/dark + custom
+
+### F5 — Sessions (P0)
+- Persist to `~/.local/share/zero/sessions/<id>/`
+- `meta.json` + `events.jsonl` + `messages.jsonl` + `checkpoints/`
+- `--resume [id]` (default: latest)
+- `--fork <id>` (copy with lineage)
+- Search: `zero search <query>` over messages/tools/results
+- Atomic writes (write-temp-then-rename)
+- Schema versioning
+- Export to Markdown/JSON
+
+### F6 — Configuration (P0)
+- Layered: defaults → user → project → env → CLI
+- `~/.config/zero/config.json` (user)
+- `./.zero/config.json` (project)
+- Runtime overlay (`--settings <path>`)
+- Feature flags for staged rollout
+- Validation on load
+
+### F7 — MCP (Model Context Protocol) (P0)
+- Connect to stdio/HTTP/SSE servers
+- Surface MCP tools into registry
+- `zero mcp add/remove/list`
+- Per-server and per-tool permissions
+- Persistent grants with revoke/clear
+- MCP server mode (zero as MCP server)
+
+### F8 — Headless / Programmatic (P0)
+- `zero exec [prompt]` — one-shot
+- `zero serve` — daemon mode
+- `--input-format stream-json` — line-delimited events
+- `--input-format stream-jsonrpc` — JSON-RPC protocol
+- `-o text|json` output formats
+- Exit codes: `0=success, 1=crash, 2=usage, 3=provider, 4=tool, 5=permission, 6=budget`
+
+### F9 — Permissions & Autonomy (P0)
+- **Side-effect classes:** read, write, shell, network, out_of_workspace
+- **Autonomy levels:** low (ask), medium (smart), high (auto)
+- `--skip-permissions-unsafe` (loud warning)
+- TUI: inline prompts (allow/deny/always)
+- Headless: default-deny unless granted
+- Persistent grants with list/revoke/clear
+- Audit log in `events.jsonl`
+
+**Permission Matrix:**
+
+| Tool Class | low | medium | high |
+|------------|-----|--------|------|
+| Read | allow | allow | allow |
+| Workspace write | prompt | allow | allow |
+| Shell | prompt | prompt | allow |
+| Network | prompt | allow | allow |
+| Out-of-workspace | deny | prompt | prompt |
+
+### F10 — Observability (P0)
+- Structured logging (stdout/stderr split)
+- Secret redaction (API keys, tokens, passwords)
+- No vendor telemetry (local only)
+- `zero doctor` — health check command
+- Crash reports (opt-in, local)
+- Performance metrics (TTFT, tokens/sec)
+
+### F11 — Subagents (P1)
+- `task` tool spawns scoped subagent
+- Built-in agents: code-review, security-review, refactor, test-gen
+- Recursion depth limit
+- Parent session linkage
+- Shared/isolated budgets
+
+### F12 — Plugins (P1)
+- User/project-scoped plugins
+- Commands, tools, prompts, hooks
+- Plugin manifest: `plugin.json`
+- `zero plugin install <path|url>`
+- No marketplace in v1 (manual install)
+
+### F13 — Distribution (P0)
+- **Single binary** via Go cross-compilation
+- npm package as a thin installer/launcher for the platform binary
+- Cross-platform: Linux, macOS, Windows
+- `zero update [--check]` self-update
+- Checksum verification
+- Rollback on failure
+
+### F14 — VS Code Extension (P1)
+- Same agent core via stream protocol
+- Inline chat panel
+- Diff view in editor
+- Diagnostics integration
+- LSP awareness
+- File watching
+
+### F15 — Memory & Context (P1)
+- `AGENTS.md` hierarchical memory
+- `@import` for modular rules
+- Auto-generated from `init` command
+- Project conventions learning
+- Skills system (reusable instruction packs)
+- Context budgeting (token tracking)
+- Auto-compaction when full
+
+### F16 — Hooks System (P1)
+- Lifecycle hooks: `PreToolUse`, `PostToolUse`, `SessionStart`, `SessionEnd`
+- User-defined shell commands
+- JSON output for conditional flow
+- `.zero/hooks/` directory
+
+### F17 — Slash Commands (P0)
+
+**Built-in (20):**
+- Session: `/clear`, `/compact`, `/context`, `/rewind`, `/resume`, `/exit`
+- Model: `/model`, `/effort`, `/style`, `/permissions`
+- Project: `/init`, `/memory`, `/skills`
+- Extensibility: `/mcp`, `/hooks`, `/agents`
+- Meta: `/help`, `/config`, `/doctor`, `/feedback`
+
+**Custom:** `./.zero/commands/*.md`
+
+### F18 — Git Integration (P1)
+- Worktree isolation per task
+- Auto-commit with messages
+- Branch management
+- PR creation
+- Diff review
+- Conflict resolution
+
+### F19 — Testing & Verification (P1)
+- `run_tests` tool
+- Self-verification loop
+- Test runner integration
+- Coverage reporting
+- Diagnostics (typecheck, lint)
+- LSP integration
+
+### F20 — Web Access (P1)
+- `web_fetch` — fetch URL content
+- `web_search` — search engine
+- Provider/permission gated
+- Rate limiting
+- Caching
+
+### F21 — Sandbox (P2)
+- Linux: bubblewrap
+- macOS: sandbox-exec
+- Windows: AppContainer
+- Filesystem policies
+- Network policies
+- Unix socket controls
+
+### F22 — Multi-modal (P2)
+- Image input (screenshots, diagrams)
+- PDF parsing
+- Voice input (optional)
+- File attachments
+- Paste support
+
+### F23 — Telemetry (P2)
+- Local-only metrics
+- Token usage tracking
+- Cost reports
+- Performance profiling
+- Crash reports (opt-in)
+- NO vendor telemetry
+
+### F24 — Advanced Features (P2)
+- Code review automation
+- Auto-refactor
+- Dependency updates
+- Security scanning
+- Documentation generation
+
+---
+
+## 5. Data Models
+
+### 5.1 Headless Event Stream
+
+```json
+{ "type": "run_start", "runId": "r1", "sessionId": "s1", "cwd": "/repo", "model": "..." }
+{ "type": "text", "delta": "..." }
+{ "type": "tool_call", "id": "t1", "name": "bash", "args": {...}, "sideEffect": "shell" }
+{ "type": "permission", "id": "t1", "decision": "needs_approval", "reason": "shell" }
+{ "type": "tool_result", "id": "t1", "status": "ok", "output": "...", "truncated": false }
+{ "type": "usage", "promptTokens": 1234, "completionTokens": 567, "costUsd": 0.012 }
+{ "type": "plan_update", "plan": [{ "id": "1", "content": "...", "status": "in_progress" }] }
+{ "type": "turn_end", "stopReason": "end_turn" }
+{ "type": "error", "code": "rate_limit", "message": "...", "recoverable": true }
+{ "type": "run_end", "status": "success", "exitCode": 0 }
+```
+
+### 5.2 Session On Disk
+
+```
+~/.local/share/zero/sessions/<id>/
+  meta.json        # { schemaVersion, id, parentId?, title, cwd, model, createdAt, updatedAt, tokens, cost }
+  events.jsonl     # append-only AgentEvent stream
+  messages.jsonl   # provider-facing messages
+  checkpoints/     # snapshots for restore/fork/compact
+```
+
+### 5.3 Tool Contract
+
+```go
+type Tool interface {
+  Name() string
+  Description() string
+  Parameters() JSONSchema
+  SideEffect() SideEffect
+  Execute(ctx ToolContext, args map[string]any) ToolResult
+}
+
+type ToolResult struct {
+  Status    string
+  Output    string
+  Truncated bool
+  Meta      map[string]any
+}
+```
+
+### 5.4 Config Schema
+
+```json
+{
+  "model": "claude-sonnet-4-6",
+  "providers": {
+    "anthropic": { "apiKey": "env:AN..._KEY" },
+    "openai": { "apiKey": "env:OP..._KEY" }
+  },
+  "permissions": {
+    "autonomy": "medium",
+    "grants": [{ "tool": "bash", "pattern": "git status", "decision": "allow" }]
+  },
+  "ui": { "theme": "dark", "syntaxHighlighting": true }
+}
+```
+
+---
+
+## 6. Non-Functional Requirements
+
+### Performance
+- TTFT < 500ms after provider's first byte
+- No full-transcript repaint per token
+- Cold start < 300ms (warm cache)
+- Tool execution < 100ms overhead
+- Session resume < 1s
+
+### Reliability
+- Crash-free TUI under resize, wide-char, non-TTY
+- Clean interrupt (no half-written state)
+- Atomic session writes
+- Recovery from malformed JSONL
+- Provider failover
+
+### Security
+- No mutation without permission decision (hard invariant)
+- API keys never logged
+- Secret redaction everywhere
+- Headless stdout free of logs
+- Workspace boundary by default
+- Sandboxing for untrusted code
+
+### Portability
+- Linux, macOS, Windows
+- Any OpenAI-compatible endpoint
+- x86_64, ARM64
+- Single binary distribution
+
+### Testability
+- Mock provider for unit tests
+- Fake MCP server/client
+- Fake terminal/PTY
+- Sandbox simulator
+- Deterministic event replay
+
+---
+
+## 7. Milestones
+
+### M0 — Foundation (Weeks 1-2)
+- Go module setup
+- Agent core skeleton
+- 8 core tools (read, write, edit, bash, grep, glob, apply_patch, update_plan)
+- OpenAI-compatible provider
+- Basic Bubble Tea TUI
+- Permission gate (read vs mutate)
+- Session persistence
+
+**Exit:** `zero` runs, can read/edit files, basic chat works.
+
+### M1 — Multi-Provider (Weeks 3-4)
+- Anthropic provider
+- Gemini provider
+- Model registry
+- Per-task model selection
+- Provider rotation
+- Cost tracking
+- Headless `exec` mode
+
+**Exit:** Switch between Claude/GPT/Gemini, see cost, run headless.
+
+### M2 — Polish (Weeks 5-8)
+- Advanced TUI features
+- Full markdown + syntax highlighting
+- Tool result truncation/pagination
+- Auto-compaction
+- Context budgeting
+- 20 slash commands
+- Doctor command
+- Search command
+
+**Exit:** Production-ready TUI, smooth streaming, all commands work.
+
+### M3 — Distribution (Weeks 9-10)
+- Single binary build
+- Self-update
+- Cross-platform CI
+- VS Code extension MVP
+- Documentation site
+
+**Exit:** `zero` ships as one binary, works on Win/Mac/Linux.
+
+### M4 — Extensibility (Weeks 11-14)
+- MCP client + permissions
+- Hooks system
+- Skills system
+- Memory (AGENTS.md)
+- Plugin loading
+- Subagents (built-in: review, security)
+
+**Exit:** Plugins work, MCP servers connect, memory persists.
+
+### M5 — Advanced (Weeks 15-20)
+- Web fetch/search
+- Git worktrees
+- Self-verification
+- Diagnostics (LSP, typecheck)
+- Testing integration
+- Code review automation
+
+**Exit:** Zero can review code, run tests, manage git.
+
+### M6 — Sandbox & Security (Weeks 21-24)
+- Linux sandbox (bubblewrap)
+- macOS sandbox
+- Windows AppContainer
+- Network policies
+- Filesystem policies
+
+**Exit:** Untrusted code runs safely in sandbox.
+
+### Beyond v1.0
+- Multi-modal (images, PDFs)
+- Voice input
+- Cloud sync (optional)
+- Team features
+- Marketplace (if demand)
+
+---
+
+## 8. Risks & Mitigations
+
+| # | Risk | Mitigation |
+|---|------|------------|
+| R1 | Scope creep | Stick to milestones, defer features |
+| R2 | Provider API changes | Abstract via interface, test with mocks |
+| R3 | TUI complexity | Use Bubble Tea/Lip Gloss, follow mature terminal agent UX patterns |
+| R4 | Performance | Benchmark early, optimize streaming |
+| R5 | Security vulnerabilities | Permission gate from day 1, security review |
+| R6 | Cross-platform bugs | CI on Win/Mac/Linux from M0 |
+| R7 | Breaking changes | Schema versioning, migration tools |
+
+---
+
+## 9. Success Metrics
+
+### v1.0 (6 months)
+- 5+ providers supported
+- 12+ tools
+- 100% test coverage on core
+- < 300ms cold start
+- Single binary < 50MB
+- Works on Win/Mac/Linux
+- 1000+ GitHub stars
+- 50+ contributors
+
+### v2.0 (12 months)
+- VS Code extension published
+- MCP server mode
+- Plugin ecosystem (20+ plugins)
+- 10K+ active users
+- Compete with Claude Code on benchmarks
+
+---
+
+## 10. Out of Scope (v1.0)
+
+- Cloud sync / hosted backend
+- Relay / remote SSH execution
+- Multi-agent mission mode
+- Plugin marketplace
+- Auto-generated wikis
+- Git authorship notes
+- Vendor telemetry
+- Voice input
+- Mobile support
+
+---
+
+## 11. Open Decisions
+
+1. **TUI library:** Bubble Tea + Lip Gloss — **Decide: Go-native TUI** (single-binary friendly, mature terminal ecosystem)
+2. **Storage:** JSONL files vs SQLite — **Decide: JSONL** (inspectable, simple)
+3. **Package structure:** Single vs monorepo — **Decide: Single** until M4
+4. **Provider abstraction:** Direct API vs wrapper — **Decide: Direct** (portability)
+5. **Patch strategy:** apply_patch vs edit_file — **Decide: Both** (apply_patch for bulk, edit_file for small)
+
+---
+
+## 12. Team & Timeline
+
+**Team (3 people):**
+- You (lead + TUI + tools)
+- Backend dev (providers + core)
+- Designer (VS Code extension + docs)
+
+**Timeline:**
+- M0-M1: 4 weeks (foundation)
+- M2-M3: 6 weeks (polish + distribution)
+- M4-M5: 8 weeks (extensibility)
+- M6: 4 weeks (sandbox)
+- **Total v1.0: 6 months**
+
+---
+
+## 13. Why This Will Work
+
+1. **Proven patterns** — Based on clean-room analysis of existing terminal coding-agent UX patterns
+2. **Multi-provider** — Unique advantage vs Claude Code/Cursor
+3. **Open source** — Community contributions
+4. **Go** — Native binary, fast startup, strong concurrency, cross-platform CLI ergonomics
+5. **npm distribution wrapper** — Keeps JavaScript ecosystem installation while the runtime stays native
+6. **Clear milestones** — Ship early, iterate
+
+---
+
+**This is the real PRD. 13 sections, 24 features, 6 months to v1.0.**

docs/README.md

@@ -1,24 +0,0 @@
-# Zero Documentation
-
-This directory contains public documentation for using, extending, and releasing
-Zero.
-
-## User Docs
-
-- [Install](INSTALL.md)
-- [Update flow](UPDATE.md)
-- [OAuth logins and subscription-backed providers](oauth-subscriptions.md)
-
-## Automation And Integrations
-
-- [Stream-JSON protocol](STREAM_JSON_PROTOCOL.md)
-- [GitHub Action](GITHUB_ACTION.md)
-- [Specialists](SPECIALISTS.md)
-
-## Maintainer Docs
-
-- [How Zero Works](HOW_ZERO_WORKS.md)
-- [npm wrapper smoke checklist](NPM_WRAPPER_SMOKE.md)
-- [Benchmarks](BENCHMARK.md)
-- [Performance benchmarks](PERFORMANCE.md)
-- [Offline agent evals](AGENT_EVALS.md)

docs/SPECIALISTS.md

@@ -1,140 +0,0 @@
-# Zero Specialists
-
-Specialists are named sub-agents that Zero can delegate focused work to through
-the `Task` tool. A specialist is a markdown manifest with YAML-style
-frontmatter plus a system prompt body.
-
-Specialists can be built in, user-scoped, or project-scoped:
-
-| Scope | Path | Notes |
-| --- | --- | --- |
-| Built-in | compiled into Zero | `worker`, `explorer`, and `code-review` ship with the binary. |
-| User | `~/.config/zero/specialists/*.md` | Available across local workspaces. |
-| Project | `.zero/specialists/*.md` | Shared with the current repository when committed. |
-
-Project specialists override user and built-in specialists with the same name.
-User specialists override built-ins.
-
-## CLI Management
-
-```bash
-zero specialist list
-zero specialist show worker
-zero specialist path
-
-zero specialist create api-review \
-  --project \
-  --description "Reviews API changes" \
-  --tools read-only,plan \
-  --prompt "Review API changes for compatibility and missing tests."
-
-zero specialist edit api-review --project
-zero specialist delete api-review --project
-```
-
-Use `--json` with `list`, `show`, `path`, `create`, or `delete` when scripting.
-`create --force` replaces an existing manifest, but refuses symlink overwrites.
-`edit` also refuses symlink manifests before opening `$VISUAL` or `$EDITOR`.
-
-## Manifest Format
-
-```markdown
----
-name: api-review
-description: Reviews API changes for compatibility and missing tests.
-tools:
-  - read-only
-  - plan
----
-
-Review API changes for behavior regressions, compatibility breaks, and missing
-tests. Report concrete findings with file paths.
-```
-
-Supported frontmatter keys:
-
-| Key | Purpose |
-| --- | --- |
-| `name` | Lowercase specialist id. Use letters, numbers, and dashes. |
-| `description` | Short summary shown in listings and task metadata. |
-| `extends` | Optional base specialist to inherit prompt/model/tools from. |
-| `model` | Optional model override. Empty means inherit the parent model. |
-| `reasoningEffort` | Optional reasoning effort override. |
-| `tools` | Array of tool categories or tool ids. |
-
-If the body is empty and `description` is set, Zero uses the description as the
-system prompt and reports a warning in `zero specialist show`.
-
-## Tool Selection
-
-Known categories:
-
-| Category | Tools |
-| --- | --- |
-| `read-only` | `read_file`, `list_directory`, `grep`, `glob` |
-| `edit` | read-only tools plus `write_file`, `edit_file`, `apply_patch` |
-| `execute` | read-only tools plus `bash` |
-| `plan` | `update_plan` |
-
-Specialist manifests cannot enable `Task`, `TaskOutput`, `TaskStop`, or
-`GenerateSpecialist`, so child specialists cannot spawn more specialists or
-author new ones.
-
-## Agent Tools
-
-Zero registers these tools for top-level agent runs:
-
-| Tool | Purpose |
-| --- | --- |
-| `Task` | Launch a specialist sub-agent for a focused prompt. |
-| `TaskOutput` | Read or block on a background specialist task's output. |
-| `TaskStop` | Stop a running background specialist task. |
-| `GenerateSpecialist` | Create a project-local specialist manifest from a description. |
-
-`GenerateSpecialist` is project-scoped only. It writes to
-`.zero/specialists`, not the user specialist directory.
-
-Example LLM-facing `Task` payload:
-
-```json
-{
-  "name": "explorer",
-  "description": "Find session storage code",
-  "prompt": "Find the files that create, load, and list sessions."
-}
-```
-
-Background task payload:
-
-```json
-{
-  "name": "worker",
-  "description": "Audit release docs",
-  "prompt": "Check the release docs for stale TypeScript references.",
-  "run_in_background": true
-}
-```
-
-The returned `task_id` is also the child session id. Use it with
-`TaskOutput`, `TaskStop`, or `Task` resume.
-
-## Background State
-
-Background specialist output is stored under:
-
-```text
-${XDG_DATA_HOME:-~/.local/share}/zero/background/
-```
-
-Each task has:
-
-- `<task_id>.ndjson` for the child process stream output
-- `<task_id>.json` for task metadata such as status, PID, parent session, and
-  timestamps
-
-Persisted metadata lets a new background manager instance read completed task
-output or stop a still-running task by id.
-
-If Zero is restarted while a background task is still marked `running`, the new
-manager marks that task `error` and clears its PID. This avoids sending
-`TaskStop` to a stale PID that may now belong to an unrelated process.

docs/STREAM_JSON_PROTOCOL.md

@@ -8,12 +8,12 @@ redacted before they are written to stdout.
 
 ## Version
 
-Current schema version: `2`
+Current schema version: `1`
 
 Every input and output event must include:
 
 ```json
-{ "schemaVersion": 2, "type": "..." }
+{ "schemaVersion": 1, "type": "..." }
 ```
 
 Output events also include `runId`.
@@ -24,42 +24,27 @@ Output events also include `runId`.
 `--file`.
 
 ```json
-{ "schemaVersion": 2, "type": "message", "role": "user", "content": "Inspect this repo." }
-{ "schemaVersion": 2, "type": "prompt", "content": "Return only blockers." }
+{ "schemaVersion": 1, "type": "message", "role": "user", "content": "Inspect this repo." }
+{ "schemaVersion": 1, "type": "prompt", "content": "Return only blockers." }
 ```
 
 Zero combines accepted input event content in order, separated by blank lines.
 Unknown fields are rejected so protocol clients catch drift early.
 
-Schema version `2` renamed sandbox permission metadata from `violation` to
-`block`. Clients should read the optional `block` object on permission and tool
-events when present.
-
 ## Output Events
 
 `zero exec --output-format stream-json` emits schema-versioned JSONL events.
 
 ```json
-{ "schemaVersion": 2, "type": "run_start", "runId": "run_20260603_abc123", "sessionId": "zero_20260603100000_abc123", "cwd": "/repo", "provider": "openai", "model": "gpt-4.1", "apiModel": "gpt-4.1" }
-{ "schemaVersion": 2, "type": "reasoning", "runId": "run_20260603_abc123", "delta": "Thinking..." }
-{ "schemaVersion": 2, "type": "text", "runId": "run_20260603_abc123", "delta": "..." }
-{ "schemaVersion": 2, "type": "tool_call", "runId": "run_20260603_abc123", "id": "call_1", "name": "read_file", "args": { "path": "README.md" }, "sideEffect": "read" }
-{ "schemaVersion": 2, "type": "permission_request", "runId": "run_20260603_abc123", "id": "call_2", "name": "write_file", "action": "prompt", "permission": "prompt", "permissionMode": "ask", "sideEffect": "write", "reason": "Creates or overwrites files." }
-{ "schemaVersion": 2, "type": "permission_decision", "runId": "run_20260603_abc123", "id": "call_2", "name": "write_file", "action": "allow", "permission": "prompt", "permissionGranted": true, "decisionReason": "approved in TUI" }
-{ "schemaVersion": 2, "type": "tool_result", "runId": "run_20260603_abc123", "id": "call_1", "status": "ok", "output": "...", "truncated": false }
-{ "schemaVersion": 2, "type": "usage", "runId": "run_20260603_abc123", "promptTokens": 12, "completionTokens": 8, "totalTokens": 20 }
-{ "schemaVersion": 2, "type": "final", "runId": "run_20260603_abc123", "text": "..." }
-{ "schemaVersion": 2, "type": "run_end", "runId": "run_20260603_abc123", "status": "success", "exitCode": 0 }
-```
-
-`reasoning` events carry live model reasoning/status deltas for providers that
-stream them separately from answer text. They are liveness/progress events only:
-they are not folded into `text` or the final answer.
-
-Permission events may include structured sandbox metadata:
-
-```json
-{ "schemaVersion": 2, "type": "permission_request", "runId": "run_20260603_abc123", "id": "call_3", "name": "bash", "action": "prompt", "permission": "prompt", "permissionMode": "ask", "sideEffect": "shell", "reason": "network access requires approval", "risk": { "level": "critical", "categories": ["network"] }, "block": { "code": "network", "toolName": "bash", "action": "prompt", "risk": { "level": "critical", "categories": ["network"] }, "reason": "network access requires approval", "recoverable": true } }
+{ "schemaVersion": 1, "type": "run_start", "runId": "run_20260603_abc123", "sessionId": "zero_20260603100000_abc123", "cwd": "/repo", "provider": "openai", "model": "gpt-4.1", "apiModel": "gpt-4.1" }
+{ "schemaVersion": 1, "type": "text", "runId": "run_20260603_abc123", "delta": "..." }
+{ "schemaVersion": 1, "type": "tool_call", "runId": "run_20260603_abc123", "id": "call_1", "name": "read_file", "args": { "path": "README.md" }, "sideEffect": "read" }
+{ "schemaVersion": 1, "type": "permission_request", "runId": "run_20260603_abc123", "id": "call_2", "name": "write_file", "action": "prompt", "permission": "prompt", "permissionMode": "ask", "sideEffect": "write", "reason": "Creates or overwrites files." }
+{ "schemaVersion": 1, "type": "permission_decision", "runId": "run_20260603_abc123", "id": "call_2", "name": "write_file", "action": "allow", "permission": "prompt", "permissionGranted": true, "decisionReason": "approved in TUI" }
+{ "schemaVersion": 1, "type": "tool_result", "runId": "run_20260603_abc123", "id": "call_1", "status": "ok", "output": "...", "truncated": false }
+{ "schemaVersion": 1, "type": "usage", "runId": "run_20260603_abc123", "promptTokens": 12, "completionTokens": 8, "totalTokens": 20 }
+{ "schemaVersion": 1, "type": "final", "runId": "run_20260603_abc123", "text": "..." }
+{ "schemaVersion": 1, "type": "run_end", "runId": "run_20260603_abc123", "status": "success", "exitCode": 0 }
 ```
 
 Errors are part of the protocol and are followed by `run_end`.
@@ -70,6 +55,6 @@ is not pre-approved, Zero may emit `permission_request` followed by a denied
 allows, denies, or always-allows the request.
 
 ```json
-{ "schemaVersion": 2, "type": "error", "runId": "run_20260603_abc123", "code": "provider_error", "message": "...", "recoverable": false }
-{ "schemaVersion": 2, "type": "run_end", "runId": "run_20260603_abc123", "status": "error", "exitCode": 3 }
+{ "schemaVersion": 1, "type": "error", "runId": "run_20260603_abc123", "code": "provider_error", "message": "...", "recoverable": false }
+{ "schemaVersion": 1, "type": "run_end", "runId": "run_20260603_abc123", "status": "error", "exitCode": 3 }
 ```

docs/UPDATE.md

@@ -1,44 +1,38 @@
-# Update Flow
+# Zero Update Flow
 
-`zero update --check` checks the latest GitHub release and compares it with the
-local CLI version.
+`zero update --check` checks the latest GitHub release for `Gitlawb/zero` and compares it with the local CLI version.
 
 ```bash
 zero update --check
 zero update --check --json
 zero update --check --repo Gitlawb/zero
+zero update --check --endpoint https://api.github.com/repos/Gitlawb/zero/releases/latest
+zero update --check --timeout 3s
 zero update --check --target windows-x64
 ```
 
-The command is intentionally check-only:
+For M2 this command is intentionally check-only:
 
 - It does not replace the running binary.
-- It exits with code `0` when the check succeeds, even when an update is
-  available.
+- It exits with code `0` when the check succeeds, even when an update is available.
 - It exits with code `1` when the release check cannot be completed.
-- `--json` prints the same result in a machine-readable format for scripts and
-  CI.
+- `--json` prints the same result in a machine-readable format for scripts and CI.
+- `--repo <owner/repo>` checks a different GitHub repository when no endpoint is provided.
+- `--endpoint <url|owner/repo>` checks a specific release API URL or repository slug.
+- `--timeout <duration>` overrides the default release check timeout.
+- `--target <platform-arch>` validates release metadata for another supported target from the current machine.
+- Release checks time out after 5 seconds by default.
+- It validates that the latest release includes the expected archive and matching `.sha256` asset for the selected platform target.
 
-Useful flags:
+Supported targets are `linux-x64`, `linux-arm64`, `macos-x64`, `macos-arm64`, `windows-x64`, and `windows-arm64`. Without `--target`, Zero checks the current platform.
 
-| Flag | Purpose |
-|---|---|
-| `--repo <owner/repo>` | Check another GitHub repository. |
-| `--endpoint <url|owner/repo>` | Check a specific release API URL or repository slug. |
-| `--timeout <duration>` | Override the default release check timeout. |
-| `--target <platform-arch>` | Validate release metadata for another supported target. |
+The release endpoint resolves in this order:
 
-Supported targets are `linux-x64`, `linux-arm64`, `macos-x64`, `macos-arm64`,
-`windows-x64`, and `windows-arm64`. Without `--target`, Zero checks the current
-platform.
+- `--endpoint` from the CLI, or `Options.Endpoint` when calling `update.Check` from code.
+- `ZERO_UPDATE_RELEASE_URL` from the environment.
+- `--repo` from the CLI, or `Options.Repository` when calling `update.Check` from code.
+- `https://api.github.com/repos/Gitlawb/zero/releases/latest`.
 
-Endpoint resolution order:
+`--endpoint`, `Options.Endpoint`, and `ZERO_UPDATE_RELEASE_URL` may be a full URL or an `owner/repo` slug.
 
-1. `--endpoint`
-2. `ZERO_UPDATE_RELEASE_URL`
-3. `--repo`
-4. `https://api.github.com/repos/Gitlawb/zero/releases/latest`
-
-Installer scripts download the matching release asset for the local platform and
-verify its `.sha256` file. If Zero is already installed, run `zero update --check`
-before reinstalling.
+Installer scripts download the matching release asset for the local platform and verify its `.sha256` file. If Zero is already installed, use this command before re-running the installer.

docs/WORK_SPLIT_PRD.md

@@ -0,0 +1,286 @@
+# Zero WorkSplit PRD v3: Go-Native Product Ownership Plan
+
+Status: Draft v3.0
+Date: 2026-06-06
+Timeline: Active sprint plan toward v0.1, then v1.0 hardening
+
+Team:
+
+- Vasanth
+- Gnanam
+- Anandan
+
+## 1. Product Goal
+
+Zero is a terminal-first AI coding agent built in Go.
+
+The goal is not just to remove the old TypeScript runtime. The goal is to make Zero feel like a mature daily-driver CLI agent:
+
+- fast startup
+- strong command UX
+- safe autonomy
+- clear permission handling
+- useful local tools
+- reliable provider/runtime behavior
+- installable, updateable, and testable on real developer machines
+
+This document replaces the older migration-era work split. The Go runtime is now the main product baseline, so future work should focus on product depth rather than repeating the old M0 migration plan.
+
+## 2. Current Baseline On Main
+
+The current main branch is Go-native for the runtime and CLI.
+
+Implemented baseline:
+
+- `cmd/zero` is the primary CLI entrypoint.
+- `cmd/zero-pr-review` exists for PR automation workflows.
+- Core runtime packages live under `internal/`.
+- Current CLI surface includes `exec`, `config`, `models`, `providers`, `doctor`, `search`, `sessions`, `plugins`, `hooks`, `mcp`, `sandbox`, `update`, `worktrees`, `verify`, `changes`, `serve`, `help`, and `version`.
+- Core areas exist for agent flow, provider/model registry, tools, TUI, sessions, search, sandbox, self-verification, stream-json, usage, redaction, hooks, MCP, plugins, update, worktrees, and git/change handling.
+- TypeScript is no longer part of the repository baseline. The npm package keeps a small Node wrapper that delegates to the Go binary.
+
+Still needed before Zero feels production-ready:
+
+- Permission and sandbox UX must become clearer and harder to misuse.
+- TUI command flows need polish beyond basic rendering.
+- Platform behavior, especially Windows sandbox/update/install paths, needs stronger product ownership.
+- Review automation should produce useful, repeatable output for maintainers.
+- Release trust needs checksums, install smoke tests, and update verification.
+
+## 3. Ownership Rules
+
+- Every feature has one DRI.
+- UI, runtime, and platform work should be split by contract, not by vague file ownership.
+- Every PR must expose user-runnable behavior or a concrete contract consumed by another owner.
+- Docs, CI, and website work can support a product PR, but they should not be the whole lane for any owner.
+- Anandan must own platform product work, not only docs/site/CI.
+- Shared contracts should be agreed before large UI or integration work lands.
+- Keep PRs large enough to move the product forward, but still reviewable.
+
+## 4. Team Roles
+
+| Owner | Primary Role | Owns | Should Not Own Alone |
+| --- | --- | --- | --- |
+| Vasanth | Product UX and TUI owner | Bubble Tea TUI, startup screen, command palette, slash command UX, tool rendering, permission prompts, sandbox violation rendering, verification/test/git output UX, themes, daily CLI feel | Provider internals, release pipeline, platform adapters |
+| Gnanam | Runtime core owner | Provider/model registry, provider factory, agent/runtime protocols, stream-json, sessions/search backends, config/doctor/usage/redaction, MCP/hooks/plugins backend, sandbox policy/grants, permission event contract | TUI composition, release packaging, platform-specific install trust |
+| Anandan | Platform product owner | Install/update/release trust, binary/package smoke tests, platform sandbox adapters, Windows behavior, PR/review automation CLI, CI as product verification, performance and release checks | TUI rendering, provider semantics, model registry ownership |
+
+## 5. Anandan Platform Product Rule
+
+Anandan should not be assigned only docs/site/CI forever. His lane must produce platform behavior that users and maintainers can run.
+
+Accepted Anandan work:
+
+- `zero sandbox policy` improvements and platform-specific sandbox behavior.
+- Windows sandbox adapter behavior and clear fallback reporting.
+- `zero update` verification, release metadata validation, and install smoke tests.
+- Package and binary checksums.
+- PR automation CLI workflows that reviewers can run locally or in CI.
+- CI jobs that prove real Zero commands work after installation.
+
+Not accepted as standalone Anandan work:
+
+- Docs-only updates.
+- Website-only updates.
+- Future VS Code notes with no current CLI behavior.
+- CI matrix changes that do not prove a user-facing command or release flow.
+- Release notes or checklists without executable validation.
+
+Docs and CI are still useful, but they must attach to real platform product behavior.
+
+## 6. Shared Contracts
+
+| Contract | DRI | Consumers | Purpose |
+| --- | --- | --- | --- |
+| Provider stream events | Gnanam | Vasanth, Anandan | Lets TUI and automation render model output, tool calls, and errors consistently. |
+| Permission and sandbox events | Gnanam | Vasanth, Anandan | Lets TUI ask for decisions and platform code enforce them without bypasses. |
+| Session event schema | Gnanam | Vasanth, Anandan | Lets TUI, resume, search, and automation share one session model. |
+| Command metadata registry | Vasanth | Gnanam, Anandan | Lets slash commands, help, shell completions, and docs use one command source. |
+| Platform/release contract | Anandan | Vasanth, Gnanam | Defines install, update, checksum, smoke, and platform capability results. |
+| Stream-json protocol | Gnanam | Anandan | Lets CI and PR automation consume Zero output without scraping text UI. |
+
+## 7. Active Workstreams
+
+### A. Safe Autonomy
+
+Top priority. Zero must be able to run tools confidently while showing users what is happening and asking when risk increases.
+
+Planned PRs:
+
+- Vasanth: `feat/tui-sandbox-permissions`
+  - permission prompt UX
+  - allow, deny, and always-allow flows
+  - sandbox violation rendering
+  - `/permissions` or equivalent permission state UI
+- Gnanam: `feat/runtime-permission-events`
+  - structured permission request events
+  - sandbox grant contract
+  - stream-json support
+  - tests for denial, approval, and violation cases
+- Anandan: `feat/windows-sandbox-platform`
+  - platform sandbox capability checks
+  - stronger Windows behavior or explicit safe fallback
+  - `zero sandbox policy --json` output useful enough for CI and support
+  - platform tests where possible
+
+Done when:
+
+- High-autonomy tool attempts produce a clear prompt, decision, and result in TUI and headless modes.
+- Denied commands do not run.
+- Sandbox violations are visible and actionable.
+- Platform fallback is explicit, tested, and not silently unsafe.
+
+### B. Command UX And Daily CLI
+
+The CLI should feel like a product, not a collection of backend commands.
+
+Planned PRs:
+
+- Vasanth: `feat/tui-command-polish`
+  - `/model`, `/provider`, `/context`, `/doctor`, `/verify`, `/resume`, and `/compact` become clear interactive flows
+  - command results render consistently
+  - startup screen stays minimal until the user asks for panels
+- Gnanam: `feat/runtime-command-contracts`
+  - stable command data APIs
+  - provider/model/session/config result structs
+  - error types suitable for TUI rendering
+- Anandan: `feat/install-update-user-flow`
+  - `zero update` and install flows are user-runnable
+  - package smoke command verifies installed binary behavior
+  - CI proves the packaged command starts and reports version/help
+
+Done when:
+
+- A new user can run Zero, inspect config/provider/model state, and understand the next action without reading source.
+
+### C. Verification And Review Automation
+
+Zero should help maintainers verify changes and review PRs with repeatable local commands.
+
+Planned PRs:
+
+- Vasanth: `feat/tui-verification-results`
+  - render verify/test/git/review results in TUI
+  - show pass/fail/blocked states clearly
+- Gnanam: `feat/runtime-self-verify-contract`
+  - structured verification results
+  - stable test/change event model
+  - redacted logs for model and automation use
+- Anandan: `feat/pr-review-cli-workflow`
+  - make `cmd/zero-pr-review` or `zero review` useful for changed files, validations, and summary output
+  - support local and CI invocation
+  - avoid brittle text scraping where stream-json can be used
+
+Done when:
+
+- A maintainer can run one local command before review and get useful changed-file, validation, and risk output.
+
+### D. Release Trust And Distribution
+
+This is Anandan's main product lane after platform sandbox work.
+
+Planned PRs:
+
+- Anandan: `feat/release-trust`
+  - release metadata validation
+  - checksums
+  - install smoke tests
+  - update verification
+  - clear error output when release artifacts are invalid
+- Vasanth:
+  - consume release/update results in CLI UX where needed
+- Gnanam:
+  - ensure config, errors, and logs are structured and redacted
+
+Done when:
+
+- Users can install, run, verify, and update Zero with confidence.
+- Maintainers can prove release artifacts work before publishing.
+
+### E. Extensibility UX
+
+This comes after safe autonomy and daily CLI polish.
+
+Planned PRs:
+
+- Vasanth:
+  - `/mcp`, `/hooks`, `/plugins`, and related UI flows
+- Gnanam:
+  - harden backend contracts and lifecycle behavior
+- Anandan:
+  - automation and docs only after executable behavior exists
+
+Done when:
+
+- Extensibility feels discoverable and safe from the terminal.
+
+## 8. Immediate Next PRs
+
+These are the next reviewable slices. Do not use the old migration-era next steps.
+
+1. Vasanth: `feat/tui-sandbox-permissions`
+2. Gnanam: `feat/runtime-permission-events`
+3. Anandan: `feat/windows-sandbox-platform`
+4. Anandan: `feat/update-release-verification`
+5. Vasanth: `feat/tui-command-polish`
+6. Gnanam: `feat/session-rewind-compaction-contract`
+
+Recommended first parallel set:
+
+- Vasanth starts the TUI permission prompt and violation display.
+- Gnanam defines the permission event/grant contract.
+- Anandan owns Windows/platform sandbox reporting and tests.
+
+## 9. Command Ownership Matrix
+
+| Command Area | Vasanth | Gnanam | Anandan |
+| --- | --- | --- | --- |
+| `exec` | TUI and interactive UX | Runtime execution contract | Package smoke and platform validation |
+| `sandbox` | Prompt and violation UX | Policy, grants, and events | Platform adapters and Windows behavior |
+| `update` | Display and user messaging | Config and redacted errors | Primary owner of update verification |
+| `verify` | Result rendering | Verification backend contract | CI and release integration |
+| `changes` | TUI rendering | Change detection backend | PR automation integration |
+| `mcp`, `hooks`, `plugins`, `serve` | Discoverability and UI | Backend lifecycle and contracts | Automation/docs after behavior exists |
+| `sessions`, `search` | Resume/search UX | Storage, query, and event model | Stream-json and automation consumers |
+
+## 10. Definition Of Done
+
+For every code PR:
+
+- Add or update tests for the changed behavior.
+- Run `go test ./...` when Go files changed.
+- Run the relevant build command when CLI entrypoints changed.
+- Run package smoke checks when install/update/npm wrapper files changed.
+- Include manual CLI output or screenshots only when they prove a user-facing flow.
+
+For docs-only PRs:
+
+- State clearly that no runtime behavior changed.
+- Do not use docs-only PRs as a substitute for platform product work.
+
+For Anandan's PRs specifically:
+
+- Must include one of:
+  - user-runnable command behavior
+  - install/update/release artifact verification
+  - platform adapter behavior
+  - CI that executes real installed Zero commands
+- Docs/CI can support the PR, but cannot be the entire value.
+
+## 11. Review Rotation
+
+- Vasanth PRs:
+  - Gnanam reviews runtime contracts.
+  - Anandan reviews platform/release impact when touched.
+- Gnanam PRs:
+  - Vasanth reviews user-facing contract impact.
+  - Anandan reviews platform/protocol impact when touched.
+- Anandan PRs:
+  - Vasanth reviews user-facing CLI behavior.
+  - Gnanam reviews runtime contract and safety impact.
+
+## 12. Retired Historical Plan
+
+The old M0-M6 TypeScript-to-Go migration schedule is now historical.
+
+Do not assign new work from the old migration next-steps section. The Go migration foundation has already landed on main. Future work should move Zero from "ported" to "mature terminal coding agent."

docs/oauth-subscriptions.md

@@ -1,222 +0,0 @@
-# OAuth logins & using ChatGPT / Claude subscriptions
-
-Zero supports two distinct things people mean by "log in with OAuth":
-
-1. **OAuth login for a provider/gateway** that issues a *standard* bearer token —
-   fully built in (`zero auth login …`). Zero attaches the token to model calls
-   automatically.
-2. **Using a ChatGPT or Claude *subscription*** (Plus / Pro / Max) instead of a
-   pay-per-token API key — only possible through a **local proxy**, for the
-   reasons documented below. Zero ships a convenience preset and this recipe.
-
----
-
-## 1. OAuth login for a provider or gateway (built in)
-
-For any OAuth 2.0 / OIDC provider that returns a normal access token usable as
-`Authorization: Bearer …` on its API, configure it with `ZERO_OAUTH_<NAME>_*`
-env vars and log in:
-
-```sh
-export ZERO_OAUTH_ACME_CLIENT_ID=…
-export ZERO_OAUTH_ACME_AUTHORIZE_URL=https://acme.example/oauth/authorize
-export ZERO_OAUTH_ACME_TOKEN_URL=https://acme.example/oauth/token
-export ZERO_OAUTH_ACME_SCOPES="openid profile"
-zero auth login acme            # browser (loopback); --device for headless
-zero auth status
-```
-
-When a login exists for a provider, the **OpenAI and Anthropic** providers send
-`Authorization: Bearer <fresh-token>` (auto-refreshed; one refresh-and-retry on a
-`401`) instead of the API key. With no login they use the API key exactly as
-before. Tokens are stored 0600 (or the OS keyring with
-`ZERO_OAUTH_STORAGE=keyring`) and never logged. See `zero auth --help`.
-
-### In the setup wizard (`/provider`)
-
-Running `/provider` opens a **"How do you want to connect?"** chooser:
-
-```text
-❯ Sign in with OAuth                 One-click browser login (OpenRouter, xAI, ChatGPT, Hugging Face)
-  Paste an API key / browse providers  Any of 20+ providers, local, or a proxy
-```
-
-Pick **Sign in with OAuth** → the list of providers that do real OAuth → choose one:
-
-```text
-❯ OpenRouter      browser sign-in · creates a key
-  xAI (Grok)      browser or device code
-  ChatGPT         browser (Codex backend, ChatGPT Plus/Pro)
-  Hugging Face    browser or device code
-```
-
-- **OpenRouter / xAI / ChatGPT / Hugging Face** are real OAuth: your browser
-  opens to approve → done (no key to paste). OpenRouter mints a key; xAI /
-  ChatGPT / Hugging Face store a refreshable bearer. Hugging Face requires a
-  one-time OAuth-app registration (no secret needed for "public" apps); the
-  preset pre-fills scopes, endpoints, and the OIDC issuer. The same chooser
-  appears in first-run onboarding. (xAI uses an opt-in preset — set
-  `ZERO_OAUTH_ALLOW_PRESETS=1` or your own `ZERO_OAUTH_XAI_*`; see below.)
-- **Device code (headless / SSH):** for a provider that supports it (xAI,
-  Hugging Face), press **d** on the list to get a code to enter on another
-  device instead of opening a browser. On an SSH session or headless Linux box
-  (no `DISPLAY`) device code is used automatically; set `ZERO_OAUTH_DEVICE=1`
-  to force it anywhere. The CLI equivalent is
-  `zero auth login <name> --device`.
-- **ChatGPT / Claude are intentionally not in this list for the proxy path** —
-  use the dedicated `chatgpt-proxy` / `custom-anthropic-compatible` preset
-  (see §2) for subscription-via-proxy. ChatGPT *is* a first-class OAuth
-  provider in this version (routes to the Codex backend) — see "Built-in OAuth
-  providers" below.
-
-### Built-in OAuth providers
-
-- **OpenRouter (no env needed)** — `zero auth openrouter` opens a browser, you
-  approve, and it **mints an OpenRouter API key** (public PKCE flow, no client_id).
-  In the interactive setup wizard, pick **OpenRouter** and press **ctrl+o** at the
-  key step to do the same inline ("Log in with OAuth"). The minted key is saved to
-  the provider profile and used normally.
-- **xAI (Grok) — opt-in preset** — xAI's flow needs an OAuth `client_id`. Zero
-  ships a built-in preset for the public Grok-CLI client, but to keep third-party
-  client identities out of the default credential path it is **off by default**.
-  Enable it with `export ZERO_OAUTH_ALLOW_PRESETS=1`, then `zero auth login xai`
-  (browser, or `--device` for headless) works one-click; the token is used directly
-  on `api.x.ai/v1`. Without the opt-in, set `ZERO_OAUTH_XAI_CLIENT_ID` (and
-  endpoints, or an issuer) yourself via `ZERO_OAUTH_XAI_*`. Either way the preset is
-  fully overridable by `ZERO_OAUTH_XAI_*` (env wins), and it requires a
-  SuperGrok / X Premium+ subscription; the client_id is an undocumented public
-  Grok-CLI client that may change without notice.
-- **ChatGPT (Codex) — opt-in preset** — `zero auth chatgpt` opens a browser, you
-  approve with your ChatGPT Plus/Pro/Business/Enterprise account, and the bearer is
-  stored. The bearer routes to `https://chatgpt.com/backend-api/codex/responses`
-  (the same endpoint the openai/codex CLI uses), with `originator: codex_cli_rs` and
-  the `chatgpt-account-id` claim injected as headers on every request. The
-  `chatgpt-account-id` is extracted from the OIDC ID token and stored alongside the
-  bearer; if the claim is missing (older ChatGPT accounts, or a rotated
-  authorization server), the Codex backend will 401 and `zero auth status chatgpt`
-  will show the warning. Like xAI, the preset uses the publicly-shipped Codex CLI
-  client identity (`app_EMoamEEZ73f0CkXaXp7hrann`) and is opt-in via
-  `ZERO_OAUTH_ALLOW_PRESETS=1`. As of mid-2026 the Codex backend is
-  Cloudflare-gated: requests from a non-Codex client can still be challenged, and
-  the `chatgpt-proxy` route in §2 is the conservative fallback.
-- **Hugging Face — opt-in preset, BYO client_id** — `zero auth login huggingface`
-  (or `--device` for headless) opens a Hugging Face OAuth flow. The bearer works on
-  the OpenAI-compatible router at `https://router.huggingface.co/v1` for hundreds
-  of OSS models (Llama, Qwen, DeepSeek, Mistral, etc.). HF does not ship a
-  globally-known client_id, so the preset ships endpoints + scopes + the OIDC
-  issuer pre-filled; you must register a "public" OAuth app (no secret) at
-  <https://huggingface.co/settings/applications/new> and set the resulting
-  `client_id` via `ZERO_OAUTH_HUGGINGFACE_CLIENT_ID`. Enable the preset with
-  `ZERO_OAUTH_ALLOW_PRESETS=1` (or omit it — the BYO client_id path uses
-  `client_credentials = none` and doesn't need the opt-in). Free tier has strict
-  rate limits; Pro removes them.
-
-Any field of a preset is overridable via `ZERO_OAUTH_<NAME>_*`. For a fully custom
-OAuth/OIDC provider, set those env vars (see `zero auth --help`) and
-`zero auth login <name>`.
-
----
-
-## 2. ChatGPT / Claude subscriptions — why a proxy is required
-
-We researched this carefully. As of mid-2026, a **subscription** OAuth token does
-**not** work as a drop-in bearer against the standard APIs:
-
-- **OpenAI (ChatGPT):** a "Sign in with ChatGPT" token only works against
-  ChatGPT's own backend (`chatgpt.com/backend-api/codex/responses`, the Responses
-  API), **not** `api.openai.com`. That backend is **Cloudflare bot-protected** —
-  non-browser / headless clients get `cf-mitigated: challenge` → `403`. It also
-  requires mimicking the official Codex client (originator + account-id header).
-  **First-class path (this version):** `zero auth chatgpt` does exactly that
-  mimicking (`originator: codex_cli_rs`, `chatgpt-account-id: <claim>`) and
-  routes requests to the Codex backend, no proxy required — see §1. The
-  `chatgpt-proxy` route below is the conservative fallback when Cloudflare
-  challenges become an issue, and is the only path that works without a
-  browser-based ChatGPT OAuth login.
-- **Anthropic (Claude):** the Messages API **rejects** subscription OAuth tokens
-  for third-party use unless the request spoofs the Claude Code identity
-  (`anthropic-beta: oauth-2025-04-20`, `claude-cli` UA, and a verbatim
-  *"You are Claude Code…"* system prompt) — and **even then** tool-using requests
-  on Max plans are routed to a disabled billing lane and `400`. Anthropic's policy
-  **prohibits** subscription-token use outside Claude Code / claude.ai, and the
-  timeline hardened through 2026: a **Feb 19 2026** docs update spelled out that
-  Free/Pro/Max OAuth tokens may not be used in third-party tools or the Agent SDK,
-  then on **April 4 2026** enforcement landed and subscription OAuth tokens
-  **stopped working in third-party harnesses** (starting with OpenClaw, then the
-  rest). As of mid-2026 the only supported ways to drive Claude from a third-party
-  tool are a standard **API key** or pay-as-you-go **"Extra Usage"** billing —
-  both per-token, not the flat subscription. The request to allow subscription use
-  (claude-code #37205) was closed *"not planned."*
-
-So Zero does **not** call those backends directly or spoof those clients — that
-would be fragile, account-risky, and (for Anthropic) against the vendor's terms.
-The robust, supported pattern is a **local proxy** that holds your subscription
-session and exposes a clean OpenAI- or Anthropic-compatible endpoint on
-`127.0.0.1`. The proxy absorbs the Cloudflare / client-spoofing surface; Zero
-just points at it.
-
-### ChatGPT via a local proxy
-
-Run a local ChatGPT OAuth proxy that exposes an OpenAI-compatible endpoint (these
-typically listen on `127.0.0.1:10531/v1`). Then use the built-in **`chatgpt-proxy`**
-preset (no API key — the proxy authenticates):
-
-```jsonc
-// ~/.config/zero/config.json (or ./.zero/config.json)
-{
-  "activeProvider": "chatgpt",
-  "providers": [
-    {
-      "name": "chatgpt",
-      "catalogID": "chatgpt-proxy",     // OpenAI-compatible, local, no key
-      "baseURL": "http://localhost:10531/v1", // override for your proxy's port
-      "model": "gpt-5.5"                 // whatever model your proxy serves
-    }
-  ]
-}
-```
-
-```sh
-zero exec --prompt "say hi"   # routes through the proxy → your ChatGPT plan
-```
-
-### Claude via a local proxy
-
-There is no single canonical Claude OAuth-proxy port, so use the generic
-**`custom-anthropic-compatible`** entry pointed at your proxy's Anthropic-compatible
-endpoint:
-
-```jsonc
-{
-  "activeProvider": "claude",
-  "providers": [
-    {
-      "name": "claude",
-      "catalogID": "custom-anthropic-compatible",
-      "baseURL": "http://localhost:<port>",  // your Claude proxy
-      "apiKey": "unused-by-proxy",
-      "model": "claude-sonnet-4.5"
-    }
-  ]
-}
-```
-
----
-
-## 3. Supported alternatives (no proxy)
-
-- **API key (recommended, simplest):** set `OPENAI_API_KEY` / `ANTHROPIC_API_KEY`
-  (or per-profile `apiKey`) and use the `openai` / `anthropic` catalog providers.
-  Bills as API usage.
-- **Anthropic subscription automation, sanctioned:** spawn the real `claude` CLI
-  (e.g. `claude -p …`) as a subprocess — the only path Anthropic recognizes as a
-  first-class subscription session.
-
----
-
-## Notes
-
-- The `chatgpt-proxy` base URL / port and model are defaults you override for your
-  setup; they are not an endorsement of any specific proxy implementation.
-- Subscription-via-proxy depends on third-party tools and undocumented vendor
-  backends; it can break without notice. The API-key path is the stable one.

go.mod

@@ -1,38 +1,45 @@
 module github.com/Gitlawb/zero
 
-go 1.26.5
+go 1.25
 
 require (
-	charm.land/bubbles/v2 v2.1.0
-	charm.land/bubbletea/v2 v2.0.7
-	charm.land/lipgloss/v2 v2.0.4
-	github.com/alecthomas/chroma/v2 v2.27.0
-	github.com/atotto/clipboard v0.1.4
-	github.com/aymanbagabas/go-udiff v0.4.1
-	github.com/charmbracelet/colorprofile v0.4.3
-	github.com/charmbracelet/x/ansi v0.11.7
-	github.com/charmbracelet/x/term v0.2.2
-	github.com/coder/websocket v1.8.15
-	github.com/ledongthuc/pdf v0.0.0-20250511090121-5959a4027728
-	golang.org/x/sys v0.46.0
-	mvdan.cc/sh/v3 v3.13.1
+	github.com/charmbracelet/bubbles v1.0.0
+	github.com/charmbracelet/bubbletea v1.3.10
+	github.com/charmbracelet/glamour v1.0.0
+	github.com/charmbracelet/lipgloss v1.1.1-0.20250404203927-76690c660834
+	github.com/charmbracelet/x/ansi v0.11.6
+	github.com/muesli/termenv v0.16.0
+	golang.org/x/sys v0.38.0
 )
 
 require (
-	github.com/charmbracelet/ultraviolet v0.0.0-20260615092913-2399af76d5b1 // indirect
-	github.com/charmbracelet/x/exp/golden v0.0.0-20260615092313-b57e5e6d29bb // indirect
-	github.com/charmbracelet/x/termios v0.1.1 // indirect
-	github.com/charmbracelet/x/windows v0.2.2 // indirect
-	github.com/clipperhouse/displaywidth v0.11.0 // indirect
-	github.com/clipperhouse/uax29/v2 v2.7.0 // indirect
-	github.com/dlclark/regexp2/v2 v2.2.1 // indirect
-	github.com/go-quicktest/qt v1.102.0 // indirect
-	github.com/lucasb-eyer/go-colorful v1.4.0 // indirect
-	github.com/mattn/go-runewidth v0.0.24 // indirect
+	github.com/alecthomas/chroma/v2 v2.20.0 // indirect
+	github.com/atotto/clipboard v0.1.4 // indirect
+	github.com/aymanbagabas/go-osc52/v2 v2.0.1 // indirect
+	github.com/aymerick/douceur v0.2.0 // indirect
+	github.com/charmbracelet/colorprofile v0.4.1 // indirect
+	github.com/charmbracelet/x/cellbuf v0.0.15 // indirect
+	github.com/charmbracelet/x/exp/slice v0.0.0-20250327172914-2fdc97757edf // indirect
+	github.com/charmbracelet/x/term v0.2.2 // indirect
+	github.com/clipperhouse/displaywidth v0.9.0 // indirect
+	github.com/clipperhouse/stringish v0.1.1 // indirect
+	github.com/clipperhouse/uax29/v2 v2.5.0 // indirect
+	github.com/dlclark/regexp2 v1.11.5 // indirect
+	github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f // indirect
+	github.com/gorilla/css v1.0.1 // indirect
+	github.com/lucasb-eyer/go-colorful v1.3.0 // indirect
+	github.com/mattn/go-isatty v0.0.20 // indirect
+	github.com/mattn/go-localereader v0.0.1 // indirect
+	github.com/mattn/go-runewidth v0.0.19 // indirect
+	github.com/microcosm-cc/bluemonday v1.0.27 // indirect
+	github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6 // indirect
 	github.com/muesli/cancelreader v0.2.2 // indirect
+	github.com/muesli/reflow v0.3.0 // indirect
 	github.com/rivo/uniseg v0.4.7 // indirect
-	github.com/rogpeppe/go-internal v1.15.0 // indirect
 	github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e // indirect
-	golang.org/x/exp v0.0.0-20260611194520-c48552f49976 // indirect
-	golang.org/x/sync v0.21.0 // indirect
+	github.com/yuin/goldmark v1.7.13 // indirect
+	github.com/yuin/goldmark-emoji v1.0.6 // indirect
+	golang.org/x/net v0.38.0 // indirect
+	golang.org/x/term v0.36.0 // indirect
+	golang.org/x/text v0.30.0 // indirect
 )

go.sum

@@ -1,70 +1,89 @@
-charm.land/bubbles/v2 v2.1.0 h1:YSnNh5cPYlYjPxRrzs5VEn3vwhtEn3jVGRBT3M7/I0g=
-charm.land/bubbles/v2 v2.1.0/go.mod h1:l97h4hym2hvWBVfmJDtrEHHCtkIKeTEb3TTJ4ZOB3wY=
-charm.land/bubbletea/v2 v2.0.7 h1:7qw2tTAVar7m7klOPBYfTB0mniv/RuexsYwMRNxSeL0=
-charm.land/bubbletea/v2 v2.0.7/go.mod h1:DGW2q8gvzHnOpMpZTORs0aySVHCox5C+2Svk0fci1qs=
-charm.land/lipgloss/v2 v2.0.4 h1:lcPeVtcp23SNra7lHy8iYE4UC2aIipVQ47sbGyyxR5Q=
-charm.land/lipgloss/v2 v2.0.4/go.mod h1:0653x8epbZSzdDfO/XPS1a/uYPOBeSsCssOpJOqDzik=
 github.com/alecthomas/assert/v2 v2.11.0 h1:2Q9r3ki8+JYXvGsDyBXwH3LcJ+WK5D0gc5E8vS6K3D0=
 github.com/alecthomas/assert/v2 v2.11.0/go.mod h1:Bze95FyfUr7x34QZrjL+XP+0qgp/zg8yS+TtBj1WA3k=
-github.com/alecthomas/chroma/v2 v2.27.0 h1:FodwmyOBgJULFYmDqibcp9pvfDLWdtPRh9v/r5BXYZs=
-github.com/alecthomas/chroma/v2 v2.27.0/go.mod h1:NjJ3ciIgrqBNeIkWZ4e46nseoLDslxU1LmfCoL+wcY8=
-github.com/alecthomas/repr v0.5.2 h1:SU73FTI9D1P5UNtvseffFSGmdNci/O6RsqzeXJtP0Qs=
-github.com/alecthomas/repr v0.5.2/go.mod h1:Fr0507jx4eOXV7AlPV6AVZLYrLIuIeSOWtW57eE/O/4=
+github.com/alecthomas/chroma/v2 v2.20.0 h1:sfIHpxPyR07/Oylvmcai3X/exDlE8+FA820NTz+9sGw=
+github.com/alecthomas/chroma/v2 v2.20.0/go.mod h1:e7tViK0xh/Nf4BYHl00ycY6rV7b8iXBksI9E359yNmA=
+github.com/alecthomas/repr v0.5.1 h1:E3G4t2QbHTSNpPKBgMTln5KLkZHLOcU7r37J4pXBuIg=
+github.com/alecthomas/repr v0.5.1/go.mod h1:Fr0507jx4eOXV7AlPV6AVZLYrLIuIeSOWtW57eE/O/4=
 github.com/atotto/clipboard v0.1.4 h1:EH0zSVneZPSuFR11BlR9YppQTVDbh5+16AmcJi4g1z4=
 github.com/atotto/clipboard v0.1.4/go.mod h1:ZY9tmq7sm5xIbd9bOK4onWV4S6X0u6GY7Vn0Yu86PYI=
-github.com/aymanbagabas/go-udiff v0.4.1 h1:OEIrQ8maEeDBXQDoGCbbTTXYJMYRCRO1fnodZ12Gv5o=
-github.com/aymanbagabas/go-udiff v0.4.1/go.mod h1:0L9PGwj20lrtmEMeyw4WKJ/TMyDtvAoK9bf2u/mNo3w=
-github.com/charmbracelet/colorprofile v0.4.3 h1:QPa1IWkYI+AOB+fE+mg/5/4HRMZcaXex9t5KX76i20Q=
-github.com/charmbracelet/colorprofile v0.4.3/go.mod h1:/zT4BhpD5aGFpqQQqw7a+VtHCzu+zrQtt1zhMt9mR4Q=
-github.com/charmbracelet/ultraviolet v0.0.0-20260615092913-2399af76d5b1 h1:4+r3uOJ69ueRBt4okgEfWZeXs3BD36HcDBmOIAUlETk=
-github.com/charmbracelet/ultraviolet v0.0.0-20260615092913-2399af76d5b1/go.mod h1:f/jRa757WUmaOZrbPspXymbg/GnbF+rwe4OLsG7aXYo=
-github.com/charmbracelet/x/ansi v0.11.7 h1:kzv1kJvjg2S3r9KHo8hDdHFQLEqn4RBCb39dAYC84jI=
-github.com/charmbracelet/x/ansi v0.11.7/go.mod h1:9qGpnAVYz+8ACONkZBUWPtL7lulP9No6p1epAihUZwQ=
-github.com/charmbracelet/x/exp/golden v0.0.0-20260615092313-b57e5e6d29bb h1:m1Uub9uM7meZ8ofVyeRfK7M/knCi3sF2gKn9kldVW0I=
-github.com/charmbracelet/x/exp/golden v0.0.0-20260615092313-b57e5e6d29bb/go.mod h1:6fMpcW6iwN/kX+xJ52eqVWsDiBTe0UJD24JLoHFe+P0=
+github.com/aymanbagabas/go-osc52/v2 v2.0.1 h1:HwpRHbFMcZLEVr42D4p7XBqjyuxQH5SMiErDT4WkJ2k=
+github.com/aymanbagabas/go-osc52/v2 v2.0.1/go.mod h1:uYgXzlJ7ZpABp8OJ+exZzJJhRNQ2ASbcXHWsFqH8hp8=
+github.com/aymanbagabas/go-udiff v0.3.1 h1:LV+qyBQ2pqe0u42ZsUEtPiCaUoqgA9gYRDs3vj1nolY=
+github.com/aymanbagabas/go-udiff v0.3.1/go.mod h1:G0fsKmG+P6ylD0r6N/KgQD/nWzgfnl8ZBcNLgcbrw8E=
+github.com/aymerick/douceur v0.2.0 h1:Mv+mAeH1Q+n9Fr+oyamOlAkUNPWPlA8PPGR0QAaYuPk=
+github.com/aymerick/douceur v0.2.0/go.mod h1:wlT5vV2O3h55X9m7iVYN0TBM0NH/MmbLnd30/FjWUq4=
+github.com/charmbracelet/bubbles v1.0.0 h1:12J8/ak/uCZEMQ6KU7pcfwceyjLlWsDLAxB5fXonfvc=
+github.com/charmbracelet/bubbles v1.0.0/go.mod h1:9d/Zd5GdnauMI5ivUIVisuEm3ave1XwXtD1ckyV6r3E=
+github.com/charmbracelet/bubbletea v1.3.10 h1:otUDHWMMzQSB0Pkc87rm691KZ3SWa4KUlvF9nRvCICw=
+github.com/charmbracelet/bubbletea v1.3.10/go.mod h1:ORQfo0fk8U+po9VaNvnV95UPWA1BitP1E0N6xJPlHr4=
+github.com/charmbracelet/colorprofile v0.4.1 h1:a1lO03qTrSIRaK8c3JRxJDZOvhvIeSco3ej+ngLk1kk=
+github.com/charmbracelet/colorprofile v0.4.1/go.mod h1:U1d9Dljmdf9DLegaJ0nGZNJvoXAhayhmidOdcBwAvKk=
+github.com/charmbracelet/glamour v1.0.0 h1:AWMLOVFHTsysl4WV8T8QgkQ0s/ZNZo7CiE4WKhk8l08=
+github.com/charmbracelet/glamour v1.0.0/go.mod h1:DSdohgOBkMr2ZQNhw4LZxSGpx3SvpeujNoXrQyH2hxo=
+github.com/charmbracelet/lipgloss v1.1.1-0.20250404203927-76690c660834 h1:ZR7e0ro+SZZiIZD7msJyA+NjkCNNavuiPBLgerbOziE=
+github.com/charmbracelet/lipgloss v1.1.1-0.20250404203927-76690c660834/go.mod h1:aKC/t2arECF6rNOnaKaVU6y4t4ZeHQzqfxedE/VkVhA=
+github.com/charmbracelet/x/ansi v0.11.6 h1:GhV21SiDz/45W9AnV2R61xZMRri5NlLnl6CVF7ihZW8=
+github.com/charmbracelet/x/ansi v0.11.6/go.mod h1:2JNYLgQUsyqaiLovhU2Rv/pb8r6ydXKS3NIttu3VGZQ=
+github.com/charmbracelet/x/cellbuf v0.0.15 h1:ur3pZy0o6z/R7EylET877CBxaiE1Sp1GMxoFPAIztPI=
+github.com/charmbracelet/x/cellbuf v0.0.15/go.mod h1:J1YVbR7MUuEGIFPCaaZ96KDl5NoS0DAWkskup+mOY+Q=
+github.com/charmbracelet/x/exp/golden v0.0.0-20241011142426-46044092ad91 h1:payRxjMjKgx2PaCWLZ4p3ro9y97+TVLZNaRZgJwSVDQ=
+github.com/charmbracelet/x/exp/golden v0.0.0-20241011142426-46044092ad91/go.mod h1:wDlXFlCrmJ8J+swcL/MnGUuYnqgQdW9rhSD61oNMb6U=
+github.com/charmbracelet/x/exp/slice v0.0.0-20250327172914-2fdc97757edf h1:rLG0Yb6MQSDKdB52aGX55JT1oi0P0Kuaj7wi1bLUpnI=
+github.com/charmbracelet/x/exp/slice v0.0.0-20250327172914-2fdc97757edf/go.mod h1:B3UgsnsBZS/eX42BlaNiJkD1pPOUa+oF1IYC6Yd2CEU=
 github.com/charmbracelet/x/term v0.2.2 h1:xVRT/S2ZcKdhhOuSP4t5cLi5o+JxklsoEObBSgfgZRk=
 github.com/charmbracelet/x/term v0.2.2/go.mod h1:kF8CY5RddLWrsgVwpw4kAa6TESp6EB5y3uxGLeCqzAI=
-github.com/charmbracelet/x/termios v0.1.1 h1:o3Q2bT8eqzGnGPOYheoYS8eEleT5ZVNYNy8JawjaNZY=
-github.com/charmbracelet/x/termios v0.1.1/go.mod h1:rB7fnv1TgOPOyyKRJ9o+AsTU/vK5WHJ2ivHeut/Pcwo=
-github.com/charmbracelet/x/windows v0.2.2 h1:IofanmuvaxnKHuV04sC0eBy/smG6kIKrWG2/jYn2GuM=
-github.com/charmbracelet/x/windows v0.2.2/go.mod h1:/8XtdKZzedat74NQFn0NGlGL4soHB0YQZrETF96h75k=
-github.com/clipperhouse/displaywidth v0.11.0 h1:lBc6kY44VFw+TDx4I8opi/EtL9m20WSEFgwIwO+UVM8=
-github.com/clipperhouse/displaywidth v0.11.0/go.mod h1:bkrFNkf81G8HyVqmKGxsPufD3JhNl3dSqnGhOoSD/o0=
-github.com/clipperhouse/uax29/v2 v2.7.0 h1:+gs4oBZ2gPfVrKPthwbMzWZDaAFPGYK72F0NJv2v7Vk=
-github.com/clipperhouse/uax29/v2 v2.7.0/go.mod h1:EFJ2TJMRUaplDxHKj1qAEhCtQPW2tJSwu5BF98AuoVM=
-github.com/coder/websocket v1.8.15 h1:6B2JPeOGlpff2Uz6vOEH1Vzpi0iUz20A+lPVhPHtNUA=
-github.com/coder/websocket v1.8.15/go.mod h1:NX3SzP+inril6yawo5CQXx8+fk145lPDC6pumgx0mVg=
-github.com/dlclark/regexp2/v2 v2.2.1 h1:mf4KkFUj0gJuarK8P+LgiS+Lit7m9N1yAwEfPbee7R0=
-github.com/dlclark/regexp2/v2 v2.2.1/go.mod h1:avUrQvPaLz2DrFNHJF0taWAFFX2C1GMSSoeiqFjcBmU=
-github.com/go-quicktest/qt v1.102.0 h1:HSQxCeh5YZH3EL3W39ixjtyaEhcWSXQHtHnMBzSs474=
-github.com/go-quicktest/qt v1.102.0/go.mod h1:p4lGIVX+8Wa6ZPNDvqcxq36XpUDLh42FLetFU7odllI=
-github.com/google/go-cmp v0.7.0 h1:wk8382ETsv4JYUZwIsn6YpYiWiBsYLSJiTsyBybVuN8=
-github.com/google/go-cmp v0.7.0/go.mod h1:pXiqmnSA92OHEEa9HXL2W4E7lf9JzCmGVUdgjX3N/iU=
+github.com/clipperhouse/displaywidth v0.9.0 h1:Qb4KOhYwRiN3viMv1v/3cTBlz3AcAZX3+y9OLhMtAtA=
+github.com/clipperhouse/displaywidth v0.9.0/go.mod h1:aCAAqTlh4GIVkhQnJpbL0T/WfcrJXHcj8C0yjYcjOZA=
+github.com/clipperhouse/stringish v0.1.1 h1:+NSqMOr3GR6k1FdRhhnXrLfztGzuG+VuFDfatpWHKCs=
+github.com/clipperhouse/stringish v0.1.1/go.mod h1:v/WhFtE1q0ovMta2+m+UbpZ+2/HEXNWYXQgCt4hdOzA=
+github.com/clipperhouse/uax29/v2 v2.5.0 h1:x7T0T4eTHDONxFJsL94uKNKPHrclyFI0lm7+w94cO8U=
+github.com/clipperhouse/uax29/v2 v2.5.0/go.mod h1:Wn1g7MK6OoeDT0vL+Q0SQLDz/KpfsVRgg6W7ihQeh4g=
+github.com/dlclark/regexp2 v1.11.5 h1:Q/sSnsKerHeCkc/jSTNq1oCm7KiVgUMZRDUoRu0JQZQ=
+github.com/dlclark/regexp2 v1.11.5/go.mod h1:DHkYz0B9wPfa6wondMfaivmHpzrQ3v9q8cnmRbL6yW8=
+github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f h1:Y/CXytFA4m6baUTXGLOoWe4PQhGxaX0KpnayAqC48p4=
+github.com/erikgeiser/coninput v0.0.0-20211004153227-1c3628e74d0f/go.mod h1:vw97MGsxSvLiUE2X8qFplwetxpGLQrlU1Q9AUEIzCaM=
+github.com/gorilla/css v1.0.1 h1:ntNaBIghp6JmvWnxbZKANoLyuXTPZ4cAMlo6RyhlbO8=
+github.com/gorilla/css v1.0.1/go.mod h1:BvnYkspnSzMmwRK+b8/xgNPLiIuNZr6vbZBTPQ2A3b0=
 github.com/hexops/gotextdiff v1.0.3 h1:gitA9+qJrrTCsiCl7+kh75nPqQt1cx4ZkudSTLoUqJM=
 github.com/hexops/gotextdiff v1.0.3/go.mod h1:pSWU5MAI3yDq+fZBTazCSJysOMbxWL1BSow5/V2vxeg=
-github.com/kr/pretty v0.3.1 h1:flRD4NNwYAUpkphVc1HcthR4KEIFJ65n8Mw5qdRn3LE=
-github.com/kr/pretty v0.3.1/go.mod h1:hoEshYVHaxMs3cyo3Yncou5ZscifuDolrwPKZanG3xk=
-github.com/kr/text v0.2.0 h1:5Nx0Ya0ZqY2ygV366QzturHI13Jq95ApcVaJBhpS+AY=
-github.com/kr/text v0.2.0/go.mod h1:eLer722TekiGuMkidMxC/pM04lWEeraHUUmBw8l2grE=
-github.com/ledongthuc/pdf v0.0.0-20250511090121-5959a4027728 h1:QwWKgMY28TAXaDl+ExRDqGQltzXqN/xypdKP86niVn8=
-github.com/ledongthuc/pdf v0.0.0-20250511090121-5959a4027728/go.mod h1:1fEHWurg7pvf5SG6XNE5Q8UZmOwex51Mkx3SLhrW5B4=
-github.com/lucasb-eyer/go-colorful v1.4.0 h1:UtrWVfLdarDgc44HcS7pYloGHJUjHV/4FwW4TvVgFr4=
-github.com/lucasb-eyer/go-colorful v1.4.0/go.mod h1:R4dSotOR9KMtayYi1e77YzuveK+i7ruzyGqttikkLy0=
-github.com/mattn/go-runewidth v0.0.24 h1:cpokDiIn0MGnhdHwuWnJBITySJ20QyNGnY2kR/ay2DU=
-github.com/mattn/go-runewidth v0.0.24/go.mod h1:XBkDxAl56ILZc9knddidhrOlY5R/pDhgLpndooCuJAs=
+github.com/lucasb-eyer/go-colorful v1.3.0 h1:2/yBRLdWBZKrf7gB40FoiKfAWYQ0lqNcbuQwVHXptag=
+github.com/lucasb-eyer/go-colorful v1.3.0/go.mod h1:R4dSotOR9KMtayYi1e77YzuveK+i7ruzyGqttikkLy0=
+github.com/mattn/go-isatty v0.0.20 h1:xfD0iDuEKnDkl03q4limB+vH+GxLEtL/jb4xVJSWWEY=
+github.com/mattn/go-isatty v0.0.20/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
+github.com/mattn/go-localereader v0.0.1 h1:ygSAOl7ZXTx4RdPYinUpg6W99U8jWvWi9Ye2JC/oIi4=
+github.com/mattn/go-localereader v0.0.1/go.mod h1:8fBrzywKY7BI3czFoHkuzRoWE9C+EiG4R1k4Cjx5p88=
+github.com/mattn/go-runewidth v0.0.12/go.mod h1:RAqKPSqVFrSLVXbA8x7dzmKdmGzieGRCM46jaSJTDAk=
+github.com/mattn/go-runewidth v0.0.19 h1:v++JhqYnZuu5jSKrk9RbgF5v4CGUjqRfBm05byFGLdw=
+github.com/mattn/go-runewidth v0.0.19/go.mod h1:XBkDxAl56ILZc9knddidhrOlY5R/pDhgLpndooCuJAs=
+github.com/microcosm-cc/bluemonday v1.0.27 h1:MpEUotklkwCSLeH+Qdx1VJgNqLlpY2KXwXFM08ygZfk=
+github.com/microcosm-cc/bluemonday v1.0.27/go.mod h1:jFi9vgW+H7c3V0lb6nR74Ib/DIB5OBs92Dimizgw2cA=
+github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6 h1:ZK8zHtRHOkbHy6Mmr5D264iyp3TiX5OmNcI5cIARiQI=
+github.com/muesli/ansi v0.0.0-20230316100256-276c6243b2f6/go.mod h1:CJlz5H+gyd6CUWT45Oy4q24RdLyn7Md9Vj2/ldJBSIo=
 github.com/muesli/cancelreader v0.2.2 h1:3I4Kt4BQjOR54NavqnDogx/MIoWBFa0StPA8ELUXHmA=
 github.com/muesli/cancelreader v0.2.2/go.mod h1:3XuTXfFS2VjM+HTLZY9Ak0l6eUKfijIfMUZ4EgX0QYo=
+github.com/muesli/reflow v0.3.0 h1:IFsN6K9NfGtjeggFP+68I4chLZV2yIKsXJFNZ+eWh6s=
+github.com/muesli/reflow v0.3.0/go.mod h1:pbwTDkVPibjO2kyvBQRBxTWEEGDGq0FlB1BIKtnHY/8=
+github.com/muesli/termenv v0.16.0 h1:S5AlUN9dENB57rsbnkPyfdGuWIlkmzJjbFf0Tf5FWUc=
+github.com/muesli/termenv v0.16.0/go.mod h1:ZRfOIKPFDYQoDFF4Olj7/QJbW60Ol/kL1pU3VfY/Cnk=
+github.com/rivo/uniseg v0.1.0/go.mod h1:J6wj4VEh+S6ZtnVlnTBMWIodfgj8LQOQFoIToxlJtxc=
+github.com/rivo/uniseg v0.2.0/go.mod h1:J6wj4VEh+S6ZtnVlnTBMWIodfgj8LQOQFoIToxlJtxc=
 github.com/rivo/uniseg v0.4.7 h1:WUdvkW8uEhrYfLC4ZzdpI2ztxP1I582+49Oc5Mq64VQ=
 github.com/rivo/uniseg v0.4.7/go.mod h1:FN3SvrM+Zdj16jyLfmOkMNblXMcoc8DfTHruCPUcx88=
-github.com/rogpeppe/go-internal v1.15.0 h1:D0RCU5rMAp+SpgkiNdrjfJ+LX4J1M32V2NeCY7EJ6hc=
-github.com/rogpeppe/go-internal v1.15.0/go.mod h1:DrUVZyrJU+txYW5/1kwtXQSMFio52ZOxX7yM1VHvnxs=
 github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e h1:JVG44RsyaB9T2KIHavMF/ppJZNG9ZpyihvCd0w101no=
 github.com/xo/terminfo v0.0.0-20220910002029-abceb7e1c41e/go.mod h1:RbqR21r5mrJuqunuUZ/Dhy/avygyECGrLceyNeo4LiM=
-golang.org/x/exp v0.0.0-20260611194520-c48552f49976 h1:X8Hz2ImujgbmetVuW+w2YkyZChE3cBpZi2P158rTG9M=
-golang.org/x/exp v0.0.0-20260611194520-c48552f49976/go.mod h1:vnf4pv9iKZXY58sQE1L86zmNWJ4159e1RkcWiLCkeEY=
-golang.org/x/sync v0.21.0 h1:HLII4xRRTtCRkxYp4HNFF0Js/Og6q2i++KXbg0gHCwM=
-golang.org/x/sync v0.21.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
-golang.org/x/sys v0.46.0 h1:noSf2Fq6F8DBgS+LysIkx7rIExoNHJsxOAtPp4rthXw=
-golang.org/x/sys v0.46.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
-mvdan.cc/sh/v3 v3.13.1 h1:DP3TfgZhDkT7lerUdnp6PTGKyxxzz6T+cOlY/xEvfWk=
-mvdan.cc/sh/v3 v3.13.1/go.mod h1:lXJ8SexMvEVcHCoDvAGLZgFJ9Wsm2sulmoNEXGhYZD0=
+github.com/yuin/goldmark v1.7.13 h1:GPddIs617DnBLFFVJFgpo1aBfe/4xcvMc3SB5t/D0pA=
+github.com/yuin/goldmark v1.7.13/go.mod h1:ip/1k0VRfGynBgxOz0yCqHrbZXhcjxyuS66Brc7iBKg=
+github.com/yuin/goldmark-emoji v1.0.6 h1:QWfF2FYaXwL74tfGOW5izeiZepUDroDJfWubQI9HTHs=
+github.com/yuin/goldmark-emoji v1.0.6/go.mod h1:ukxJDKFpdFb5x0a5HqbdlcKtebh086iJpI31LTKmWuA=
+golang.org/x/exp v0.0.0-20231006140011-7918f672742d h1:jtJma62tbqLibJ5sFQz8bKtEM8rJBtfilJ2qTU199MI=
+golang.org/x/exp v0.0.0-20231006140011-7918f672742d/go.mod h1:ldy0pHrwJyGW56pPQzzkH36rKxoZW1tw7ZJpeKx+hdo=
+golang.org/x/net v0.38.0 h1:vRMAPTMaeGqVhG5QyLJHqNDwecKTomGeqbnfZyKlBI8=
+golang.org/x/net v0.38.0/go.mod h1:ivrbrMbzFq5J41QOQh0siUuly180yBYtLp+CKbEaFx8=
+golang.org/x/sys v0.0.0-20210809222454-d867a43fc93e/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.38.0 h1:3yZWxaJjBmCWXqhN1qh02AkOnCQ1poK6oF+a7xWL6Gc=
+golang.org/x/sys v0.38.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
+golang.org/x/term v0.36.0 h1:zMPR+aF8gfksFprF/Nc/rd1wRS1EI6nDBGyWAvDzx2Q=
+golang.org/x/term v0.36.0/go.mod h1:Qu394IJq6V6dCBRgwqshf3mPF85AqzYEzofzRdZkWss=
+golang.org/x/text v0.30.0 h1:yznKA/E9zq54KzlzBEAWn1NXSQ8DIp/NYMy88xJjl4k=
+golang.org/x/text v0.30.0/go.mod h1:yDdHFIX9t+tORqspjENWgzaCVXgk0yYnYuSZ8UzzBVM=

internal/acp/agent.go

@@ -1,594 +0,0 @@
-package acp
-
-import (
-	"context"
-	"encoding/base64"
-	"encoding/json"
-	"errors"
-	"log"
-	"strings"
-	"sync"
-
-	"github.com/Gitlawb/zero/internal/agent"
-	"github.com/Gitlawb/zero/internal/config"
-	"github.com/Gitlawb/zero/internal/sandbox"
-	"github.com/Gitlawb/zero/internal/sessions"
-	"github.com/Gitlawb/zero/internal/tools"
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-// Deps are the ZERO capabilities the ACP Agent drives. The CLI fills these with
-// real implementations; tests inject fakes (e.g. a canned provider) to drive the
-// full ACP flow without a live model. Keeping auth/model/keys behind these deps
-// means the editor only hosts the thread — ZERO owns BYOK and telemetry-free
-// operation.
-type Deps struct {
-	ResolveConfig func(workspaceRoot string, overrides config.Overrides) (config.ResolvedConfig, error)
-	NewProvider   func(profile config.ProviderProfile) (zeroruntime.Provider, error)
-	RunAgent      func(ctx context.Context, prompt string, provider zeroruntime.Provider, opts agent.Options) (agent.Result, error)
-	// BuildWorkspace builds the SCOPED tool registry and the sandbox engine for a
-	// validated workspace root, so ACP shell tools (bash/exec_command) are confined
-	// exactly like the exec surface — never run unconfined on the host.
-	BuildWorkspace func(workspaceRoot string, resolved config.ResolvedConfig) (*tools.Registry, *sandbox.Engine, error)
-	// ResolveWorkspaceRoot validates + normalizes a client-supplied cwd (must be an
-	// existing directory; never the bare root). It is the file-tool confinement root.
-	ResolveWorkspaceRoot func(cwd string) (string, error)
-	Store                *sessions.Store
-	AgentInfo            Implementation
-}
-
-// Agent is the ACP agent server bound to one JSON-RPC connection (one editor).
-type Agent struct {
-	conn *Conn
-	deps Deps
-
-	mu          sync.Mutex
-	clientCaps  ClientCapabilities
-	initialized bool
-	sessions    map[string]*acpSession
-}
-
-type turnRecord struct {
-	user      string
-	assistant string
-}
-
-type acpSession struct {
-	id  string
-	cwd string
-
-	// turnMu serializes prompt turns for one session: concurrent session/prompt
-	// calls run one at a time so they can't interleave history or clobber the
-	// single cancel slot.
-	turnMu sync.Mutex
-
-	mu      sync.Mutex
-	mode    agent.PermissionMode
-	model   string // override; "" => config default
-	cancel  context.CancelFunc
-	history []turnRecord
-}
-
-// NewAgent builds the ACP server and registers its method handlers on conn.
-func NewAgent(conn *Conn, deps Deps) *Agent {
-	a := &Agent{conn: conn, deps: deps, sessions: make(map[string]*acpSession)}
-	conn.Handle(MethodInitialize, a.handleInitialize)
-	conn.Handle(MethodSessionNew, a.handleSessionNew)
-	conn.Handle(MethodSessionLoad, a.handleSessionLoad)
-	conn.Handle(MethodSessionPrompt, a.handleSessionPrompt)
-	conn.Handle(MethodSessionSetMode, a.handleSetMode)
-	conn.Handle(MethodSessionSetConfigOption, a.handleSetConfigOption)
-	conn.Handle(MethodZeroSetModel, a.handleZeroSetModel)
-	conn.HandleNotify(MethodSessionCancel, a.handleCancel)
-	return a
-}
-
-// Serve runs the connection read loop until the stream closes or ctx is done.
-func (a *Agent) Serve(ctx context.Context) error { return a.conn.Serve(ctx) }
-
-// ---- initialize ----
-
-func (a *Agent) handleInitialize(_ context.Context, params json.RawMessage) (any, error) {
-	var p InitializeParams
-	if len(params) > 0 {
-		_ = json.Unmarshal(params, &p)
-	}
-	negotiated := ProtocolVersion
-	if p.ProtocolVersion > 0 && p.ProtocolVersion < ProtocolVersion {
-		negotiated = p.ProtocolVersion
-	}
-	a.mu.Lock()
-	a.clientCaps = p.ClientCapabilities
-	a.initialized = true
-	a.mu.Unlock()
-
-	info := a.deps.AgentInfo
-	return InitializeResult{
-		ProtocolVersion: negotiated,
-		AgentCapabilities: AgentCapabilities{
-			// Only advertise what ZERO actually implements: session/load (loadSession)
-			// and image prompts. session/resume + the session-capability sub-object
-			// are intentionally omitted since there is no resume handler yet.
-			LoadSession:        true,
-			PromptCapabilities: PromptCapabilities{Image: true},
-		},
-		AgentInfo: &info,
-		// ZERO owns credentials (BYOK) and does not delegate auth to the editor.
-		AuthMethods: []AuthMethod{},
-	}, nil
-}
-
-// ---- session lifecycle ----
-
-func (a *Agent) handleSessionNew(_ context.Context, params json.RawMessage) (any, error) {
-	var p NewSessionParams
-	if err := json.Unmarshal(params, &p); err != nil {
-		return nil, RPCError(codeInvalidParams, "invalid session/new params")
-	}
-	root, err := a.deps.ResolveWorkspaceRoot(p.Cwd)
-	if err != nil {
-		return nil, RPCError(codeInvalidParams, err.Error())
-	}
-	meta, err := a.deps.Store.Create(sessions.CreateInput{Title: "ACP session", Cwd: root})
-	if err != nil {
-		return nil, RPCError(codeInternalError, "create session: "+err.Error())
-	}
-	sess := a.registerSession(meta.SessionID, root, nil)
-	return NewSessionResult{
-		SessionID: sess.id,
-		Modes:     a.modeState(sess),
-	}, nil
-}
-
-func (a *Agent) handleSessionLoad(_ context.Context, params json.RawMessage) (any, error) {
-	var p LoadSessionParams
-	if err := json.Unmarshal(params, &p); err != nil {
-		return nil, RPCError(codeInvalidParams, "invalid session/load params")
-	}
-	meta, err := a.deps.Store.Get(p.SessionID)
-	if err != nil || meta == nil {
-		return nil, RPCError(codeInvalidParams, "session not found: "+p.SessionID)
-	}
-	cwdInput := p.Cwd
-	if strings.TrimSpace(cwdInput) == "" {
-		cwdInput = meta.Cwd
-	}
-	root, err := a.deps.ResolveWorkspaceRoot(cwdInput)
-	if err != nil {
-		return nil, RPCError(codeInvalidParams, err.Error())
-	}
-	// Load history BEFORE publishing the session so no concurrent prompt observes
-	// a half-initialized session (registerSession sets history under the lock and
-	// reuses an already-live session rather than orphaning its in-flight turn).
-	history, historyErr := a.loadHistory(meta.SessionID)
-	sess := a.registerSession(meta.SessionID, root, history)
-	a.warnPersistence(
-		&notifier{conn: a.conn, sessionID: sess.id},
-		"load session history",
-		"Could not load session history. The session is open, but earlier turns may be missing until storage recovers.",
-		historyErr,
-	)
-	return LoadSessionResult{
-		Modes: a.modeState(sess),
-	}, nil
-}
-
-// ---- prompt turn ----
-
-func (a *Agent) handleSessionPrompt(ctx context.Context, params json.RawMessage) (any, error) {
-	var p PromptParams
-	if err := json.Unmarshal(params, &p); err != nil {
-		return nil, RPCError(codeInvalidParams, "invalid session/prompt params")
-	}
-	sess := a.session(p.SessionID)
-	if sess == nil {
-		return nil, RPCError(codeInvalidParams, "unknown session: "+p.SessionID)
-	}
-
-	// Serialize turns for this session so two prompts can't interleave history or
-	// fight over the single cancel slot. session/cancel still works concurrently
-	// (it doesn't take turnMu).
-	sess.turnMu.Lock()
-	defer sess.turnMu.Unlock()
-
-	userText := promptText(p.Prompt)
-	images := promptImages(p.Prompt)
-
-	turnCtx, cancel := context.WithCancel(ctx)
-	sess.setCancel(cancel)
-	defer func() {
-		cancel()
-		sess.setCancel(nil)
-	}()
-
-	reason, err := a.runTurn(turnCtx, sess, userText, images)
-	if err != nil {
-		return nil, err
-	}
-	return PromptResult{StopReason: reason}, nil
-}
-
-func (a *Agent) runTurn(ctx context.Context, sess *acpSession, userText string, images []zeroruntime.ImageBlock) (string, error) {
-	overrides := config.Overrides{}
-	if model := sess.currentModel(); model != "" {
-		overrides.Provider.Model = model
-	}
-	resolved, err := a.deps.ResolveConfig(sess.cwd, overrides)
-	if err != nil {
-		return "", RPCError(codeInternalError, "config: "+err.Error())
-	}
-	provider, err := a.deps.NewProvider(resolved.Provider)
-	if err != nil {
-		return "", RPCError(codeInternalError, "provider: "+err.Error())
-	}
-	// Build the SCOPED registry + sandbox engine for this session's workspace so
-	// shell/file tools are confined to the workspace exactly like the exec surface.
-	registry, sandboxEngine, err := a.deps.BuildWorkspace(sess.cwd, resolved)
-	if err != nil {
-		return "", RPCError(codeInternalError, "workspace: "+err.Error())
-	}
-	note := &notifier{conn: a.conn, sessionID: sess.id}
-
-	opts := agent.Options{
-		Cwd:            sess.cwd,
-		SessionID:      sess.id,
-		ProviderName:   resolved.Provider.Name,
-		Model:          resolved.Provider.Model,
-		Registry:       registry,
-		Sandbox:        sandboxEngine,
-		PermissionMode: sess.currentMode(),
-		MaxTurns:       resolved.MaxTurns,
-		Images:         images,
-		OnText:         note.text,
-		OnReasoning:    note.thought,
-		OnToolCall:     note.toolCall,
-		OnToolResult: func(result agent.ToolResult) {
-			note.toolResult(result)
-			if result.Name == "update_plan" {
-				a.emitPlan(registry, note)
-			}
-		},
-		OnPermissionRequest: func(ctx context.Context, req agent.PermissionRequest) (agent.PermissionDecision, error) {
-			return a.requestPermission(ctx, sess.id, req)
-		},
-	}
-
-	agentPrompt := buildPrompt(sess.snapshotHistory(), userText)
-	result, runErr := a.deps.RunAgent(ctx, agentPrompt, provider, opts)
-
-	reason, stopErr := stopReasonFor(result, runErr)
-	if stopErr != nil {
-		return "", RPCError(codeInternalError, stopErr.Error())
-	}
-	if err := a.persistTurn(sess, userText, result.FinalAnswer); err != nil {
-		a.warnPersistence(
-			note,
-			"save session history",
-			"Could not save session history. This turn is available in memory, but future resume may miss it until storage recovers.",
-			err,
-		)
-	}
-	return reason, nil
-}
-
-func stopReasonFor(result agent.Result, err error) (string, error) {
-	if err != nil {
-		if errors.Is(err, context.Canceled) {
-			return StopCancelled, nil
-		}
-		return "", err
-	}
-	if result.FinishReason == "length" {
-		return StopMaxTokens, nil
-	}
-	if result.FinishReason == "content_filter" {
-		return StopRefusal, nil
-	}
-	return StopEndTurn, nil
-}
-
-// requestPermission forwards a ZERO permission prompt to the client as an ACP
-// session/request_permission request and maps the outcome back. Failure to reach
-// the client fails closed to deny.
-func (a *Agent) requestPermission(ctx context.Context, sessionID string, req agent.PermissionRequest) (agent.PermissionDecision, error) {
-	params := RequestPermissionParams{
-		SessionID: sessionID,
-		ToolCall:  permissionToolCall(req),
-		Options:   buildPermissionOptions(req),
-	}
-	var result RequestPermissionResult
-	if err := a.conn.Call(ctx, MethodSessionRequestPerm, params, &result); err != nil {
-		if errors.Is(err, context.Canceled) {
-			return agent.PermissionDecision{Action: agent.PermissionDecisionCancel, Reason: "cancelled"}, nil
-		}
-		return agent.PermissionDecision{Action: agent.PermissionDecisionDeny, Reason: "permission request failed: " + err.Error()}, nil
-	}
-	return decisionFromOutcome(result.Outcome, req.AvailableDecisions), nil
-}
-
-func (a *Agent) emitPlan(registry *tools.Registry, note *notifier) {
-	t, ok := registry.Get("update_plan")
-	if !ok {
-		return
-	}
-	planner, ok := t.(interface{ CurrentPlan() []tools.PlanItem })
-	if !ok {
-		return
-	}
-	note.plan(planner.CurrentPlan())
-}
-
-// ---- mode + model selection ----
-
-func (a *Agent) handleSetMode(_ context.Context, params json.RawMessage) (any, error) {
-	var p SetSessionModeParams
-	if err := json.Unmarshal(params, &p); err != nil {
-		return nil, RPCError(codeInvalidParams, "invalid set_mode params")
-	}
-	sess := a.session(p.SessionID)
-	if sess == nil {
-		return nil, RPCError(codeInvalidParams, "unknown session: "+p.SessionID)
-	}
-	mode := agent.PermissionMode(p.ModeID)
-	switch mode {
-	case agent.PermissionModeAuto, agent.PermissionModeAsk:
-		sess.setMode(mode)
-		(&notifier{conn: a.conn, sessionID: sess.id}).currentMode(string(mode))
-		return SetSessionModeResult{}, nil
-	case agent.PermissionModeUnsafe:
-		// Unsafe = run every tool with no prompt. The TUI gates this behind an
-		// explicit --skip-permissions-unsafe operator flag; an editor client must
-		// not be able to grant itself unconfined, no-prompt access over the wire.
-		return nil, RPCError(codeInvalidParams, "mode not permitted over ACP: "+p.ModeID)
-	default:
-		return nil, RPCError(codeInvalidParams, "unknown mode: "+p.ModeID)
-	}
-}
-
-func (a *Agent) handleSetConfigOption(_ context.Context, params json.RawMessage) (any, error) {
-	var p SetSessionConfigOptionParams
-	if err := json.Unmarshal(params, &p); err != nil {
-		return nil, RPCError(codeInvalidParams, "invalid set_config_option params")
-	}
-	sess := a.session(p.SessionID)
-	if sess == nil {
-		return nil, RPCError(codeInvalidParams, "unknown session: "+p.SessionID)
-	}
-	if p.ConfigID != configIDModel {
-		return nil, RPCError(codeInvalidParams, "unknown config option: "+p.ConfigID)
-	}
-	sess.setModel(p.Value)
-	return SetSessionConfigOptionResult{}, nil
-}
-
-func (a *Agent) handleZeroSetModel(_ context.Context, params json.RawMessage) (any, error) {
-	var p ZeroSetModelParams
-	if err := json.Unmarshal(params, &p); err != nil {
-		return nil, RPCError(codeInvalidParams, "invalid _zero/set_model params")
-	}
-	sess := a.session(p.SessionID)
-	if sess == nil {
-		return nil, RPCError(codeInvalidParams, "unknown session: "+p.SessionID)
-	}
-	sess.setModel(p.Model)
-	return ZeroSetModelResult{Model: p.Model}, nil
-}
-
-func (a *Agent) handleCancel(_ context.Context, params json.RawMessage) {
-	var p CancelParams
-	if err := json.Unmarshal(params, &p); err != nil {
-		return
-	}
-	if sess := a.session(p.SessionID); sess != nil {
-		sess.invokeCancel()
-	}
-}
-
-// ---- advertising helpers ----
-
-func (a *Agent) modeState(s *acpSession) *SessionModeState {
-	// Only auto/ask are offered over ACP; Unsafe is gated to the operator (see
-	// handleSetMode) so a client can't grant itself no-prompt host access.
-	return &SessionModeState{
-		CurrentModeID: string(s.currentMode()),
-		AvailableModes: []SessionMode{
-			{ID: string(agent.PermissionModeAuto), Name: "Auto", Description: "Run safe tools automatically; ask before risky ones."},
-			{ID: string(agent.PermissionModeAsk), Name: "Ask", Description: "Ask before every tool that changes state."},
-		},
-	}
-}
-
-// ---- persistence + continuity ----
-
-func (a *Agent) persistTurn(sess *acpSession, user, assistant string) error {
-	defer sess.appendHistory(turnRecord{user: user, assistant: assistant})
-	if a.deps.Store == nil {
-		return nil
-	}
-	events := []sessions.AppendEventInput{
-		{
-			Type:    sessions.EventMessage,
-			Payload: map[string]any{"role": "user", "content": user},
-		},
-	}
-	if assistant != "" {
-		events = append(events, sessions.AppendEventInput{
-			Type:    sessions.EventMessage,
-			Payload: map[string]any{"role": "assistant", "content": assistant},
-		})
-	}
-	_, err := a.deps.Store.AppendEvents(sess.id, events)
-	return err
-}
-
-func (a *Agent) loadHistory(sessionID string) ([]turnRecord, error) {
-	if a.deps.Store == nil {
-		return nil, nil
-	}
-	events, err := a.deps.Store.ReadEvents(sessionID)
-	if err != nil {
-		return nil, err
-	}
-	var records []turnRecord
-	var pendingUser string
-	havePending := false
-	for _, e := range events {
-		if e.Type != sessions.EventMessage {
-			continue
-		}
-		raw, err := json.Marshal(e.Payload)
-		if err != nil {
-			continue
-		}
-		var msg struct {
-			Role    string `json:"role"`
-			Content string `json:"content"`
-		}
-		if json.Unmarshal(raw, &msg) != nil {
-			continue
-		}
-		switch msg.Role {
-		case "user":
-			if havePending {
-				records = append(records, turnRecord{user: pendingUser})
-			}
-			pendingUser = msg.Content
-			havePending = true
-		case "assistant":
-			records = append(records, turnRecord{user: pendingUser, assistant: msg.Content})
-			pendingUser = ""
-			havePending = false
-		}
-	}
-	if havePending {
-		records = append(records, turnRecord{user: pendingUser})
-	}
-	return records, nil
-}
-
-func (a *Agent) warnPersistence(note *notifier, action string, message string, err error) {
-	if err == nil {
-		return
-	}
-	sessionID := ""
-	if note != nil {
-		sessionID = note.sessionID
-	}
-	log.Printf("zero acp: failed to %s for session %s: %v", action, sessionID, err)
-	if note != nil {
-		note.text("\n\n[zero warning] " + message + "\n")
-	}
-}
-
-// buildPrompt prepends prior conversation as context, since agent.Run drives a
-// single seeded turn. Mirrors how headless resume folds history into the prompt.
-func buildPrompt(history []turnRecord, userText string) string {
-	if len(history) == 0 {
-		return userText
-	}
-	var b strings.Builder
-	b.WriteString("Conversation so far:\n")
-	for _, t := range history {
-		b.WriteString("User: ")
-		b.WriteString(t.user)
-		b.WriteString("\n")
-		if t.assistant != "" {
-			b.WriteString("Assistant: ")
-			b.WriteString(t.assistant)
-			b.WriteString("\n")
-		}
-	}
-	b.WriteString("\n---\nContinue with this request:\n")
-	b.WriteString(userText)
-	return b.String()
-}
-
-func promptImages(blocks []ContentBlock) []zeroruntime.ImageBlock {
-	var images []zeroruntime.ImageBlock
-	for _, blk := range blocks {
-		if blk.Type != "image" || blk.Data == "" {
-			continue
-		}
-		data, err := base64.StdEncoding.DecodeString(blk.Data)
-		if err != nil {
-			continue
-		}
-		images = append(images, zeroruntime.ImageBlock{MediaType: blk.MimeType, Data: data})
-	}
-	return images
-}
-
-// ---- session registry + accessors ----
-
-// registerSession publishes a session under the agent's lock. If one is already
-// registered for id (e.g. a re-load of an in-flight session) the existing live
-// session is returned unchanged rather than orphaning its turn or resetting its
-// mode/model. history is set BEFORE publishing so no concurrent prompt can read a
-// half-initialized session.
-func (a *Agent) registerSession(id, cwd string, history []turnRecord) *acpSession {
-	a.mu.Lock()
-	defer a.mu.Unlock()
-	if existing := a.sessions[id]; existing != nil {
-		return existing
-	}
-	sess := &acpSession{id: id, cwd: cwd, mode: agent.PermissionModeAuto, history: history}
-	a.sessions[id] = sess
-	return sess
-}
-
-func (a *Agent) session(id string) *acpSession {
-	a.mu.Lock()
-	defer a.mu.Unlock()
-	return a.sessions[id]
-}
-
-func (s *acpSession) setCancel(cancel context.CancelFunc) {
-	s.mu.Lock()
-	s.cancel = cancel
-	s.mu.Unlock()
-}
-
-func (s *acpSession) invokeCancel() {
-	s.mu.Lock()
-	cancel := s.cancel
-	s.mu.Unlock()
-	if cancel != nil {
-		cancel()
-	}
-}
-
-func (s *acpSession) setMode(mode agent.PermissionMode) {
-	s.mu.Lock()
-	s.mode = mode
-	s.mu.Unlock()
-}
-
-func (s *acpSession) currentMode() agent.PermissionMode {
-	s.mu.Lock()
-	defer s.mu.Unlock()
-	return s.mode
-}
-
-func (s *acpSession) setModel(model string) {
-	s.mu.Lock()
-	s.model = model
-	s.mu.Unlock()
-}
-
-func (s *acpSession) currentModel() string {
-	s.mu.Lock()
-	defer s.mu.Unlock()
-	return s.model
-}
-
-func (s *acpSession) appendHistory(rec turnRecord) {
-	s.mu.Lock()
-	s.history = append(s.history, rec)
-	s.mu.Unlock()
-}
-
-func (s *acpSession) snapshotHistory() []turnRecord {
-	s.mu.Lock()
-	defer s.mu.Unlock()
-	return append([]turnRecord(nil), s.history...)
-}

internal/acp/agent_test.go

@@ -1,332 +0,0 @@
-package acp
-
-import (
-	"context"
-	"encoding/json"
-	"fmt"
-	"io"
-	"os"
-	"path/filepath"
-	"strings"
-	"testing"
-	"time"
-
-	"github.com/Gitlawb/zero/internal/agent"
-	"github.com/Gitlawb/zero/internal/config"
-	"github.com/Gitlawb/zero/internal/sandbox"
-	"github.com/Gitlawb/zero/internal/sessions"
-	"github.com/Gitlawb/zero/internal/tools"
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-// fakeProvider streams a canned assistant message and ends the turn — enough to
-// drive the real agent.Run loop without a live model.
-type fakeProvider struct{ text string }
-
-func (f fakeProvider) StreamCompletion(_ context.Context, _ zeroruntime.CompletionRequest) (<-chan zeroruntime.StreamEvent, error) {
-	ch := make(chan zeroruntime.StreamEvent, 4)
-	go func() {
-		defer close(ch)
-		ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventText, Content: f.text}
-		ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventDone}
-	}()
-	return ch, nil
-}
-
-func testDeps(t *testing.T) Deps {
-	t.Helper()
-	store := sessions.NewStore(sessions.StoreOptions{RootDir: t.TempDir()})
-	return Deps{
-		ResolveConfig: func(_ string, o config.Overrides) (config.ResolvedConfig, error) {
-			model := "fake-model"
-			if o.Provider.Model != "" {
-				model = o.Provider.Model
-			}
-			return config.ResolvedConfig{
-				Provider: config.ProviderProfile{Name: "fake", Model: model},
-				MaxTurns: 4,
-			}, nil
-		},
-		NewProvider: func(config.ProviderProfile) (zeroruntime.Provider, error) {
-			return fakeProvider{text: "Hello from ZERO"}, nil
-		},
-		RunAgent: agent.Run,
-		BuildWorkspace: func(string, config.ResolvedConfig) (*tools.Registry, *sandbox.Engine, error) {
-			r := tools.NewRegistry()
-			r.Register(tools.NewUpdatePlanTool())
-			return r, nil, nil
-		},
-		ResolveWorkspaceRoot: func(cwd string) (string, error) { return cwd, nil },
-		Store:                store,
-		AgentInfo:            Implementation{Name: "zero", Version: "test"},
-	}
-}
-
-// clientHarness wires a client Conn to an Agent over in-memory pipes and collects
-// session/update text chunks.
-type clientHarness struct {
-	client  *Conn
-	updates chan string
-	stop    func()
-}
-
-func newHarness(t *testing.T, deps Deps) *clientHarness {
-	t.Helper()
-	ar, bw := io.Pipe() // agent -> client
-	br, aw := io.Pipe() // client -> agent
-	agentConn := NewConn(ar, aw)
-	client := NewConn(br, bw)
-	a := NewAgent(agentConn, deps)
-
-	h := &clientHarness{client: client, updates: make(chan string, 128)}
-	client.HandleNotify(MethodSessionUpdate, func(_ context.Context, params json.RawMessage) {
-		var probe struct {
-			Update struct {
-				SessionUpdate string `json:"sessionUpdate"`
-				Content       struct {
-					Text string `json:"text"`
-				} `json:"content"`
-			} `json:"update"`
-		}
-		if json.Unmarshal(params, &probe) != nil {
-			return
-		}
-		if probe.Update.SessionUpdate == UpdateAgentMessageChunk {
-			h.updates <- probe.Update.Content.Text
-		}
-	})
-
-	ctx, cancel := context.WithCancel(context.Background())
-	go func() { _ = a.Serve(ctx) }()
-	go func() { _ = client.Serve(ctx) }()
-	h.stop = func() {
-		cancel()
-		_ = aw.Close()
-		_ = bw.Close()
-	}
-	return h
-}
-
-func TestACPEndToEndPrompt(t *testing.T) {
-	h := newHarness(t, testDeps(t))
-	defer h.stop()
-
-	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
-	defer cancel()
-
-	// initialize
-	var initRes InitializeResult
-	if err := h.client.Call(ctx, MethodInitialize, InitializeParams{ProtocolVersion: ProtocolVersion}, &initRes); err != nil {
-		t.Fatalf("initialize: %v", err)
-	}
-	if initRes.ProtocolVersion != ProtocolVersion {
-		t.Fatalf("protocol version = %d", initRes.ProtocolVersion)
-	}
-	if !initRes.AgentCapabilities.LoadSession || !initRes.AgentCapabilities.PromptCapabilities.Image {
-		t.Fatalf("unexpected capabilities: %+v", initRes.AgentCapabilities)
-	}
-
-	// session/new
-	var newRes NewSessionResult
-	if err := h.client.Call(ctx, MethodSessionNew, NewSessionParams{Cwd: t.TempDir(), McpServers: []McpServer{}}, &newRes); err != nil {
-		t.Fatalf("session/new: %v", err)
-	}
-	if newRes.SessionID == "" {
-		t.Fatal("session/new returned empty sessionId")
-	}
-	if newRes.Modes == nil || newRes.Modes.CurrentModeID != string(agent.PermissionModeAuto) {
-		t.Fatalf("expected auto mode, got %+v", newRes.Modes)
-	}
-
-	// session/prompt
-	var promptRes PromptResult
-	if err := h.client.Call(ctx, MethodSessionPrompt, PromptParams{
-		SessionID: newRes.SessionID,
-		Prompt:    []ContentBlock{TextBlock("hi")},
-	}, &promptRes); err != nil {
-		t.Fatalf("session/prompt: %v", err)
-	}
-	if promptRes.StopReason != StopEndTurn {
-		t.Fatalf("stopReason = %q, want %q", promptRes.StopReason, StopEndTurn)
-	}
-
-	// The streamed agent_message_chunk(s) should carry the assistant text.
-	if got := drainText(t, h.updates); !strings.Contains(got, "Hello from ZERO") {
-		t.Fatalf("streamed text = %q, want it to contain the assistant message", got)
-	}
-}
-
-func TestACPUnknownSessionPromptErrors(t *testing.T) {
-	h := newHarness(t, testDeps(t))
-	defer h.stop()
-	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
-	defer cancel()
-	err := h.client.Call(ctx, MethodSessionPrompt, PromptParams{SessionID: "nope", Prompt: []ContentBlock{TextBlock("x")}}, &PromptResult{})
-	if err == nil {
-		t.Fatal("expected error for unknown session")
-	}
-}
-
-func TestACPSetModeUpdatesSession(t *testing.T) {
-	h := newHarness(t, testDeps(t))
-	defer h.stop()
-	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
-	defer cancel()
-
-	var newRes NewSessionResult
-	if err := h.client.Call(ctx, MethodSessionNew, NewSessionParams{Cwd: t.TempDir(), McpServers: []McpServer{}}, &newRes); err != nil {
-		t.Fatalf("session/new: %v", err)
-	}
-	// auto/ask are accepted.
-	if err := h.client.Call(ctx, MethodSessionSetMode, SetSessionModeParams{SessionID: newRes.SessionID, ModeID: string(agent.PermissionModeAsk)}, &SetSessionModeResult{}); err != nil {
-		t.Fatalf("set_mode ask: %v", err)
-	}
-	// Unsafe must be rejected over ACP — a client can't self-grant no-prompt host access.
-	if err := h.client.Call(ctx, MethodSessionSetMode, SetSessionModeParams{SessionID: newRes.SessionID, ModeID: string(agent.PermissionModeUnsafe)}, &SetSessionModeResult{}); err == nil {
-		t.Fatal("expected Unsafe mode to be rejected over ACP")
-	}
-	// An unknown mode must be rejected.
-	if err := h.client.Call(ctx, MethodSessionSetMode, SetSessionModeParams{SessionID: newRes.SessionID, ModeID: "bogus"}, &SetSessionModeResult{}); err == nil {
-		t.Fatal("expected error for unknown mode")
-	}
-}
-
-// TestACPRunTurnWiresSandboxAndScopedRegistry proves the sandbox engine and the
-// scoped registry from BuildWorkspace actually reach agent.Options — i.e. ACP
-// shell tools run confined, not unconfined on the host.
-func TestACPRunTurnWiresSandboxAndScopedRegistry(t *testing.T) {
-	deps := testDeps(t)
-	reg := tools.NewRegistry()
-	reg.Register(tools.NewUpdatePlanTool())
-	engine := sandbox.NewEngine(sandbox.EngineOptions{WorkspaceRoot: t.TempDir()})
-	deps.BuildWorkspace = func(string, config.ResolvedConfig) (*tools.Registry, *sandbox.Engine, error) {
-		return reg, engine, nil
-	}
-	var captured agent.Options
-	deps.RunAgent = func(_ context.Context, _ string, _ zeroruntime.Provider, opts agent.Options) (agent.Result, error) {
-		captured = opts
-		return agent.Result{FinalAnswer: "ok"}, nil
-	}
-
-	h := newHarness(t, deps)
-	defer h.stop()
-	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
-	defer cancel()
-
-	var newRes NewSessionResult
-	if err := h.client.Call(ctx, MethodSessionNew, NewSessionParams{Cwd: t.TempDir(), McpServers: []McpServer{}}, &newRes); err != nil {
-		t.Fatalf("session/new: %v", err)
-	}
-	if err := h.client.Call(ctx, MethodSessionPrompt, PromptParams{SessionID: newRes.SessionID, Prompt: []ContentBlock{TextBlock("hi")}}, &PromptResult{}); err != nil {
-		t.Fatalf("session/prompt: %v", err)
-	}
-	if captured.Sandbox != engine {
-		t.Fatal("sandbox engine was not wired into agent.Options (shell tools would run unconfined)")
-	}
-	if captured.Registry != reg {
-		t.Fatal("scoped registry was not wired into agent.Options")
-	}
-}
-
-// TestACPRejectsInvalidCwd confirms session/new fails when the workspace root
-// resolver rejects the client cwd (e.g. filesystem root).
-func TestACPRejectsInvalidCwd(t *testing.T) {
-	deps := testDeps(t)
-	deps.ResolveWorkspaceRoot = func(string) (string, error) {
-		return "", fmt.Errorf("cwd must not be the filesystem root")
-	}
-	h := newHarness(t, deps)
-	defer h.stop()
-	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
-	defer cancel()
-	if err := h.client.Call(ctx, MethodSessionNew, NewSessionParams{Cwd: "/", McpServers: []McpServer{}}, &NewSessionResult{}); err == nil {
-		t.Fatal("expected session/new to reject an invalid cwd")
-	}
-}
-
-func TestACPPromptWarnsWhenTurnPersistenceFails(t *testing.T) {
-	deps := testDeps(t)
-	h := newHarness(t, deps)
-	defer h.stop()
-	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
-	defer cancel()
-
-	var newRes NewSessionResult
-	if err := h.client.Call(ctx, MethodSessionNew, NewSessionParams{Cwd: t.TempDir(), McpServers: []McpServer{}}, &newRes); err != nil {
-		t.Fatalf("session/new: %v", err)
-	}
-	metadataPath := filepath.Join(deps.Store.RootDir, newRes.SessionID, sessions.MetadataFile)
-	if err := os.Remove(metadataPath); err != nil {
-		t.Fatalf("remove metadata: %v", err)
-	}
-
-	var promptRes PromptResult
-	if err := h.client.Call(ctx, MethodSessionPrompt, PromptParams{
-		SessionID: newRes.SessionID,
-		Prompt:    []ContentBlock{TextBlock("hi")},
-	}, &promptRes); err != nil {
-		t.Fatalf("session/prompt: %v", err)
-	}
-	if promptRes.StopReason != StopEndTurn {
-		t.Fatalf("stopReason = %q, want %q", promptRes.StopReason, StopEndTurn)
-	}
-	got := drainTextUntil(t, h.updates, func(text string) bool {
-		return strings.Contains(text, "Hello from ZERO") &&
-			strings.Contains(text, "Could not save session history")
-	})
-	if !strings.Contains(got, "Could not save session history") {
-		t.Fatalf("streamed text = %q, want persistence warning", got)
-	}
-}
-
-func TestACPLoadWarnsWhenHistoryReadFails(t *testing.T) {
-	deps := testDeps(t)
-	cwd := t.TempDir()
-	meta, err := deps.Store.Create(sessions.CreateInput{Title: "ACP session", Cwd: cwd})
-	if err != nil {
-		t.Fatalf("create session: %v", err)
-	}
-	eventsPath := filepath.Join(deps.Store.RootDir, meta.SessionID, sessions.EventsFile)
-	if err := os.WriteFile(eventsPath, []byte("{bad json}\n"), 0o600); err != nil {
-		t.Fatalf("write corrupt events: %v", err)
-	}
-	h := newHarness(t, deps)
-	defer h.stop()
-	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
-	defer cancel()
-
-	if err := h.client.Call(ctx, MethodSessionLoad, LoadSessionParams{SessionID: meta.SessionID, Cwd: cwd, McpServers: []McpServer{}}, &LoadSessionResult{}); err != nil {
-		t.Fatalf("session/load: %v", err)
-	}
-	got := drainTextUntil(t, h.updates, func(text string) bool {
-		return strings.Contains(text, "Could not load session history")
-	})
-	if !strings.Contains(got, "Could not load session history") {
-		t.Fatalf("streamed text = %q, want load warning", got)
-	}
-}
-
-// drainText collects streamed chunks for a short window and concatenates them.
-func drainText(t *testing.T, ch <-chan string) string {
-	t.Helper()
-	return drainTextUntil(t, ch, func(text string) bool {
-		return strings.Contains(text, "Hello from ZERO")
-	})
-}
-
-func drainTextUntil(t *testing.T, ch <-chan string, done func(string) bool) string {
-	t.Helper()
-	var b strings.Builder
-	deadline := time.After(2 * time.Second)
-	for {
-		select {
-		case s := <-ch:
-			b.WriteString(s)
-			if done(b.String()) {
-				return b.String()
-			}
-		case <-deadline:
-			return b.String()
-		}
-	}
-}

internal/acp/jsonrpc.go

@@ -1,331 +0,0 @@
-// Package acp implements the Agent Client Protocol (ACP) surface for ZERO: a
-// JSON-RPC 2.0 peer spoken over stdio (newline-delimited JSON) so editors such
-// as Zed, JetBrains, and Neovim can drive ZERO's agent core as a backend.
-//
-// The protocol shapes (method names, message fields) are derived solely from the
-// public, Apache-licensed ACP specification at agentclientprotocol.com and the
-// agentclientprotocol/agent-client-protocol repository (schema/v1). All logic
-// here is written originally against ZERO's own interfaces.
-package acp
-
-import (
-	"bufio"
-	"bytes"
-	"context"
-	"encoding/json"
-	"errors"
-	"fmt"
-	"io"
-	"sync"
-)
-
-// JSON-RPC 2.0 standard error codes.
-const (
-	codeParseError     = -32700
-	codeInvalidRequest = -32600
-	codeMethodNotFound = -32601
-	codeInvalidParams  = -32602
-	codeInternalError  = -32603
-)
-
-// rpcError is a JSON-RPC 2.0 error object.
-type rpcError struct {
-	Code    int             `json:"code"`
-	Message string          `json:"message"`
-	Data    json.RawMessage `json:"data,omitempty"`
-}
-
-func (e *rpcError) Error() string {
-	if e == nil {
-		return "<nil rpc error>"
-	}
-	return fmt.Sprintf("jsonrpc error %d: %s", e.Code, e.Message)
-}
-
-// RPCError builds a protocol error to return from a handler. Use it when a
-// handler needs to control the wire-level error code; any other error a handler
-// returns is reported as a generic internal error.
-func RPCError(code int, message string) error {
-	return &rpcError{Code: code, Message: message}
-}
-
-// rpcMessage is the union of request, response, and notification frames. The
-// shape is disambiguated by which fields are present (per JSON-RPC 2.0):
-//   - request:      method + id
-//   - notification: method, no id
-//   - response:     id + (result | error), no method
-type rpcMessage struct {
-	JSONRPC string          `json:"jsonrpc"`
-	ID      json.RawMessage `json:"id,omitempty"`
-	Method  string          `json:"method,omitempty"`
-	Params  json.RawMessage `json:"params,omitempty"`
-	Result  json.RawMessage `json:"result,omitempty"`
-	Error   *rpcError       `json:"error,omitempty"`
-}
-
-func (m rpcMessage) isResponse() bool { return m.Method == "" && len(m.ID) > 0 }
-func (m rpcMessage) isRequest() bool  { return m.Method != "" && len(m.ID) > 0 }
-func (m rpcMessage) isNotify() bool   { return m.Method != "" && len(m.ID) == 0 }
-
-// HandlerFunc handles an inbound request and returns the result to encode, or an
-// error. Returning an *rpcError controls the wire code; any other error becomes
-// a generic internal error.
-type HandlerFunc func(ctx context.Context, params json.RawMessage) (any, error)
-
-// NotifyFunc handles an inbound notification (no response is sent).
-type NotifyFunc func(ctx context.Context, params json.RawMessage)
-
-// Conn is a JSON-RPC 2.0 peer over a single ndjson stream pair. It both serves
-// inbound requests/notifications (via registered handlers) and issues outbound
-// requests/notifications — needed because ACP is bidirectional (the agent calls
-// the client for session/request_permission, fs/*, terminal/*).
-type Conn struct {
-	reader *bufio.Reader
-	w      io.Writer
-
-	writeMu sync.Mutex // serializes all writes to w
-
-	handlers  map[string]HandlerFunc
-	notifiers map[string]NotifyFunc
-
-	mu      sync.Mutex
-	nextID  int64
-	pending map[int64]chan rpcMessage
-	closed  bool
-
-	wg sync.WaitGroup // tracks in-flight inbound handlers
-}
-
-// NewConn builds a peer reading ndjson from r and writing ndjson to w.
-func NewConn(r io.Reader, w io.Writer) *Conn {
-	return &Conn{
-		reader:    bufio.NewReader(r),
-		w:         w,
-		handlers:  make(map[string]HandlerFunc),
-		notifiers: make(map[string]NotifyFunc),
-		pending:   make(map[int64]chan rpcMessage),
-	}
-}
-
-// Handle registers a request handler for method.
-func (c *Conn) Handle(method string, fn HandlerFunc) { c.handlers[method] = fn }
-
-// HandleNotify registers a notification handler for method.
-func (c *Conn) HandleNotify(method string, fn NotifyFunc) { c.notifiers[method] = fn }
-
-// Serve runs the read loop until the stream ends, ctx is cancelled, or a fatal
-// decode error occurs. Inbound requests and notifications are dispatched on
-// their own goroutines so a long-running handler (e.g. session/prompt) never
-// blocks the loop from delivering session/cancel or a permission response.
-func (c *Conn) Serve(ctx context.Context) error {
-	ctx, cancel := context.WithCancel(ctx)
-	// On exit, cancel in-flight handlers (so a blocked outbound Call unblocks via
-	// ctx) and then wait for them to finish writing their responses. Without this,
-	// a finite input stream (e.g. piped ndjson that EOFs right after a request)
-	// would race the dispatch goroutine and drop the response.
-	defer func() {
-		cancel()
-		c.wg.Wait()
-	}()
-
-	for {
-		// One ndjson value per line. Framing per line (rather than streaming the
-		// decoder) keeps a single malformed line from making the whole connection
-		// unrecoverable — we report -32700 and continue.
-		line, err := c.reader.ReadBytes('\n')
-		if len(bytes.TrimSpace(line)) > 0 {
-			c.handleLine(ctx, line)
-		}
-		if err != nil {
-			c.failAllPending(err)
-			if errors.Is(err, io.EOF) || ctx.Err() != nil {
-				return nil
-			}
-			return err
-		}
-	}
-}
-
-// handleLine decodes and dispatches one ndjson frame. A parse failure replies
-// with -32700 (id null) and keeps the connection alive.
-func (c *Conn) handleLine(ctx context.Context, line []byte) {
-	var msg rpcMessage
-	if err := json.Unmarshal(bytes.TrimSpace(line), &msg); err != nil {
-		c.writeError(json.RawMessage("null"), &rpcError{Code: codeParseError, Message: "parse error"})
-		return
-	}
-	if msg.JSONRPC != "" && msg.JSONRPC != "2.0" {
-		if len(msg.ID) > 0 {
-			c.writeError(msg.ID, &rpcError{Code: codeInvalidRequest, Message: "unsupported jsonrpc version"})
-		}
-		return
-	}
-	switch {
-	case msg.isResponse():
-		c.deliver(msg)
-	case msg.isRequest():
-		c.wg.Add(1)
-		go func(m rpcMessage) {
-			defer c.wg.Done()
-			c.dispatchRequest(ctx, m)
-		}(msg)
-	case msg.isNotify():
-		if fn := c.notifiers[msg.Method]; fn != nil {
-			c.wg.Add(1)
-			go func(m rpcMessage) {
-				defer c.wg.Done()
-				fn(ctx, m.Params)
-			}(msg)
-		}
-	default:
-		// Malformed frame; reply only if we can identify a request id.
-		if len(msg.ID) > 0 {
-			c.writeError(msg.ID, &rpcError{Code: codeInvalidRequest, Message: "invalid request"})
-		}
-	}
-}
-
-func (c *Conn) dispatchRequest(ctx context.Context, msg rpcMessage) {
-	fn := c.handlers[msg.Method]
-	if fn == nil {
-		c.writeError(msg.ID, &rpcError{Code: codeMethodNotFound, Message: "method not found: " + msg.Method})
-		return
-	}
-	result, err := fn(ctx, msg.Params)
-	if err != nil {
-		var re *rpcError
-		if errors.As(err, &re) {
-			c.writeError(msg.ID, re)
-		} else {
-			c.writeError(msg.ID, &rpcError{Code: codeInternalError, Message: err.Error()})
-		}
-		return
-	}
-	c.writeResult(msg.ID, result)
-}
-
-// Call issues an outbound request and blocks until the response arrives, ctx is
-// cancelled, or the connection closes.
-func (c *Conn) Call(ctx context.Context, method string, params any, result any) error {
-	raw, err := marshalParams(params)
-	if err != nil {
-		return err
-	}
-
-	c.mu.Lock()
-	if c.closed {
-		c.mu.Unlock()
-		return errors.New("acp: connection closed")
-	}
-	c.nextID++
-	id := c.nextID
-	ch := make(chan rpcMessage, 1)
-	c.pending[id] = ch
-	c.mu.Unlock()
-
-	defer func() {
-		c.mu.Lock()
-		delete(c.pending, id)
-		c.mu.Unlock()
-	}()
-
-	idRaw, _ := json.Marshal(id)
-	if err := c.write(rpcMessage{JSONRPC: "2.0", ID: idRaw, Method: method, Params: raw}); err != nil {
-		return err
-	}
-
-	select {
-	case <-ctx.Done():
-		return ctx.Err()
-	case resp := <-ch:
-		if resp.Error != nil {
-			return resp.Error
-		}
-		if result != nil && len(resp.Result) > 0 {
-			return json.Unmarshal(resp.Result, result)
-		}
-		return nil
-	}
-}
-
-// Notify issues an outbound notification (no response expected).
-func (c *Conn) Notify(method string, params any) error {
-	raw, err := marshalParams(params)
-	if err != nil {
-		return err
-	}
-	return c.write(rpcMessage{JSONRPC: "2.0", Method: method, Params: raw})
-}
-
-func (c *Conn) deliver(msg rpcMessage) {
-	var id int64
-	if err := json.Unmarshal(msg.ID, &id); err != nil {
-		return
-	}
-	c.mu.Lock()
-	ch := c.pending[id]
-	delete(c.pending, id)
-	c.mu.Unlock()
-	if ch == nil {
-		return
-	}
-	// Non-blocking: the pending channel is cap-1 and may have been abandoned (e.g.
-	// a Call cancelled by session/cancel). A duplicate or late response frame must
-	// never block the read-loop goroutine.
-	select {
-	case ch <- msg:
-	default:
-	}
-}
-
-func (c *Conn) failAllPending(err error) {
-	c.mu.Lock()
-	c.closed = true
-	pending := c.pending
-	c.pending = make(map[int64]chan rpcMessage)
-	c.mu.Unlock()
-	for _, ch := range pending {
-		select {
-		case ch <- rpcMessage{Error: &rpcError{Code: codeInternalError, Message: err.Error()}}:
-		default:
-		}
-	}
-}
-
-func (c *Conn) writeResult(id json.RawMessage, result any) {
-	raw, err := json.Marshal(result)
-	if err != nil {
-		c.writeError(id, &rpcError{Code: codeInternalError, Message: err.Error()})
-		return
-	}
-	_ = c.write(rpcMessage{JSONRPC: "2.0", ID: id, Result: raw})
-}
-
-func (c *Conn) writeError(id json.RawMessage, e *rpcError) {
-	_ = c.write(rpcMessage{JSONRPC: "2.0", ID: id, Error: e})
-}
-
-func (c *Conn) write(msg rpcMessage) error {
-	msg.JSONRPC = "2.0"
-	data, err := json.Marshal(msg)
-	if err != nil {
-		return err
-	}
-	c.writeMu.Lock()
-	defer c.writeMu.Unlock()
-	if _, err := c.w.Write(append(data, '\n')); err != nil {
-		return err
-	}
-	return nil
-}
-
-// marshalParams encodes params, treating nil as an absent params field.
-func marshalParams(params any) (json.RawMessage, error) {
-	if params == nil {
-		return nil, nil
-	}
-	if raw, ok := params.(json.RawMessage); ok {
-		return raw, nil
-	}
-	return json.Marshal(params)
-}

internal/acp/jsonrpc_test.go

@@ -1,199 +0,0 @@
-package acp
-
-import (
-	"context"
-	"encoding/json"
-	"io"
-	"testing"
-	"time"
-)
-
-// connPair wires two Conns together over in-memory pipes and serves both.
-func connPair(t *testing.T) (a, b *Conn, stop func()) {
-	t.Helper()
-	ar, bw := io.Pipe() // b -> a
-	br, aw := io.Pipe() // a -> b
-	a = NewConn(ar, aw)
-	b = NewConn(br, bw)
-	ctx, cancel := context.WithCancel(context.Background())
-	go func() { _ = a.Serve(ctx) }()
-	go func() { _ = b.Serve(ctx) }()
-	return a, b, func() {
-		cancel()
-		_ = aw.Close()
-		_ = bw.Close()
-	}
-}
-
-func TestConnRequestResponse(t *testing.T) {
-	a, b, stop := connPair(t)
-	defer stop()
-
-	b.Handle("add", func(_ context.Context, params json.RawMessage) (any, error) {
-		var in struct{ X, Y int }
-		if err := json.Unmarshal(params, &in); err != nil {
-			return nil, RPCError(codeInvalidParams, "bad params")
-		}
-		return map[string]int{"sum": in.X + in.Y}, nil
-	})
-
-	var out struct{ Sum int }
-	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
-	defer cancel()
-	if err := a.Call(ctx, "add", map[string]int{"X": 2, "Y": 3}, &out); err != nil {
-		t.Fatalf("call: %v", err)
-	}
-	if out.Sum != 5 {
-		t.Fatalf("sum = %d, want 5", out.Sum)
-	}
-}
-
-func TestConnNotification(t *testing.T) {
-	a, b, stop := connPair(t)
-	defer stop()
-
-	got := make(chan string, 1)
-	b.HandleNotify("ping", func(_ context.Context, params json.RawMessage) {
-		var in struct{ Msg string }
-		_ = json.Unmarshal(params, &in)
-		got <- in.Msg
-	})
-
-	if err := a.Notify("ping", map[string]string{"Msg": "hello"}); err != nil {
-		t.Fatalf("notify: %v", err)
-	}
-	select {
-	case msg := <-got:
-		if msg != "hello" {
-			t.Fatalf("got %q, want hello", msg)
-		}
-	case <-time.After(2 * time.Second):
-		t.Fatal("notification not delivered")
-	}
-}
-
-func TestConnMethodNotFound(t *testing.T) {
-	a, _, stop := connPair(t)
-	defer stop()
-
-	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
-	defer cancel()
-	err := a.Call(ctx, "does_not_exist", nil, nil)
-	var re *rpcError
-	if !asRPCError(err, &re) {
-		t.Fatalf("expected rpcError, got %v", err)
-	}
-	if re.Code != codeMethodNotFound {
-		t.Fatalf("code = %d, want %d", re.Code, codeMethodNotFound)
-	}
-}
-
-func TestConnHandlerError(t *testing.T) {
-	a, b, stop := connPair(t)
-	defer stop()
-	b.Handle("boom", func(_ context.Context, _ json.RawMessage) (any, error) {
-		return nil, RPCError(codeInvalidParams, "nope")
-	})
-	ctx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
-	defer cancel()
-	err := a.Call(ctx, "boom", nil, nil)
-	var re *rpcError
-	if !asRPCError(err, &re) || re.Code != codeInvalidParams {
-		t.Fatalf("expected invalid-params rpcError, got %v", err)
-	}
-}
-
-// TestConnBidirectionalDuringHandler proves that while one peer is inside a
-// request handler it can issue an outbound request back to the caller and the
-// caller answers it — exactly the session/prompt -> session/request_permission
-// pattern. If the read loop blocked on the handler, this would deadlock.
-func TestConnBidirectionalDuringHandler(t *testing.T) {
-	a, b, stop := connPair(t)
-	defer stop()
-
-	// a answers an "approve?" callback.
-	a.Handle("approve", func(_ context.Context, _ json.RawMessage) (any, error) {
-		return map[string]bool{"ok": true}, nil
-	})
-
-	// b's "run" handler calls back to a mid-flight.
-	b.Handle("run", func(ctx context.Context, _ json.RawMessage) (any, error) {
-		var approval struct{ OK bool }
-		if err := b.Call(ctx, "approve", nil, &approval); err != nil {
-			return nil, err
-		}
-		return map[string]bool{"ran": approval.OK}, nil
-	})
-
-	var out struct{ Ran bool }
-	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
-	defer cancel()
-	if err := a.Call(ctx, "run", nil, &out); err != nil {
-		t.Fatalf("run: %v", err)
-	}
-	if !out.Ran {
-		t.Fatal("expected ran=true via mid-handler callback")
-	}
-}
-
-// TestConnSurvivesMalformedLine proves a single bad ndjson line yields a -32700
-// and does NOT tear down the connection — a following valid request still works.
-func TestConnSurvivesMalformedLine(t *testing.T) {
-	clientR, serverW := io.Pipe() // server -> client
-	serverR, clientW := io.Pipe() // client -> server
-	server := NewConn(serverR, serverW)
-	server.Handle("ping", func(_ context.Context, _ json.RawMessage) (any, error) { return "pong", nil })
-	ctx, cancel := context.WithCancel(context.Background())
-	defer func() {
-		cancel()
-		_ = serverW.Close()
-		_ = clientW.Close()
-	}()
-	go func() { _ = server.Serve(ctx) }()
-
-	go func() {
-		_, _ = clientW.Write([]byte("this is not json\n"))
-		_, _ = clientW.Write([]byte(`{"jsonrpc":"2.0","id":1,"method":"ping"}` + "\n"))
-	}()
-
-	dec := json.NewDecoder(clientR)
-	var sawParseError, sawPong bool
-	for i := 0; i < 2; i++ {
-		var msg struct {
-			Result any `json:"result"`
-			Error  *struct {
-				Code int `json:"code"`
-			} `json:"error"`
-		}
-		done := make(chan error, 1)
-		go func() { done <- dec.Decode(&msg) }()
-		select {
-		case err := <-done:
-			if err != nil {
-				t.Fatalf("decode response %d: %v", i, err)
-			}
-		case <-time.After(3 * time.Second):
-			t.Fatal("timed out waiting for a response")
-		}
-		if msg.Error != nil && msg.Error.Code == codeParseError {
-			sawParseError = true
-		}
-		if r, ok := msg.Result.(string); ok && r == "pong" {
-			sawPong = true
-		}
-	}
-	if !sawParseError {
-		t.Error("expected a -32700 parse error for the malformed line")
-	}
-	if !sawPong {
-		t.Error("expected the valid request to still be answered (connection survived the bad line)")
-	}
-}
-
-func asRPCError(err error, target **rpcError) bool {
-	re, ok := err.(*rpcError)
-	if ok {
-		*target = re
-	}
-	return ok
-}

internal/acp/permission.go

@@ -1,125 +0,0 @@
-package acp
-
-import (
-	"encoding/json"
-
-	"github.com/Gitlawb/zero/internal/agent"
-)
-
-// permission.go maps ZERO's permission prompt model onto ACP's
-// session/request_permission request/option/outcome model. The option id on the
-// wire carries the ZERO decision action verbatim, so mapping the client's
-// selection back to a ZERO decision is exact and lossless.
-
-// buildPermissionOptions turns the decisions ZERO offers for a tool call into ACP
-// PermissionOptions. Only the actions ZERO actually presented (AvailableDecisions)
-// are surfaced; the optionId is the ZERO action string for a clean round-trip.
-func buildPermissionOptions(req agent.PermissionRequest) []PermissionOption {
-	actions := req.AvailableDecisions
-	if len(actions) == 0 {
-		// Sensible default if ZERO didn't enumerate: allow once / reject.
-		actions = []agent.PermissionDecisionAction{
-			agent.PermissionDecisionAllow,
-			agent.PermissionDecisionDeny,
-		}
-	}
-	options := make([]PermissionOption, 0, len(actions))
-	for _, action := range actions {
-		kind, name := optionKindFor(action)
-		if kind == "" {
-			continue // skip actions that have no clean ACP option (e.g. cancel)
-		}
-		options = append(options, PermissionOption{
-			OptionID: string(action),
-			Name:     name,
-			Kind:     kind,
-		})
-	}
-	return options
-}
-
-// optionKindFor maps a ZERO decision action to an ACP PermissionOptionKind and a
-// human label. Returns an empty kind for actions that ACP expresses through the
-// outcome rather than an option (cancel).
-func optionKindFor(action agent.PermissionDecisionAction) (kind, name string) {
-	switch action {
-	case agent.PermissionDecisionAllow, agent.PermissionDecisionAllowStrict:
-		return PermAllowOnce, "Allow"
-	case agent.PermissionDecisionAllowForSession:
-		return PermAllowAlways, "Allow for this session"
-	case agent.PermissionDecisionAllowPrefix:
-		return PermAllowAlways, "Allow this command for the session"
-	case agent.PermissionDecisionAlwaysAllow:
-		return PermAllowAlways, "Always allow"
-	case agent.PermissionDecisionAlwaysAllowPrefix:
-		return PermAllowAlways, "Always allow this command"
-	case agent.PermissionDecisionDeny:
-		return PermRejectOnce, "Reject"
-	case agent.PermissionDecisionCancel:
-		return "", "" // expressed as outcome=cancelled, not an option
-	default:
-		return "", ""
-	}
-}
-
-// decisionFromOutcome maps the client's permission outcome back to a ZERO
-// decision. A cancelled outcome cancels the run; a selected option id is the ZERO
-// action verbatim (validated against what was offered); anything unrecognized
-// fails closed to deny.
-func decisionFromOutcome(outcome RequestPermissionOutcome, offered []agent.PermissionDecisionAction) agent.PermissionDecision {
-	switch outcome.Outcome {
-	case OutcomeCancelled:
-		return agent.PermissionDecision{Action: agent.PermissionDecisionCancel, Reason: "client cancelled"}
-	case OutcomeSelected:
-		action := agent.PermissionDecisionAction(outcome.OptionID)
-		// Bind to what was actually offered for THIS call: a client must not be able
-		// to return a broader grant (always_allow / allow_for_session) that wasn't
-		// presented. Anything not offered fails closed to deny.
-		if actionOffered(action, offered) {
-			return agent.PermissionDecision{Action: action}
-		}
-		return agent.PermissionDecision{Action: agent.PermissionDecisionDeny, Reason: "permission option was not offered"}
-	default:
-		return agent.PermissionDecision{Action: agent.PermissionDecisionDeny, Reason: "no permission outcome"}
-	}
-}
-
-func actionOffered(action agent.PermissionDecisionAction, offered []agent.PermissionDecisionAction) bool {
-	for _, a := range offered {
-		if a == action {
-			return true
-		}
-	}
-	return false
-}
-
-// permissionToolCall builds the ToolCall descriptor embedded in a
-// session/request_permission request from a ZERO permission request.
-func permissionToolCall(req agent.PermissionRequest) ToolCallUpdate {
-	args := marshalArgs(req.Args)
-	return ToolCallUpdate{
-		ToolCallID: req.ToolCallID,
-		Title:      toolTitle(req.ToolName, string(args)),
-		Kind:       toolKindFor(req.ToolName),
-		Status:     ToolStatusPending,
-		RawInput:   rawInputBytes(args),
-	}
-}
-
-func marshalArgs(args map[string]any) []byte {
-	if len(args) == 0 {
-		return nil
-	}
-	data, err := json.Marshal(args)
-	if err != nil {
-		return nil
-	}
-	return data
-}
-
-func rawInputBytes(data []byte) json.RawMessage {
-	if len(data) == 0 || !json.Valid(data) {
-		return nil
-	}
-	return json.RawMessage(data)
-}

internal/acp/permission_test.go

@@ -1,81 +0,0 @@
-package acp
-
-import (
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/agent"
-)
-
-func TestBuildPermissionOptions(t *testing.T) {
-	req := agent.PermissionRequest{
-		ToolCallID: "tc1",
-		ToolName:   "bash",
-		AvailableDecisions: []agent.PermissionDecisionAction{
-			agent.PermissionDecisionAllow,
-			agent.PermissionDecisionAllowForSession,
-			agent.PermissionDecisionAlwaysAllow,
-			agent.PermissionDecisionDeny,
-			agent.PermissionDecisionCancel, // must be dropped (expressed as outcome)
-		},
-	}
-	opts := buildPermissionOptions(req)
-	if len(opts) != 4 {
-		t.Fatalf("expected 4 options (cancel dropped), got %d: %+v", len(opts), opts)
-	}
-	// optionId must carry the ZERO action verbatim for a clean round trip.
-	if opts[0].OptionID != string(agent.PermissionDecisionAllow) || opts[0].Kind != PermAllowOnce {
-		t.Errorf("allow option = %+v", opts[0])
-	}
-	if opts[1].Kind != PermAllowAlways || opts[3].Kind != PermRejectOnce {
-		t.Errorf("kinds = %q, %q", opts[1].Kind, opts[3].Kind)
-	}
-}
-
-func TestBuildPermissionOptionsDefault(t *testing.T) {
-	opts := buildPermissionOptions(agent.PermissionRequest{ToolName: "x"})
-	if len(opts) != 2 {
-		t.Fatalf("expected default allow+deny, got %d", len(opts))
-	}
-}
-
-func TestDecisionFromOutcome(t *testing.T) {
-	offered := []agent.PermissionDecisionAction{
-		agent.PermissionDecisionAllow,
-		agent.PermissionDecisionAlwaysAllow,
-		agent.PermissionDecisionDeny,
-	}
-	if d := decisionFromOutcome(RequestPermissionOutcome{Outcome: OutcomeCancelled}, offered); d.Action != agent.PermissionDecisionCancel {
-		t.Errorf("cancelled -> %q, want cancel", d.Action)
-	}
-	if d := decisionFromOutcome(RequestPermissionOutcome{Outcome: OutcomeSelected, OptionID: "allow"}, offered); d.Action != agent.PermissionDecisionAllow {
-		t.Errorf("selected allow -> %q", d.Action)
-	}
-	if d := decisionFromOutcome(RequestPermissionOutcome{Outcome: OutcomeSelected, OptionID: "always_allow"}, offered); d.Action != agent.PermissionDecisionAlwaysAllow {
-		t.Errorf("selected always_allow -> %q", d.Action)
-	}
-	// Unknown option fails closed to deny.
-	if d := decisionFromOutcome(RequestPermissionOutcome{Outcome: OutcomeSelected, OptionID: "bogus"}, offered); d.Action != agent.PermissionDecisionDeny {
-		t.Errorf("unknown option -> %q, want deny", d.Action)
-	}
-	// Missing/empty outcome fails closed to deny.
-	if d := decisionFromOutcome(RequestPermissionOutcome{}, offered); d.Action != agent.PermissionDecisionDeny {
-		t.Errorf("empty outcome -> %q, want deny", d.Action)
-	}
-}
-
-func TestPermissionToolCall(t *testing.T) {
-	tc := permissionToolCall(agent.PermissionRequest{
-		ToolCallID: "tc9",
-		ToolName:   "read_file",
-		Args:       map[string]any{"path": "a.go"},
-	})
-	if tc.ToolCallID != "tc9" || tc.Kind != ToolKindRead || tc.Status != ToolStatusPending {
-		t.Fatalf("unexpected toolCall: %+v", tc)
-	}
-	if tc.Title != "read_file a.go" {
-		t.Errorf("title = %q", tc.Title)
-	}
-	if len(tc.RawInput) == 0 {
-		t.Error("expected rawInput from args")
-	}
-}

internal/acp/translate.go

@@ -1,215 +0,0 @@
-package acp
-
-import (
-	"encoding/json"
-	"strings"
-	"unicode/utf8"
-
-	"github.com/Gitlawb/zero/internal/agent"
-	"github.com/Gitlawb/zero/internal/tools"
-)
-
-// translate.go maps ZERO's agent events onto ACP session/update payloads. The
-// mapping functions are pure (no I/O) so they can be unit-tested directly; the
-// notifier wires them to a live JSON-RPC connection.
-
-func agentMessageChunk(delta string) ContentChunk {
-	return ContentChunk{SessionUpdate: UpdateAgentMessageChunk, Content: TextBlock(delta)}
-}
-
-func agentThoughtChunk(delta string) ContentChunk {
-	return ContentChunk{SessionUpdate: UpdateAgentThoughtChunk, Content: TextBlock(delta)}
-}
-
-// toolKindFor maps a ZERO tool name to the closest ACP ToolKind so editors can
-// pick an icon/affordance. Unknown tools fall back to "other".
-func toolKindFor(name string) string {
-	switch name {
-	case "read_file", "read_minified_file", "list_directory":
-		return ToolKindRead
-	case "glob", "grep":
-		return ToolKindSearch
-	case "write_file", "edit_file", "apply_patch":
-		return ToolKindEdit
-	case "bash", "exec_command", "write_stdin":
-		return ToolKindExecute
-	case "web_fetch", "web_search":
-		return ToolKindFetch
-	case "update_plan":
-		return ToolKindThink
-	default:
-		return ToolKindOther
-	}
-}
-
-// toolTitle builds a concise human title, e.g. "read_file src/main.go".
-func toolTitle(name, rawArgs string) string {
-	if hint := primaryArgHint(rawArgs); hint != "" {
-		return name + " " + hint
-	}
-	return name
-}
-
-// primaryArgHint extracts the most relevant argument (path/pattern/command) from
-// raw JSON arguments. Best-effort; returns "" when it can't parse.
-func primaryArgHint(rawArgs string) string {
-	if strings.TrimSpace(rawArgs) == "" {
-		return ""
-	}
-	var m map[string]any
-	if err := json.Unmarshal([]byte(rawArgs), &m); err != nil {
-		return ""
-	}
-	for _, key := range []string{"path", "file_path", "pattern", "query", "command", "url", "cwd"} {
-		if v, ok := m[key].(string); ok && strings.TrimSpace(v) != "" {
-			return truncateHint(v)
-		}
-	}
-	return ""
-}
-
-func truncateHint(s string) string {
-	s = strings.TrimSpace(s)
-	const max = 60
-	if utf8.RuneCountInString(s) > max {
-		return string([]rune(s)[:max]) + "…"
-	}
-	return s
-}
-
-// rawInput returns the tool arguments as a raw JSON object when they parse, else
-// nil (so a malformed/empty arg string never produces invalid JSON on the wire).
-func rawInput(args string) json.RawMessage {
-	if strings.TrimSpace(args) == "" || !json.Valid([]byte(args)) {
-		return nil
-	}
-	return json.RawMessage(args)
-}
-
-// toolCallStart maps an advertised ZERO tool call to the initial ACP "tool_call"
-// update (status in_progress — ZERO executes immediately after advertising).
-func toolCallStart(call agent.ToolCall) ToolCallUpdate {
-	return ToolCallUpdate{
-		SessionUpdate: UpdateToolCall,
-		ToolCallID:    call.ID,
-		Title:         toolTitle(call.Name, call.Arguments),
-		Kind:          toolKindFor(call.Name),
-		Status:        ToolStatusInProgress,
-		RawInput:      rawInput(call.Arguments),
-	}
-}
-
-// toolCallResult maps a finished ZERO tool result to a "tool_call_update".
-func toolCallResult(result agent.ToolResult) ToolCallUpdate {
-	status := ToolStatusCompleted
-	if result.Status == tools.StatusError {
-		status = ToolStatusFailed
-	}
-	upd := ToolCallUpdate{
-		SessionUpdate: UpdateToolCallUpdate,
-		ToolCallID:    result.ToolCallID,
-		Status:        status,
-	}
-	if content := toolResultContent(result); len(content) > 0 {
-		upd.Content = content
-	}
-	if locs := toolResultLocations(result); len(locs) > 0 {
-		upd.Locations = locs
-	}
-	return upd
-}
-
-func toolResultContent(result agent.ToolResult) []ToolCallContent {
-	text := strings.TrimRight(result.Output, "\n")
-	if text == "" {
-		text = result.Display.Summary
-	}
-	if text == "" {
-		return nil
-	}
-	return []ToolCallContent{ToolContent(TextBlock(text))}
-}
-
-func toolResultLocations(result agent.ToolResult) []ToolCallLocation {
-	locs := make([]ToolCallLocation, 0, len(result.ChangedFiles))
-	for _, f := range result.ChangedFiles {
-		if strings.TrimSpace(f) == "" {
-			continue
-		}
-		locs = append(locs, ToolCallLocation{Path: f})
-	}
-	return locs
-}
-
-// planUpdate maps ZERO's plan items to an ACP "plan" update.
-func planUpdate(items []tools.PlanItem) PlanUpdate {
-	entries := make([]PlanEntry, 0, len(items))
-	for _, it := range items {
-		entries = append(entries, PlanEntry{
-			Content:  it.Content,
-			Priority: PlanPriorityMedium,
-			Status:   planStatusToACP(it.Status),
-		})
-	}
-	return PlanUpdate{SessionUpdate: UpdatePlan, Entries: entries}
-}
-
-// planStatusToACP maps ZERO's plan status (pending/in_progress/completed/failed)
-// to ACP's PlanEntryStatus (which has no "failed"; a failed step is terminal, so
-// it maps to completed).
-func planStatusToACP(s string) string {
-	switch s {
-	case "in_progress":
-		return PlanStatusInProgress
-	case "completed", "failed":
-		return PlanStatusCompleted
-	default:
-		return PlanStatusPending
-	}
-}
-
-// promptText concatenates the text content blocks of an inbound prompt.
-func promptText(blocks []ContentBlock) string {
-	var b strings.Builder
-	for _, blk := range blocks {
-		if blk.Type == "text" {
-			b.WriteString(blk.Text)
-		}
-	}
-	return b.String()
-}
-
-// notifier sends translated updates over a connection for one session.
-type notifier struct {
-	conn      *Conn
-	sessionID string
-}
-
-func (n *notifier) send(update any) {
-	_ = n.conn.Notify(MethodSessionUpdate, SessionNotification{SessionID: n.sessionID, Update: update})
-}
-
-func (n *notifier) text(delta string) {
-	if delta != "" {
-		n.send(agentMessageChunk(delta))
-	}
-}
-
-func (n *notifier) thought(delta string) {
-	if delta != "" {
-		n.send(agentThoughtChunk(delta))
-	}
-}
-
-func (n *notifier) toolCall(call agent.ToolCall)       { n.send(toolCallStart(call)) }
-func (n *notifier) toolResult(result agent.ToolResult) { n.send(toolCallResult(result)) }
-
-func (n *notifier) plan(items []tools.PlanItem) {
-	if len(items) > 0 {
-		n.send(planUpdate(items))
-	}
-}
-
-func (n *notifier) currentMode(modeID string) {
-	n.send(CurrentModeUpdate{SessionUpdate: UpdateCurrentMode, CurrentModeID: modeID})
-}

internal/acp/translate_test.go

@@ -1,162 +0,0 @@
-package acp
-
-import (
-	"strings"
-	"testing"
-	"unicode/utf8"
-
-	"github.com/Gitlawb/zero/internal/agent"
-	"github.com/Gitlawb/zero/internal/tools"
-)
-
-func TestAgentMessageAndThoughtChunks(t *testing.T) {
-	m := agentMessageChunk("hello")
-	if m.SessionUpdate != UpdateAgentMessageChunk || m.Content.Type != "text" || m.Content.Text != "hello" {
-		t.Fatalf("unexpected message chunk: %+v", m)
-	}
-	th := agentThoughtChunk("thinking")
-	if th.SessionUpdate != UpdateAgentThoughtChunk || th.Content.Text != "thinking" {
-		t.Fatalf("unexpected thought chunk: %+v", th)
-	}
-}
-
-func TestToolKindFor(t *testing.T) {
-	cases := map[string]string{
-		"read_file":      ToolKindRead,
-		"list_directory": ToolKindRead,
-		"grep":           ToolKindSearch,
-		"glob":           ToolKindSearch,
-		"edit_file":      ToolKindEdit,
-		"apply_patch":    ToolKindEdit,
-		"bash":           ToolKindExecute,
-		"exec_command":   ToolKindExecute,
-		"web_fetch":      ToolKindFetch,
-		"update_plan":    ToolKindThink,
-		"some_mcp_tool":  ToolKindOther,
-	}
-	for name, want := range cases {
-		if got := toolKindFor(name); got != want {
-			t.Errorf("toolKindFor(%q) = %q, want %q", name, got, want)
-		}
-	}
-}
-
-func TestToolTitleAndHint(t *testing.T) {
-	if got := toolTitle("read_file", `{"path":"src/main.go"}`); got != "read_file src/main.go" {
-		t.Errorf("title = %q", got)
-	}
-	if got := toolTitle("bash", `{"command":"go test ./..."}`); got != "bash go test ./..." {
-		t.Errorf("title = %q", got)
-	}
-	if got := toolTitle("mystery", `not json`); got != "mystery" {
-		t.Errorf("malformed args should yield bare name, got %q", got)
-	}
-	if got := toolTitle("noargs", ``); got != "noargs" {
-		t.Errorf("empty args should yield bare name, got %q", got)
-	}
-}
-
-func TestToolCallStart(t *testing.T) {
-	upd := toolCallStart(agent.ToolCall{ID: "tc1", Name: "read_file", Arguments: `{"path":"a.go"}`})
-	if upd.SessionUpdate != UpdateToolCall {
-		t.Fatalf("sessionUpdate = %q", upd.SessionUpdate)
-	}
-	if upd.ToolCallID != "tc1" || upd.Status != ToolStatusInProgress || upd.Kind != ToolKindRead {
-		t.Fatalf("unexpected start: %+v", upd)
-	}
-	if string(upd.RawInput) != `{"path":"a.go"}` {
-		t.Fatalf("rawInput = %s", upd.RawInput)
-	}
-	// Malformed args must not produce invalid JSON on the wire.
-	if got := toolCallStart(agent.ToolCall{ID: "x", Name: "bash", Arguments: "broken"}); got.RawInput != nil {
-		t.Fatalf("malformed args should drop rawInput, got %s", got.RawInput)
-	}
-}
-
-func TestToolCallResult(t *testing.T) {
-	ok := toolCallResult(agent.ToolResult{
-		ToolCallID:   "tc1",
-		Name:         "edit_file",
-		Status:       tools.StatusOK,
-		Output:       "applied\n",
-		ChangedFiles: []string{"a.go", ""},
-	})
-	if ok.SessionUpdate != UpdateToolCallUpdate || ok.Status != ToolStatusCompleted {
-		t.Fatalf("unexpected ok result: %+v", ok)
-	}
-	if len(ok.Content) != 1 || ok.Content[0].Type != "content" || ok.Content[0].Content.Text != "applied" {
-		t.Fatalf("unexpected content: %+v", ok.Content)
-	}
-	if len(ok.Locations) != 1 || ok.Locations[0].Path != "a.go" {
-		t.Fatalf("blank changed files should be dropped, got %+v", ok.Locations)
-	}
-
-	failed := toolCallResult(agent.ToolResult{ToolCallID: "tc2", Status: tools.StatusError, Output: "boom"})
-	if failed.Status != ToolStatusFailed {
-		t.Fatalf("error result should be failed, got %q", failed.Status)
-	}
-}
-
-func TestPlanUpdateAndStatus(t *testing.T) {
-	upd := planUpdate([]tools.PlanItem{
-		{Content: "step a", Status: "completed"},
-		{Content: "step b", Status: "in_progress"},
-		{Content: "step c", Status: "failed"},
-		{Content: "step d", Status: "weird"},
-	})
-	if upd.SessionUpdate != UpdatePlan || len(upd.Entries) != 4 {
-		t.Fatalf("unexpected plan: %+v", upd)
-	}
-	want := []string{PlanStatusCompleted, PlanStatusInProgress, PlanStatusCompleted, PlanStatusPending}
-	for i, w := range want {
-		if upd.Entries[i].Status != w {
-			t.Errorf("entry %d status = %q, want %q", i, upd.Entries[i].Status, w)
-		}
-		if upd.Entries[i].Priority != PlanPriorityMedium {
-			t.Errorf("entry %d priority = %q", i, upd.Entries[i].Priority)
-		}
-	}
-}
-
-func TestPromptText(t *testing.T) {
-	got := promptText([]ContentBlock{
-		TextBlock("hello "),
-		ImageBlock("base64", "image/png"),
-		TextBlock("world"),
-	})
-	if got != "hello world" {
-		t.Fatalf("promptText = %q", got)
-	}
-}
-
-func TestToolTitleTruncateHintRuneSafe(t *testing.T) {
-	// A 61-character string containing multi-byte UTF-8 runes (emojis / CJK characters).
-	// We want to verify that it is truncated without cutting any runes or producing invalid UTF-8.
-	longPath := "📁/项目/非常长的路径名称/测试/🚀/emoji-and-cjk-characters-which-are-very-long-and-exceed-sixty-characters"
-
-	// Create JSON args for read_file
-	rawArgs := `{"path":"` + longPath + `"}`
-	got := toolTitle("read_file", rawArgs)
-
-	expectedPrefix := "read_file "
-	if !strings.HasPrefix(got, expectedPrefix) {
-		t.Fatalf("expected title to start with %q, got %q", expectedPrefix, got)
-	}
-
-	hint := strings.TrimPrefix(got, expectedPrefix)
-	// Hint should end with the ellipsis character
-	if !strings.HasSuffix(hint, "…") {
-		t.Fatalf("expected truncated hint to end with ellipsis, got %q", hint)
-	}
-
-	// Check that we don't have invalid UTF-8 runes
-	if !utf8.ValidString(hint) {
-		t.Fatalf("truncated hint is not a valid UTF-8 string: %q", hint)
-	}
-
-	// The rune count of the hint (excluding ellipsis) should be exactly 60
-	runes := []rune(strings.TrimSuffix(hint, "…"))
-	if len(runes) != 60 {
-		t.Fatalf("expected exactly 60 runes before ellipsis, got %d (hint: %q)", len(runes), hint)
-	}
-}

internal/acp/types.go

@@ -1,378 +0,0 @@
-package acp
-
-import "encoding/json"
-
-// ProtocolVersion is the ACP protocol version ZERO speaks. Wire compatibility is
-// negotiated during initialize; v1 is the current stable version.
-const ProtocolVersion = 1
-
-// Method names exactly as they appear on the wire (public ACP spec).
-const (
-	MethodInitialize             = "initialize"
-	MethodAuthenticate           = "authenticate"
-	MethodSessionNew             = "session/new"
-	MethodSessionLoad            = "session/load"
-	MethodSessionPrompt          = "session/prompt"
-	MethodSessionCancel          = "session/cancel" // notification
-	MethodSessionUpdate          = "session/update" // notification (agent -> client)
-	MethodSessionSetMode         = "session/set_mode"
-	MethodSessionSetConfigOption = "session/set_config_option"
-	MethodSessionRequestPerm     = "session/request_permission" // agent -> client
-	MethodFSReadTextFile         = "fs/read_text_file"          // agent -> client
-	MethodFSWriteTextFile        = "fs/write_text_file"         // agent -> client
-
-	// Vendor-prefixed ZERO extensions (clients that don't support them ignore the
-	// method and degrade cleanly, per the spec's _-prefixed convention).
-	MethodZeroSetModel = "_zero/set_model"
-)
-
-// SessionUpdate discriminator values (the "sessionUpdate" field).
-const (
-	UpdateAgentMessageChunk = "agent_message_chunk"
-	UpdateAgentThoughtChunk = "agent_thought_chunk"
-	UpdateUserMessageChunk  = "user_message_chunk"
-	UpdateToolCall          = "tool_call"
-	UpdateToolCallUpdate    = "tool_call_update"
-	UpdatePlan              = "plan"
-	UpdateAvailableCommands = "available_commands_update"
-	UpdateCurrentMode       = "current_mode_update"
-)
-
-// ---- initialize ----
-
-type Implementation struct {
-	Name    string `json:"name"`
-	Version string `json:"version,omitempty"`
-}
-
-type FileSystemCapabilities struct {
-	ReadTextFile  bool `json:"readTextFile"`
-	WriteTextFile bool `json:"writeTextFile"`
-}
-
-type ClientCapabilities struct {
-	FS       FileSystemCapabilities `json:"fs"`
-	Terminal bool                   `json:"terminal"`
-}
-
-type PromptCapabilities struct {
-	Image           bool `json:"image"`
-	Audio           bool `json:"audio"`
-	EmbeddedContext bool `json:"embeddedContext"`
-}
-
-type AgentCapabilities struct {
-	LoadSession        bool               `json:"loadSession"`
-	PromptCapabilities PromptCapabilities `json:"promptCapabilities"`
-}
-
-type AuthMethod struct {
-	ID          string `json:"id"`
-	Name        string `json:"name"`
-	Description string `json:"description,omitempty"`
-}
-
-type InitializeParams struct {
-	ProtocolVersion    int                `json:"protocolVersion"`
-	ClientCapabilities ClientCapabilities `json:"clientCapabilities"`
-	ClientInfo         *Implementation    `json:"clientInfo,omitempty"`
-}
-
-type InitializeResult struct {
-	ProtocolVersion   int               `json:"protocolVersion"`
-	AgentCapabilities AgentCapabilities `json:"agentCapabilities"`
-	AgentInfo         *Implementation   `json:"agentInfo,omitempty"`
-	AuthMethods       []AuthMethod      `json:"authMethods"`
-}
-
-// ---- content blocks ----
-
-// ContentBlock is the polymorphic content type. ZERO emits "text" (and "image"
-// on tool content); it parses "text", "image", and "resource"/"resource_link"
-// from inbound prompts. A single struct with omitempty fields covers both
-// directions since the field names do not collide across the variants ZERO uses.
-type ContentBlock struct {
-	Type     string          `json:"type"`
-	Text     string          `json:"text,omitempty"`
-	Data     string          `json:"data,omitempty"`     // image/audio: base64
-	MimeType string          `json:"mimeType,omitempty"` // image/audio
-	URI      string          `json:"uri,omitempty"`      // resource_link
-	Name     string          `json:"name,omitempty"`     // resource_link
-	Resource json.RawMessage `json:"resource,omitempty"` // embedded resource
-}
-
-func TextBlock(text string) ContentBlock { return ContentBlock{Type: "text", Text: text} }
-
-func ImageBlock(base64Data, mimeType string) ContentBlock {
-	return ContentBlock{Type: "image", Data: base64Data, MimeType: mimeType}
-}
-
-// ---- sessions ----
-
-// McpServer mirrors the editor-provided MCP server entry. ZERO owns its own MCP
-// configuration (BYOK), so these are accepted for spec compliance; ZERO's
-// configured servers remain authoritative.
-type McpServer struct {
-	Name    string          `json:"name"`
-	Command string          `json:"command,omitempty"`
-	Args    []string        `json:"args,omitempty"`
-	Env     json.RawMessage `json:"env,omitempty"`
-	URL     string          `json:"url,omitempty"`
-}
-
-type NewSessionParams struct {
-	Cwd                   string      `json:"cwd"`
-	McpServers            []McpServer `json:"mcpServers"`
-	AdditionalDirectories []string    `json:"additionalDirectories,omitempty"`
-}
-
-type NewSessionResult struct {
-	SessionID     string                `json:"sessionId"`
-	ConfigOptions []SessionConfigOption `json:"configOptions,omitempty"`
-	Modes         *SessionModeState     `json:"modes,omitempty"`
-}
-
-type LoadSessionParams struct {
-	SessionID             string      `json:"sessionId"`
-	Cwd                   string      `json:"cwd"`
-	McpServers            []McpServer `json:"mcpServers"`
-	AdditionalDirectories []string    `json:"additionalDirectories,omitempty"`
-}
-
-type LoadSessionResult struct {
-	ConfigOptions []SessionConfigOption `json:"configOptions,omitempty"`
-	Modes         *SessionModeState     `json:"modes,omitempty"`
-}
-
-// ---- prompt turn ----
-
-type PromptParams struct {
-	SessionID string         `json:"sessionId"`
-	Prompt    []ContentBlock `json:"prompt"`
-}
-
-// StopReason values (why a prompt turn ended).
-const (
-	StopEndTurn   = "end_turn"
-	StopMaxTokens = "max_tokens"
-	StopRefusal   = "refusal"
-	StopCancelled = "cancelled"
-)
-
-type PromptResult struct {
-	StopReason string `json:"stopReason"`
-}
-
-type CancelParams struct {
-	SessionID string `json:"sessionId"`
-}
-
-// ---- session/update notification ----
-
-type SessionNotification struct {
-	SessionID string `json:"sessionId"`
-	Update    any    `json:"update"`
-}
-
-// AgentMessageChunk / AgentThoughtChunk / UserMessageChunk all carry a single
-// ContentBlock under "content"; the variant is set via SessionUpdate.
-type ContentChunk struct {
-	SessionUpdate string       `json:"sessionUpdate"`
-	Content       ContentBlock `json:"content"`
-}
-
-// ToolKind classifies a tool call for client rendering.
-const (
-	ToolKindRead    = "read"
-	ToolKindEdit    = "edit"
-	ToolKindDelete  = "delete"
-	ToolKindMove    = "move"
-	ToolKindSearch  = "search"
-	ToolKindExecute = "execute"
-	ToolKindThink   = "think"
-	ToolKindFetch   = "fetch"
-	ToolKindOther   = "other"
-)
-
-// ToolCallStatus values.
-const (
-	ToolStatusPending    = "pending"
-	ToolStatusInProgress = "in_progress"
-	ToolStatusCompleted  = "completed"
-	ToolStatusFailed     = "failed"
-)
-
-// ToolCallUpdate is used for both the initial "tool_call" and subsequent
-// "tool_call_update" notifications (distinguished by SessionUpdate). It also
-// appears inside session/request_permission.
-type ToolCallUpdate struct {
-	SessionUpdate string             `json:"sessionUpdate,omitempty"`
-	ToolCallID    string             `json:"toolCallId"`
-	Title         string             `json:"title,omitempty"`
-	Kind          string             `json:"kind,omitempty"`
-	Status        string             `json:"status,omitempty"`
-	RawInput      json.RawMessage    `json:"rawInput,omitempty"`
-	Content       []ToolCallContent  `json:"content,omitempty"`
-	Locations     []ToolCallLocation `json:"locations,omitempty"`
-}
-
-// ToolCallContent is a tool call's rendered output. ZERO emits "content" (a
-// text/image block) and "diff" (a file change); "terminal" is part of the spec
-// but unused because ZERO executes locally.
-type ToolCallContent struct {
-	Type string `json:"type"`
-	// type == "content"
-	Content *ContentBlock `json:"content,omitempty"`
-	// type == "diff"
-	Path    string `json:"path,omitempty"`
-	OldText string `json:"oldText,omitempty"`
-	NewText string `json:"newText,omitempty"`
-}
-
-func ToolContent(block ContentBlock) ToolCallContent {
-	return ToolCallContent{Type: "content", Content: &block}
-}
-
-func ToolDiff(path, oldText, newText string) ToolCallContent {
-	return ToolCallContent{Type: "diff", Path: path, OldText: oldText, NewText: newText}
-}
-
-type ToolCallLocation struct {
-	Path string `json:"path"`
-	Line *int   `json:"line,omitempty"`
-}
-
-// ---- plan ----
-
-const (
-	PlanStatusPending    = "pending"
-	PlanStatusInProgress = "in_progress"
-	PlanStatusCompleted  = "completed"
-
-	PlanPriorityHigh   = "high"
-	PlanPriorityMedium = "medium"
-	PlanPriorityLow    = "low"
-)
-
-type PlanEntry struct {
-	Content  string `json:"content"`
-	Priority string `json:"priority"`
-	Status   string `json:"status"`
-}
-
-type PlanUpdate struct {
-	SessionUpdate string      `json:"sessionUpdate"`
-	Entries       []PlanEntry `json:"entries"`
-}
-
-type AvailableCommand struct {
-	Name        string `json:"name"`
-	Description string `json:"description"`
-}
-
-type AvailableCommandsUpdate struct {
-	SessionUpdate     string             `json:"sessionUpdate"`
-	AvailableCommands []AvailableCommand `json:"availableCommands"`
-}
-
-type CurrentModeUpdate struct {
-	SessionUpdate string `json:"sessionUpdate"`
-	CurrentModeID string `json:"currentModeId"`
-}
-
-// ---- permissions ----
-
-const (
-	PermAllowOnce    = "allow_once"
-	PermAllowAlways  = "allow_always"
-	PermRejectOnce   = "reject_once"
-	PermRejectAlways = "reject_always"
-)
-
-type PermissionOption struct {
-	OptionID string `json:"optionId"`
-	Name     string `json:"name"`
-	Kind     string `json:"kind"`
-}
-
-type RequestPermissionParams struct {
-	SessionID string             `json:"sessionId"`
-	ToolCall  ToolCallUpdate     `json:"toolCall"`
-	Options   []PermissionOption `json:"options"`
-}
-
-// RequestPermissionOutcome is a tagged union: {"outcome":"cancelled"} or
-// {"outcome":"selected","optionId":"..."}.
-type RequestPermissionOutcome struct {
-	Outcome  string `json:"outcome"`
-	OptionID string `json:"optionId,omitempty"`
-}
-
-const (
-	OutcomeSelected  = "selected"
-	OutcomeCancelled = "cancelled"
-)
-
-type RequestPermissionResult struct {
-	Outcome RequestPermissionOutcome `json:"outcome"`
-}
-
-// ---- session modes ----
-
-type SessionMode struct {
-	ID          string `json:"id"`
-	Name        string `json:"name"`
-	Description string `json:"description,omitempty"`
-}
-
-type SessionModeState struct {
-	CurrentModeID  string        `json:"currentModeId"`
-	AvailableModes []SessionMode `json:"availableModes"`
-}
-
-type SetSessionModeParams struct {
-	SessionID string `json:"sessionId"`
-	ModeID    string `json:"modeId"`
-}
-
-type SetSessionModeResult struct{}
-
-// ---- session config options (model selection) ----
-
-type SessionConfigOptionValue struct {
-	ID   string `json:"id"`
-	Name string `json:"name"`
-}
-
-type SessionConfigOption struct {
-	ID          string                     `json:"id"`
-	Name        string                     `json:"name"`
-	Description string                     `json:"description,omitempty"`
-	Value       string                     `json:"value"`
-	Values      []SessionConfigOptionValue `json:"values,omitempty"`
-}
-
-type SetSessionConfigOptionParams struct {
-	SessionID string `json:"sessionId"`
-	ConfigID  string `json:"configId"`
-	Value     string `json:"value"`
-}
-
-type SetSessionConfigOptionResult struct {
-	ConfigOptions []SessionConfigOption `json:"configOptions"`
-}
-
-// ---- vendor: _zero/set_model ----
-
-type ZeroSetModelParams struct {
-	SessionID string `json:"sessionId"`
-	Model     string `json:"model"`
-}
-
-type ZeroSetModelResult struct {
-	Model string `json:"model"`
-}
-
-// configIDModel is the SessionConfigOption id ZERO uses to expose model choice
-// through the standard session/set_config_option method.
-const configIDModel = "model"

internal/agent/acceptance_gate_test.go

@@ -1,248 +0,0 @@
-package agent
-
-import (
-	"context"
-	"strings"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/tools"
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-func enabledSelfCorrector(t *testing.T) *SelfCorrector {
-	t.Helper()
-	// checker/verifier nil → no LSP / project-test pass runs; we only need a
-	// non-nil SelfCorrector to enable the task-grounded acceptance gate.
-	return NewSelfCorrector(t.TempDir(), nil, nil, SelfCorrectConfig{Enabled: true})
-}
-
-// BUG #2(a): a final message that admits the model guessed / could not meet the
-// objective must be reported as INCOMPLETE, never success — and immediately
-// (an admitted guess is not worth re-prompting).
-func TestAcceptanceSelfReportAdmissionIsIncomplete(t *testing.T) {
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		textTurn("I couldn't analyze the board image, so I wrote the common opening move e2e4 as my best guess."),
-	}}
-
-	result, err := Run(context.Background(), "write the best chess move", provider, Options{
-		Registry:                tools.NewRegistry(),
-		MaxTurns:                10,
-		RequireCompletionSignal: true,
-		SelfCorrect:             enabledSelfCorrector(t),
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if !result.Incomplete {
-		t.Fatalf("admitted guess must be Incomplete, got success; final=%q", result.FinalAnswer)
-	}
-	if !strings.Contains(result.IncompleteReason, "objective was not met") {
-		t.Fatalf("IncompleteReason = %q, want it to cite the admission", result.IncompleteReason)
-	}
-	if len(provider.requests) != 1 {
-		t.Fatalf("an admission should downgrade immediately (1 turn), got %d", len(provider.requests))
-	}
-}
-
-// BUG #2(a): the inability detector must generalize over the verb the model uses
-// (so re-phrasing can't defeat it) while NOT misreading success-y negations
-// ("could not find any issues") as admissions.
-func TestSelfReportedIncompletionMatching(t *testing.T) {
-	cases := []struct {
-		text  string
-		admit bool
-	}{
-		// real chess re-run wording that slipped a fixed phrase list:
-		{"…which I cannot do without proper image analysis capabilities.", true},
-		{"I cannot visually analyze the chess board image.", true},
-		{"I could not recover the original values, so this is my best guess.", true},
-		{"I was unable to complete the optimization.", true},
-		{"The result may not be correct.", true},
-		// multi-occurrence: an early guarded use ("could not find any") must NOT mask a
-		// later genuine admission with the same stem (review #5):
-		{"I could not find any examples in the codebase, so I could not implement the feature.", true},
-		// genuine successes that must NOT be flagged:
-		{"I implemented the function and all tests pass.", false},
-		{"I verified the fix; I could not find any remaining issues.", false},
-		{"I cannot reproduce the bug, so the fix holds.", false},
-		{"Done — the output matches the required format and values.", false},
-		// behavior descriptions that must NOT be flagged — these phrases describe
-		// implemented behavior, not an admission (review #1, removed from the list):
-		{"The parser will fall back to UTF-8 when no encoding is set.", false},
-		{"I replaced the placeholder value with the computed key.", false},
-		{"The retry uses exponential backoff as a fallback.", false},
-		{"Without proper error handling the function would crash, so I added it.", false},
-		// real false positive (conversational recap): retelling a PAST exchange is
-		// not an admission about the current objective — the sentence carries its
-		// own narrative context ("You asked …", "at the time"):
-		{"You asked if I could work autonomously on zero if you gave me a GitHub account — I was honest that my sandbox had no repo, no build system, and no network/integration wired up yet, so I couldn't actually do it at the time.", false},
-		{"When we last spoke, I couldn't reach the GitHub API from the sandbox.", false},
-		{"Last session I was unable to build the project; that is what we should retry now.", false},
-		// quoted admissions (the model citing its own earlier message or a log
-		// line) must NOT be flagged:
-		{`My earlier reply said "I couldn't do it" but the tooling is now installed.`, false},
-		{"The run log contains `I cannot access the network`, which is expected inside the sandbox.", false},
-		// current-objective admissions must STILL be flagged, including past tense
-		// and multi-sentence finals whose other sentences are narrative:
-		{"I couldn't complete the refactor.", true},
-		{"I hit an error I could not resolve, so the migration is unfinished.", true},
-		{"You asked me to port the parser. I couldn't finish the second half.", true},
-		// unbalanced delimiters must not swallow the tail (only BALANCED quoted
-		// spans are stripped), so an admission after a stray quote still fires:
-		{"The log ends with a stray \" and I couldn't finish the migration.", true},
-		{"There is an unterminated `code span and I cannot proceed with the fix.", true},
-		{"The doc opens with “a curly quote and I was unable to complete the task.", true},
-	}
-	for _, c := range cases {
-		got := selfReportedIncompletion(c.text) != ""
-		if got != c.admit {
-			t.Errorf("selfReportedIncompletion(%q) admitted=%v, want %v", c.text, got, c.admit)
-		}
-	}
-}
-
-// BUG #2(a): a "cannot analyze / cannot determine" admission (present tense) must
-// be caught even when the model also stamps a bogus "PASS" based on a shape check.
-// Mirrors the chess-best-move final ("I cannot analyze the position… PASS: I wrote
-// the move and verified by reading it back").
-func TestAcceptanceCannotPhrasingIsIncomplete(t *testing.T) {
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		textTurn("I cannot analyze the specific position from the image, so I cannot determine the best move. PASS: I wrote e2e4 to the file and verified it by reading it back."),
-	}}
-
-	result, err := Run(context.Background(), "write the best move", provider, Options{
-		Registry:                tools.NewRegistry(),
-		MaxTurns:                10,
-		RequireCompletionSignal: true,
-		SelfCorrect:             enabledSelfCorrector(t),
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if !result.Incomplete {
-		t.Fatalf("a 'cannot analyze/determine' admission must be Incomplete despite the bogus PASS; final=%q", result.FinalAnswer)
-	}
-}
-
-// A pending/in_progress update_plan item is an AMBIGUOUS signal (the model may have
-// finished without re-marking the step), so it must NOT by itself force INCOMPLETE.
-// The gate nudges, but if the model keeps claiming completion with no continuation
-// cue and no admission, the run finalizes as SUCCESS — trusting the completion claim
-// over stale plan bookkeeping. (Only a continuation cue or self-report admission
-// forces INCOMPLETE; see TestCompletionGate* and TestAcceptance*Admission*.)
-func TestPendingPlanAloneDoesNotForceIncomplete(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewUpdatePlanTool())
-	done := "All set." // confident; no cue, no admission — only the plan is stale
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		planTurn("completed", "in_progress"),
-		textTurn(done), textTurn(done), textTurn(done), textTurn(done), textTurn(done),
-	}}
-
-	result, err := Run(context.Background(), "do the multi-step task", provider, Options{
-		Registry:                registry,
-		MaxTurns:                10,
-		RequireCompletionSignal: true,
-		SelfCorrect:             enabledSelfCorrector(t),
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.Incomplete {
-		t.Fatalf("pending-plan alone must NOT force Incomplete (stale bookkeeping != unfinished); reason=%q", result.IncompleteReason)
-	}
-	if result.FinalAnswer != done {
-		t.Fatalf("final answer = %q, want the completion claim %q", result.FinalAnswer, done)
-	}
-	if !someRequestContains(provider.requests, continueNudgeMarker) {
-		t.Fatalf("expected at least one continue-nudge before accepting completion")
-	}
-}
-
-// BUG #2(b): when self-correct is on, the run must pass a one-time task-grounded
-// acceptance check before success. A model that, when challenged, confirms its
-// work meets the criterion (no admission) then finalizes as success — no regression.
-func TestAcceptanceGroundedCompletionSucceeds(t *testing.T) {
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		textTurn("I implemented the function."),
-		textTurn("PASS. I ran the verification and it passes; the output meets the task requirement."),
-	}}
-
-	result, err := Run(context.Background(), "implement the function", provider, Options{
-		Registry:                tools.NewRegistry(),
-		MaxTurns:                10,
-		RequireCompletionSignal: true,
-		SelfCorrect:             enabledSelfCorrector(t),
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.Incomplete {
-		t.Fatalf("a verified completion must succeed, got Incomplete (%q)", result.IncompleteReason)
-	}
-	if result.FinalAnswer != "PASS. I ran the verification and it passes; the output meets the task requirement." {
-		t.Fatalf("final answer = %q", result.FinalAnswer)
-	}
-	// Exactly one acceptance pass: the first "done" turn was challenged, the second confirmed.
-	if len(provider.requests) != 2 {
-		t.Fatalf("expected 2 turns (1 acceptance challenge + 1 confirmation), got %d", len(provider.requests))
-	}
-	if !someRequestContains(provider.requests, acceptanceNudgeMarker) {
-		t.Fatalf("expected the task-grounded acceptance check to be demanded")
-	}
-}
-
-// With the gate OFF (interactive/TUI default), even an admission is accepted as the
-// final answer exactly as before — proving no behavior change for non-headless callers.
-func TestAcceptanceGateOffLeavesAdmissionAsSuccess(t *testing.T) {
-	admission := "I couldn't determine the value, so this is a guess."
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		textTurn(admission),
-	}}
-
-	result, err := Run(context.Background(), "do a thing", provider, Options{
-		Registry: tools.NewRegistry(),
-		MaxTurns: 10,
-		// RequireCompletionSignal false, SelfCorrect nil — the legacy path.
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.Incomplete {
-		t.Fatalf("legacy path must never set Incomplete; final=%q", result.FinalAnswer)
-	}
-	if result.FinalAnswer != admission {
-		t.Fatalf("final answer = %q, want the text verbatim (legacy behavior)", result.FinalAnswer)
-	}
-	if len(provider.requests) != 1 {
-		t.Fatalf("legacy path must not re-prompt; got %d turns", len(provider.requests))
-	}
-}
-
-// The acceptance check is scoped to runs with self-correct on: a headless run
-// WITHOUT a SelfCorrector accepts a confident completion immediately (the (b) gate
-// is inert), so enabling RequireCompletionSignal alone adds no acceptance turn.
-func TestAcceptanceGateRequiresSelfCorrect(t *testing.T) {
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		textTurn("Done. Implemented and it works."),
-	}}
-
-	result, err := Run(context.Background(), "implement it", provider, Options{
-		Registry:                tools.NewRegistry(),
-		MaxTurns:                10,
-		RequireCompletionSignal: true,
-		// SelfCorrect deliberately nil → acceptance gate (b) must not fire.
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.Incomplete {
-		t.Fatalf("confident completion without self-correct must succeed; reason=%q", result.IncompleteReason)
-	}
-	if len(provider.requests) != 1 {
-		t.Fatalf("acceptance gate must not fire without self-correct; got %d turns", len(provider.requests))
-	}
-	if someRequestContains(provider.requests, acceptanceNudgeMarker) {
-		t.Fatalf("acceptance nudge must not be injected when self-correct is off")
-	}
-}

internal/agent/additional_permissions_validation_test.go

@@ -1,146 +0,0 @@
-package agent
-
-import (
-	"context"
-	"strings"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/tools"
-)
-
-// A shell call that sets sandbox_permissions: with_additional_permissions but
-// omits (or malforms) additional_permissions can never succeed: the same
-// validation runs again at grant time regardless of the user's decision. It
-// must be rejected as a plain tool error BEFORE any permission prompt, not
-// surfaced as a confusing "approved but denied" prompt outcome.
-func TestExecuteToolCallRejectsMissingAdditionalPermissionsBeforePrompt(t *testing.T) {
-	root := t.TempDir()
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewBashTool(root))
-
-	promptCalls := 0
-	result, err := executeToolCall(
-		context.Background(),
-		registry,
-		ToolCall{
-			ID:        "c1",
-			Name:      "bash",
-			Arguments: `{"command":"echo hi","sandbox_permissions":"with_additional_permissions"}`,
-		},
-		PermissionModeAuto,
-		Options{
-			OnPermissionRequest: func(ctx context.Context, request PermissionRequest) (PermissionDecision, error) {
-				promptCalls++
-				return PermissionDecision{Action: PermissionDecisionAllow}, nil
-			},
-		},
-	)
-	if err != nil {
-		t.Fatalf("executeToolCall: %v", err)
-	}
-	if result.Status != tools.StatusError {
-		t.Fatalf("status = %q, want error", result.Status)
-	}
-	if promptCalls != 0 {
-		t.Fatalf("OnPermissionRequest called %d times, want 0 (must reject before prompting)", promptCalls)
-	}
-}
-
-// The mismatched-flag case (additional_permissions present without the
-// sandbox_permissions flag) must be rejected the same way.
-func TestExecuteToolCallRejectsAdditionalPermissionsWithoutFlagBeforePrompt(t *testing.T) {
-	root := t.TempDir()
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewBashTool(root))
-
-	promptCalls := 0
-	result, err := executeToolCall(
-		context.Background(),
-		registry,
-		ToolCall{
-			ID:        "c1",
-			Name:      "bash",
-			Arguments: `{"command":"echo hi","additional_permissions":{"network":{"enabled":true}}}`,
-		},
-		PermissionModeAuto,
-		Options{
-			OnPermissionRequest: func(ctx context.Context, request PermissionRequest) (PermissionDecision, error) {
-				promptCalls++
-				return PermissionDecision{Action: PermissionDecisionAllow}, nil
-			},
-		},
-	)
-	if err != nil {
-		t.Fatalf("executeToolCall: %v", err)
-	}
-	if result.Status != tools.StatusError {
-		t.Fatalf("status = %q, want error", result.Status)
-	}
-	if promptCalls != 0 {
-		t.Fatalf("OnPermissionRequest called %d times, want 0 (must reject before prompting)", promptCalls)
-	}
-}
-
-// The rejection message must show a concrete, correctly-shaped example: a
-// model that gets this wrong once (attaching sandbox_permissions out of habit
-// from an earlier call, without a valid payload) needs enough in the error to
-// self-correct on retry instead of repeating the identical mistake.
-func TestExecuteToolCallMissingAdditionalPermissionsErrorIsActionable(t *testing.T) {
-	root := t.TempDir()
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewBashTool(root))
-
-	result, err := executeToolCall(
-		context.Background(),
-		registry,
-		ToolCall{
-			ID:        "c1",
-			Name:      "bash",
-			Arguments: `{"command":"ping example.com","sandbox_permissions":"with_additional_permissions"}`,
-		},
-		PermissionModeAuto,
-		Options{},
-	)
-	if err != nil {
-		t.Fatalf("executeToolCall: %v", err)
-	}
-	if !strings.Contains(result.Output, `{"network": {"enabled": true}}`) {
-		t.Fatalf("output = %q, want a concrete additional_permissions example", result.Output)
-	}
-	if !strings.Contains(result.Output, "omit sandbox_permissions") {
-		t.Fatalf("output = %q, want guidance that the flag can simply be omitted", result.Output)
-	}
-}
-
-// A well-formed additional_permissions payload must still reach the normal
-// permission prompt: this fix only rejects malformed payloads.
-func TestExecuteToolCallStillPromptsForValidAdditionalPermissions(t *testing.T) {
-	root := t.TempDir()
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewBashTool(root))
-
-	promptCalls := 0
-	_, err := executeToolCall(
-		context.Background(),
-		registry,
-		ToolCall{
-			ID:   "c1",
-			Name: "bash",
-			Arguments: `{"command":"echo hi","sandbox_permissions":"with_additional_permissions",` +
-				`"additional_permissions":{"network":{"enabled":true}}}`,
-		},
-		PermissionModeAuto,
-		Options{
-			OnPermissionRequest: func(ctx context.Context, request PermissionRequest) (PermissionDecision, error) {
-				promptCalls++
-				return PermissionDecision{Action: PermissionDecisionDeny}, nil
-			},
-		},
-	)
-	if err != nil {
-		t.Fatalf("executeToolCall: %v", err)
-	}
-	if promptCalls != 1 {
-		t.Fatalf("OnPermissionRequest called %d times, want exactly 1 for a valid elevation request", promptCalls)
-	}
-}

internal/agent/ask_user_test.go

@@ -137,42 +137,6 @@ func TestRunAskUserHandlerErrorDegradesGracefully(t *testing.T) {
 	}
 }
 
-func TestRunAskUserCancellationAbortsRun(t *testing.T) {
-	registry := registryWithAskUser()
-	args := `{"questions":[{"question":"Which framework?"}]}`
-	provider := providerCallingAskUserThenAnswer(args, "done")
-
-	result, err := Run(context.Background(), "clarify", provider, Options{
-		Registry: registry,
-		OnAskUser: func(_ context.Context, _ AskUserRequest) (AskUserResponse, error) {
-			return AskUserResponse{}, context.Canceled
-		},
-	})
-
-	// A canceled prompt must abort with that error — NOT fabricate a headless
-	// answer and run on to a final answer.
-	if !errors.Is(err, context.Canceled) {
-		t.Fatalf("expected run to abort with context.Canceled, got err=%v answer=%q", err, result.FinalAnswer)
-	}
-	// The model must not be asked for a follow-up turn after cancellation.
-	if len(provider.requests) != 1 {
-		t.Fatalf("expected the run to stop after the canceled ask_user (1 turn), got %d", len(provider.requests))
-	}
-	// The recorded tool result must reflect cancellation, not a synthetic answer.
-	var toolMsg string
-	for _, m := range result.Messages {
-		if m.Role == zeroruntime.MessageRoleTool {
-			toolMsg = m.Content
-		}
-	}
-	if !strings.Contains(strings.ToLower(toolMsg), "canceled") {
-		t.Fatalf("expected a canceled tool result, got %q", toolMsg)
-	}
-	if strings.Contains(strings.ToLower(toolMsg), "no interactive user") {
-		t.Fatalf("must not fabricate a headless answer on cancellation, got %q", toolMsg)
-	}
-}
-
 func TestRunAskUserRedactsSecretsInAnswers(t *testing.T) {
 	registry := registryWithAskUser()
 	secret := "sk-ant-api03-ABCDEFGHIJKLMNOP1234567890"

internal/agent/async_diagnostics.go

@@ -1,185 +0,0 @@
-package agent
-
-import (
-	"context"
-	"path/filepath"
-	"sort"
-	"strings"
-	"sync"
-	"time"
-)
-
-// Post-edit LSP diagnostics used to run synchronously inside edit_file /
-// write_file: every edit blocked its tool result on a fresh-publish wait with
-// a quiet-period debounce (≥300ms floor that reset on each publish, 10s cap) —
-// the largest recurring latency tax in an edit-heavy session. Collection is
-// now asynchronous: a file is enqueued the moment a mutating tool finishes, a
-// single worker checks it while the rest of the tool batch (and any
-// self-correct verification) runs, and the loop drains completed results just
-// before building the NEXT provider request, appending errors as a user-role
-// nudge. The model always issues another request after an edit turn — it has
-// to read its tool results — so it still sees introduced errors at the same
-// decision point as before, without any tool call stalling on the language
-// server.
-//
-// A file re-edited before its check runs is checked once in its final state
-// (the checker re-reads the file); a file re-edited after its check completed
-// is re-enqueued and the newer result wins. When the worker is still busy at
-// drain time, the drain gives up after a short wait and the results are
-// delivered on the following turn instead — they stay accurate because any
-// further edit re-enqueues the file.
-
-// asyncDiagnosticsDrainTimeout bounds how long a turn waits on the in-flight
-// check before deferring delivery to the next turn. Most checks finish well
-// under it (debounce floor + analysis); a cold language server on a large
-// repo does not get to stall the loop the way the old inline 10s cap could.
-// A var so tests can shorten the wait.
-var asyncDiagnosticsDrainTimeout = 3 * time.Second
-
-// asyncDiagnosticsFinalDrainTimeout is the wait used when the run is about to
-// FINALIZE (natural completion or the max-turns summary): there is no later
-// turn to defer to, so the gate waits out the old inline-era budget rather
-// than silently dropping an error the last edit introduced. A var so tests
-// can shorten the wait.
-var asyncDiagnosticsFinalDrainTimeout = 10 * time.Second
-
-// asyncDiagnosticsNudge prefixes the drained diagnostics blocks; phrased like
-// the old inline block so the model reacts the same way.
-const asyncDiagnosticsNudge = "Diagnostics after your recent edits (fix any errors you introduced):\n"
-
-type asyncDiagnostics struct {
-	check         func(context.Context, string) string
-	workspaceRoot string
-
-	mu      sync.Mutex
-	queue   []string
-	queued  map[string]bool
-	results map[string]string
-	// working is non-nil while the worker goroutine runs and is closed when it
-	// exits, so drain can wait for quiescence without polling.
-	working chan struct{}
-}
-
-// newAsyncDiagnostics returns nil when check is nil, and every method on a nil
-// receiver is a no-op — callers wire it unconditionally.
-func newAsyncDiagnostics(check func(context.Context, string) string, workspaceRoot string) *asyncDiagnostics {
-	if check == nil {
-		return nil
-	}
-	return &asyncDiagnostics{
-		check:         check,
-		workspaceRoot: workspaceRoot,
-		queued:        map[string]bool{},
-		results:       map[string]string{},
-	}
-}
-
-// enqueue schedules changed files for a background check. Paths are
-// workspace-relative from Result.ChangedFiles (absolute when under a granted
-// extra write root), mirroring SelfCorrector's resolution.
-func (diagnostics *asyncDiagnostics) enqueue(ctx context.Context, files []string) {
-	if diagnostics == nil || len(files) == 0 {
-		return
-	}
-	diagnostics.mu.Lock()
-	defer diagnostics.mu.Unlock()
-	for _, file := range files {
-		path := diagnostics.absPath(file)
-		if diagnostics.queued[path] {
-			continue
-		}
-		diagnostics.queued[path] = true
-		// A completed result for this file describes a pre-edit state now.
-		delete(diagnostics.results, path)
-		diagnostics.queue = append(diagnostics.queue, path)
-	}
-	if diagnostics.working == nil && len(diagnostics.queue) > 0 {
-		done := make(chan struct{})
-		diagnostics.working = done
-		go diagnostics.work(ctx, done)
-	}
-}
-
-func (diagnostics *asyncDiagnostics) work(ctx context.Context, done chan struct{}) {
-	defer close(done)
-	for {
-		diagnostics.mu.Lock()
-		if len(diagnostics.queue) == 0 || ctx.Err() != nil {
-			diagnostics.working = nil
-			diagnostics.mu.Unlock()
-			return
-		}
-		path := diagnostics.queue[0]
-		diagnostics.queue = diagnostics.queue[1:]
-		// Un-mark before checking so an edit that lands DURING the check
-		// re-enqueues the file and it is checked again in its final state.
-		delete(diagnostics.queued, path)
-		diagnostics.mu.Unlock()
-
-		block := diagnostics.check(ctx, path)
-
-		diagnostics.mu.Lock()
-		if block != "" {
-			diagnostics.results[path] = block
-		} else {
-			delete(diagnostics.results, path)
-		}
-		diagnostics.mu.Unlock()
-	}
-}
-
-// drain waits briefly for the worker to go quiet, then formats and consumes
-// all completed error blocks as one nudge. Returns "" when there is nothing
-// to report or the worker is still busy (results then surface next turn).
-func (diagnostics *asyncDiagnostics) drain(ctx context.Context) string {
-	return diagnostics.drainWithin(ctx, asyncDiagnosticsDrainTimeout)
-}
-
-// drainFinal is drain with the finalization budget: called before the run's
-// last model request, where an undelivered result would otherwise be lost.
-func (diagnostics *asyncDiagnostics) drainFinal(ctx context.Context) string {
-	return diagnostics.drainWithin(ctx, asyncDiagnosticsFinalDrainTimeout)
-}
-
-func (diagnostics *asyncDiagnostics) drainWithin(ctx context.Context, timeout time.Duration) string {
-	if diagnostics == nil {
-		return ""
-	}
-	diagnostics.mu.Lock()
-	busy := diagnostics.working
-	diagnostics.mu.Unlock()
-	if busy != nil {
-		timer := time.NewTimer(timeout)
-		defer timer.Stop()
-		select {
-		case <-busy:
-		case <-timer.C:
-			return ""
-		case <-ctx.Done():
-			return ""
-		}
-	}
-	diagnostics.mu.Lock()
-	defer diagnostics.mu.Unlock()
-	if len(diagnostics.results) == 0 {
-		return ""
-	}
-	paths := make([]string, 0, len(diagnostics.results))
-	for path := range diagnostics.results {
-		paths = append(paths, path)
-	}
-	sort.Strings(paths)
-	blocks := make([]string, 0, len(paths))
-	for _, path := range paths {
-		blocks = append(blocks, diagnostics.results[path])
-	}
-	diagnostics.results = map[string]string{}
-	return asyncDiagnosticsNudge + strings.Join(blocks, "\n")
-}
-
-func (diagnostics *asyncDiagnostics) absPath(path string) string {
-	if filepath.IsAbs(path) {
-		return path
-	}
-	return filepath.Join(diagnostics.workspaceRoot, path)
-}

internal/agent/async_diagnostics_test.go

@@ -1,338 +0,0 @@
-package agent
-
-import (
-	"context"
-	"path/filepath"
-	"strings"
-	"sync"
-	"testing"
-	"time"
-
-	"github.com/Gitlawb/zero/internal/tools"
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-func TestAsyncDiagnosticsNilCollectorNoOps(t *testing.T) {
-	var diagnostics *asyncDiagnostics
-	if diagnostics != newAsyncDiagnostics(nil, "/ws") {
-		t.Fatal("nil check must yield a nil collector")
-	}
-	diagnostics.enqueue(context.Background(), []string{"a.go"})
-	if nudge := diagnostics.drain(context.Background()); nudge != "" {
-		t.Fatalf("nil collector drain = %q, want empty", nudge)
-	}
-}
-
-func TestAsyncDiagnosticsCollectsAndDrainsOnce(t *testing.T) {
-	// A real absolute root: a literal "/ws" is not filepath.IsAbs on Windows
-	// (no drive letter), which is exactly what the assertions below check.
-	root := t.TempDir()
-	var mu sync.Mutex
-	var checked []string
-	check := func(_ context.Context, absPath string) string {
-		mu.Lock()
-		checked = append(checked, absPath)
-		mu.Unlock()
-		if filepath.Base(absPath) == "clean.go" {
-			return ""
-		}
-		return "ERR " + filepath.Base(absPath)
-	}
-	diagnostics := newAsyncDiagnostics(check, root)
-	diagnostics.enqueue(context.Background(), []string{"broken.go", "clean.go"})
-
-	nudge := diagnostics.drain(context.Background())
-	if !strings.HasPrefix(nudge, asyncDiagnosticsNudge) {
-		t.Fatalf("nudge missing prefix: %q", nudge)
-	}
-	if !strings.Contains(nudge, "ERR broken.go") || strings.Contains(nudge, "clean.go") {
-		t.Fatalf("nudge = %q, want broken.go errors only", nudge)
-	}
-	mu.Lock()
-	for _, path := range checked {
-		if !filepath.IsAbs(path) || !strings.HasPrefix(path, root+string(filepath.Separator)) {
-			t.Fatalf("check received %q, want path under workspace root %q", path, root)
-		}
-	}
-	mu.Unlock()
-	if second := diagnostics.drain(context.Background()); second != "" {
-		t.Fatalf("second drain = %q, want empty (results are consumed)", second)
-	}
-}
-
-// A check still running at drain time defers delivery to the next drain
-// instead of blocking the turn.
-func TestAsyncDiagnosticsSlowCheckDefersToNextDrain(t *testing.T) {
-	previous := asyncDiagnosticsDrainTimeout
-	asyncDiagnosticsDrainTimeout = 20 * time.Millisecond
-	defer func() { asyncDiagnosticsDrainTimeout = previous }()
-
-	release := make(chan struct{})
-	check := func(_ context.Context, absPath string) string {
-		<-release
-		return "ERR " + filepath.Base(absPath)
-	}
-	diagnostics := newAsyncDiagnostics(check, "/ws")
-	diagnostics.enqueue(context.Background(), []string{"slow.go"})
-
-	start := time.Now()
-	if nudge := diagnostics.drain(context.Background()); nudge != "" {
-		t.Fatalf("drain during in-flight check = %q, want empty", nudge)
-	}
-	if elapsed := time.Since(start); elapsed > time.Second {
-		t.Fatalf("drain blocked %v, want bounded by the (shortened) timeout", elapsed)
-	}
-
-	close(release)
-	deadline := time.Now().Add(2 * time.Second)
-	for {
-		if nudge := diagnostics.drain(context.Background()); nudge != "" {
-			if !strings.Contains(nudge, "ERR slow.go") {
-				t.Fatalf("deferred nudge = %q", nudge)
-			}
-			return
-		}
-		if time.Now().After(deadline) {
-			t.Fatal("deferred diagnostics never delivered")
-		}
-		time.Sleep(5 * time.Millisecond)
-	}
-}
-
-// Re-enqueueing a file after its check completed replaces the stale result.
-func TestAsyncDiagnosticsReEditReplacesResult(t *testing.T) {
-	var mu sync.Mutex
-	response := "ERR old"
-	check := func(context.Context, string) string {
-		mu.Lock()
-		defer mu.Unlock()
-		return response
-	}
-	diagnostics := newAsyncDiagnostics(check, "/ws")
-
-	diagnostics.enqueue(context.Background(), []string{"a.go"})
-	waitForIdle(t, diagnostics)
-	mu.Lock()
-	response = "ERR new"
-	mu.Unlock()
-	diagnostics.enqueue(context.Background(), []string{"a.go"})
-
-	nudge := diagnostics.drain(context.Background())
-	if !strings.Contains(nudge, "ERR new") || strings.Contains(nudge, "ERR old") {
-		t.Fatalf("nudge = %q, want only the re-check result", nudge)
-	}
-}
-
-func waitForIdle(t *testing.T, diagnostics *asyncDiagnostics) {
-	t.Helper()
-	deadline := time.Now().Add(2 * time.Second)
-	for {
-		diagnostics.mu.Lock()
-		busy := diagnostics.working
-		diagnostics.mu.Unlock()
-		if busy == nil {
-			return
-		}
-		if time.Now().After(deadline) {
-			t.Fatal("worker never went idle")
-		}
-		time.Sleep(2 * time.Millisecond)
-	}
-}
-
-// changedFilesTool is a fake mutating tool reporting a changed file without
-// touching the filesystem.
-type changedFilesTool struct{}
-
-func (changedFilesTool) Name() string        { return "fake_edit" }
-func (changedFilesTool) Description() string { return "test mutating tool" }
-func (changedFilesTool) Parameters() tools.Schema {
-	return tools.Schema{Type: "object", Properties: map[string]tools.PropertySchema{}}
-}
-func (changedFilesTool) Safety() tools.Safety {
-	return tools.Safety{SideEffect: tools.SideEffectWrite, Permission: tools.PermissionAllow, Reason: "test"}
-}
-func (changedFilesTool) Run(context.Context, map[string]any) tools.Result {
-	return tools.Result{Status: tools.StatusOK, Output: "edited", ChangedFiles: []string{"main.go"}}
-}
-
-// End-to-end through Run: an edit turn does NOT carry diagnostics in the tool
-// result (the old inline path), but the next request contains the background
-// diagnostics nudge, before the model finalizes.
-func TestRunDeliversAsyncDiagnosticsNudgeNextTurn(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(changedFilesTool{})
-	provider := &mockProvider{
-		turns: [][]zeroruntime.StreamEvent{
-			{
-				{Type: zeroruntime.StreamEventToolCallStart, ToolCallID: "call-1", ToolName: "fake_edit"},
-				{Type: zeroruntime.StreamEventToolCallDelta, ToolCallID: "call-1", ArgumentsFragment: `{}`},
-				{Type: zeroruntime.StreamEventToolCallEnd, ToolCallID: "call-1"},
-				{Type: zeroruntime.StreamEventDone},
-			},
-			{
-				{Type: zeroruntime.StreamEventText, Content: "done"},
-				{Type: zeroruntime.StreamEventDone},
-			},
-		},
-	}
-	root := t.TempDir()
-	var checkedPath string
-	var mu sync.Mutex
-
-	result, err := Run(context.Background(), "fix main.go", provider, Options{
-		Registry: registry,
-		Cwd:      root,
-		FileDiagnostics: func(_ context.Context, absPath string) string {
-			mu.Lock()
-			checkedPath = absPath
-			mu.Unlock()
-			return "main.go:1:1 error: boom"
-		},
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.FinalAnswer != "done" {
-		t.Fatalf("final answer = %q", result.FinalAnswer)
-	}
-	if len(provider.requests) != 2 {
-		t.Fatalf("expected 2 provider turns, got %d", len(provider.requests))
-	}
-
-	// The tool result itself must NOT block on / embed diagnostics.
-	for _, message := range provider.requests[1].Messages {
-		if message.Role == zeroruntime.MessageRoleTool && strings.Contains(message.Content, "boom") {
-			t.Fatalf("diagnostics leaked into the tool result: %q", message.Content)
-		}
-	}
-	// The nudge must be present in the second request as a user message.
-	found := false
-	for _, message := range provider.requests[1].Messages {
-		if message.Role == zeroruntime.MessageRoleUser && strings.HasPrefix(message.Content, asyncDiagnosticsNudge) {
-			if !strings.Contains(message.Content, "boom") {
-				t.Fatalf("nudge missing diagnostics: %q", message.Content)
-			}
-			found = true
-		}
-	}
-	if !found {
-		t.Fatal("no async diagnostics nudge in the follow-up request")
-	}
-	mu.Lock()
-	defer mu.Unlock()
-	if want := filepath.Join(root, "main.go"); checkedPath != want {
-		t.Fatalf("checked path = %q, want %q", checkedPath, want)
-	}
-}
-
-// When the run hits the maxTurns ceiling right after an edit turn, the
-// final-answer request must still carry the diagnostics nudge — otherwise an
-// error introduced by the last edit would go unreported in the summary.
-func TestRunDrainsAsyncDiagnosticsBeforeMaxTurnsFinalAnswer(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(changedFilesTool{})
-	provider := &mockProvider{
-		turns: [][]zeroruntime.StreamEvent{
-			{
-				{Type: zeroruntime.StreamEventToolCallStart, ToolCallID: "call-1", ToolName: "fake_edit"},
-				{Type: zeroruntime.StreamEventToolCallDelta, ToolCallID: "call-1", ArgumentsFragment: `{}`},
-				{Type: zeroruntime.StreamEventToolCallEnd, ToolCallID: "call-1"},
-				{Type: zeroruntime.StreamEventDone},
-			},
-			{
-				{Type: zeroruntime.StreamEventText, Content: "summary"},
-				{Type: zeroruntime.StreamEventDone},
-			},
-		},
-	}
-
-	result, err := Run(context.Background(), "fix main.go", provider, Options{
-		Registry: registry,
-		Cwd:      t.TempDir(),
-		MaxTurns: 1,
-		FileDiagnostics: func(context.Context, string) string {
-			return "main.go:1:1 error: boom"
-		},
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.FinalAnswer != "summary" {
-		t.Fatalf("final answer = %q", result.FinalAnswer)
-	}
-	if len(provider.requests) != 2 {
-		t.Fatalf("expected the max-turns final-answer request, got %d requests", len(provider.requests))
-	}
-	found := false
-	for _, message := range provider.requests[1].Messages {
-		if message.Role == zeroruntime.MessageRoleUser && strings.HasPrefix(message.Content, asyncDiagnosticsNudge) {
-			found = true
-		}
-	}
-	if !found {
-		t.Fatal("max-turns final-answer request missing the diagnostics nudge")
-	}
-}
-
-// When the model finalizes while a slow check is still in flight (the
-// per-turn drain missed it), the finalization gate must wait it out and give
-// the model one more turn with the nudge instead of dropping the errors.
-func TestRunFinalizationGateDeliversLateDiagnostics(t *testing.T) {
-	previous := asyncDiagnosticsDrainTimeout
-	asyncDiagnosticsDrainTimeout = time.Millisecond
-	defer func() { asyncDiagnosticsDrainTimeout = previous }()
-
-	registry := tools.NewRegistry()
-	registry.Register(changedFilesTool{})
-	provider := &mockProvider{
-		turns: [][]zeroruntime.StreamEvent{
-			{
-				{Type: zeroruntime.StreamEventToolCallStart, ToolCallID: "call-1", ToolName: "fake_edit"},
-				{Type: zeroruntime.StreamEventToolCallDelta, ToolCallID: "call-1", ArgumentsFragment: `{}`},
-				{Type: zeroruntime.StreamEventToolCallEnd, ToolCallID: "call-1"},
-				{Type: zeroruntime.StreamEventDone},
-			},
-			{
-				{Type: zeroruntime.StreamEventText, Content: "done"},
-				{Type: zeroruntime.StreamEventDone},
-			},
-			{
-				{Type: zeroruntime.StreamEventText, Content: "done after fix"},
-				{Type: zeroruntime.StreamEventDone},
-			},
-		},
-	}
-
-	result, err := Run(context.Background(), "fix main.go", provider, Options{
-		Registry: registry,
-		Cwd:      t.TempDir(),
-		FileDiagnostics: func(context.Context, string) string {
-			// Slower than the (shortened) per-turn drain, well under the
-			// finalization budget.
-			time.Sleep(100 * time.Millisecond)
-			return "main.go:1:1 error: boom"
-		},
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.FinalAnswer != "done after fix" {
-		t.Fatalf("final answer = %q, want the post-nudge answer", result.FinalAnswer)
-	}
-	if len(provider.requests) != 3 {
-		t.Fatalf("expected a third request carrying the late nudge, got %d requests", len(provider.requests))
-	}
-	found := false
-	for _, message := range provider.requests[2].Messages {
-		if message.Role == zeroruntime.MessageRoleUser && strings.HasPrefix(message.Content, asyncDiagnosticsNudge) {
-			if !strings.Contains(message.Content, "boom") {
-				t.Fatalf("nudge missing diagnostics: %q", message.Content)
-			}
-			found = true
-		}
-	}
-	if !found {
-		t.Fatal("finalization gate did not deliver the late diagnostics nudge")
-	}
-}

internal/agent/command_prefix.go

@@ -1,515 +0,0 @@
-package agent
-
-import (
-	"fmt"
-	"runtime"
-	"strings"
-
-	"github.com/Gitlawb/zero/internal/sandbox"
-	"github.com/Gitlawb/zero/internal/tools"
-	"mvdan.cc/sh/v3/syntax"
-)
-
-func proposedCommandPrefix(toolName string, args map[string]any) []string {
-	if !isShellCommandTool(toolName) {
-		return nil
-	}
-	command, ok := firstStringArg(args, "command", "cmd", "script", "shell")
-	if !ok {
-		return nil
-	}
-	segments, ok := safeShellCommandSegments(command)
-	if !ok {
-		return nil
-	}
-	if requested, ok := requestedPrefixRule(args); ok {
-		if safeRequestedPrefixForSegments(requested, segments) {
-			return requested
-		}
-		return nil
-	}
-	// Only propose approving a prefix of one segment when every other segment
-	// in the command is independently known-safe. Once a prefix is approved,
-	// shellExecutionArgsForApproval escalates the whole command (every
-	// segment, not just the approved one) to bypass the sandbox, so offering a
-	// prefix that leaves an MSYS-prone (or otherwise unsafe) segment uncovered
-	// would let that segment run unsandboxed without ever being reviewed.
-	for index, tokens := range segments {
-		if knownSafeCommandSegment(tokens) {
-			continue
-		}
-		if !sandbox.ValidCommandPrefix(tokens) || !otherSegmentsKnownSafe(segments, index) {
-			return nil
-		}
-		return append([]string(nil), tokens...)
-	}
-	if len(segments) == 0 || !sandbox.ValidCommandPrefix(segments[0]) {
-		return nil
-	}
-	return append([]string(nil), segments[0]...)
-}
-
-// otherSegmentsKnownSafe reports whether every segment other than the one at
-// skip is known-safe on its own.
-func otherSegmentsKnownSafe(segments [][]string, skip int) bool {
-	for index, tokens := range segments {
-		if index == skip {
-			continue
-		}
-		if !knownSafeCommandSegment(tokens) {
-			return false
-		}
-	}
-	return true
-}
-
-func matchCommandPrefix(toolName string, args map[string]any, options Options) (sandbox.CommandPrefixGrant, bool, bool) {
-	if !isShellCommandTool(toolName) || options.Sandbox == nil {
-		return sandbox.CommandPrefixGrant{}, false, false
-	}
-	if shellCommandAdditionalPermissionsRequested(args) {
-		return sandbox.CommandPrefixGrant{}, false, false
-	}
-	command, ok := firstStringArg(args, "command", "cmd", "script", "shell")
-	if !ok {
-		return sandbox.CommandPrefixGrant{}, false, false
-	}
-	segments, ok := safeShellCommandSegments(command)
-	if !ok {
-		return sandbox.CommandPrefixGrant{}, false, false
-	}
-	var matched sandbox.CommandPrefixGrant
-	matchedAny := false
-	matchedSession := false
-	for _, tokens := range segments {
-		if grant, ok := options.Sandbox.LookupCommandPrefix(toolName, tokens); ok {
-			if !matchedAny {
-				matched = grant
-			}
-			matchedAny = true
-			continue
-		}
-		if grant, ok := options.Sandbox.LookupCommandPrefixForSession(toolName, tokens); ok {
-			if !matchedAny {
-				matched = grant
-			}
-			matchedAny = true
-			matchedSession = true
-			continue
-		}
-		if knownSafeCommandSegment(tokens) {
-			continue
-		}
-		return sandbox.CommandPrefixGrant{}, false, false
-	}
-	if matchedAny {
-		return matched, true, matchedSession
-	}
-	return sandbox.CommandPrefixGrant{}, false, false
-}
-
-func shellExecutionArgsForApproval(toolName string, args map[string]any, action PermissionDecisionAction, options Options) map[string]any {
-	if !isShellCommandTool(toolName) || !shellPrefixApprovalBypassesSandbox(action) {
-		return args
-	}
-	if options.Sandbox == nil || !options.Sandbox.UnsandboxedExecutionAllowed() {
-		return args
-	}
-	if shellCommandAdditionalPermissionsRequested(args) || shellCommandRequiresEscalated(args) {
-		return args
-	}
-	planned := cloneArgs(args)
-	if planned == nil {
-		planned = map[string]any{}
-	}
-	planned["sandbox_permissions"] = string(tools.SandboxPermissionsRequireEscalated)
-	return planned
-}
-
-func shellPrefixApprovalBypassesSandbox(action PermissionDecisionAction) bool {
-	return action == PermissionDecisionAllowPrefix || action == PermissionDecisionAlwaysAllowPrefix
-}
-
-func shellCommandRequiresEscalated(args map[string]any) bool {
-	raw, ok := args["sandbox_permissions"]
-	if !ok || raw == nil {
-		return false
-	}
-	value, ok := raw.(string)
-	if !ok {
-		value = fmt.Sprint(raw)
-	}
-	return strings.TrimSpace(value) == string(tools.SandboxPermissionsRequireEscalated)
-}
-
-func isShellCommandTool(toolName string) bool {
-	return toolName == "bash" || toolName == "exec_command"
-}
-
-func firstStringArg(args map[string]any, names ...string) (string, bool) {
-	for _, name := range names {
-		if raw, ok := args[name].(string); ok {
-			value := strings.TrimSpace(raw)
-			if value != "" {
-				return value, true
-			}
-		}
-	}
-	return "", false
-}
-
-func requestedPrefixRule(args map[string]any) ([]string, bool) {
-	raw, ok := args["prefix_rule"]
-	if !ok {
-		return nil, false
-	}
-	switch value := raw.(type) {
-	case []string:
-		return cleanPrefixRule(value), true
-	case []any:
-		out := make([]string, 0, len(value))
-		for _, item := range value {
-			part, ok := item.(string)
-			if !ok {
-				return nil, true
-			}
-			out = append(out, part)
-		}
-		return cleanPrefixRule(out), true
-	default:
-		return nil, true
-	}
-}
-
-func cleanPrefixRule(prefix []string) []string {
-	cleaned := make([]string, 0, len(prefix))
-	for _, part := range prefix {
-		part = strings.TrimSpace(part)
-		if part == "" {
-			return nil
-		}
-		cleaned = append(cleaned, part)
-	}
-	return cleaned
-}
-
-func safeRequestedPrefixForSegments(prefix []string, segments [][]string) bool {
-	if len(prefix) == 0 || !sandbox.ValidCommandPrefix(prefix) {
-		return false
-	}
-	matched := false
-	for _, command := range segments {
-		if len(prefix) > len(command) {
-			if knownSafeCommandSegment(command) {
-				continue
-			}
-			return false
-		}
-		if hasStringPrefix(command, prefix) {
-			matched = true
-			continue
-		}
-		if knownSafeCommandSegment(command) {
-			continue
-		}
-		return false
-	}
-	return matched
-}
-
-func safeShellCommandSegments(command string) ([][]string, bool) {
-	command = strings.TrimSpace(command)
-	if command == "" {
-		return nil, false
-	}
-	file, err := syntax.NewParser().Parse(strings.NewReader(command), "")
-	if err != nil || len(file.Stmts) == 0 {
-		return nil, false
-	}
-	segments := make([][]string, 0, len(file.Stmts))
-	for _, stmt := range file.Stmts {
-		if !collectSafeShellStatement(stmt, &segments) {
-			return nil, false
-		}
-	}
-	if len(segments) == 0 {
-		return nil, false
-	}
-	return segments, true
-}
-
-func collectSafeShellStatement(stmt *syntax.Stmt, segments *[][]string) bool {
-	if stmt == nil || stmt.Negated || stmt.Background || stmt.Coprocess || stmt.Disown || len(stmt.Redirs) > 0 {
-		return false
-	}
-	return collectSafeShellCommand(stmt.Cmd, segments)
-}
-
-func collectSafeShellCommand(cmd syntax.Command, segments *[][]string) bool {
-	switch node := cmd.(type) {
-	case *syntax.CallExpr:
-		tokens, ok := literalCallTokens(node)
-		if !ok {
-			return false
-		}
-		*segments = append(*segments, tokens)
-		return true
-	case *syntax.BinaryCmd:
-		switch node.Op {
-		case syntax.AndStmt, syntax.OrStmt, syntax.Pipe:
-		default:
-			return false
-		}
-		return collectSafeShellStatement(node.X, segments) && collectSafeShellStatement(node.Y, segments)
-	default:
-		return false
-	}
-}
-
-func literalCallTokens(call *syntax.CallExpr) ([]string, bool) {
-	if call == nil || len(call.Assigns) > 0 || len(call.Args) == 0 {
-		return nil, false
-	}
-	tokens := make([]string, 0, len(call.Args))
-	for _, word := range call.Args {
-		if len(word.Parts) != 1 {
-			return nil, false
-		}
-		lit, ok := word.Parts[0].(*syntax.Lit)
-		if !ok || strings.ContainsAny(lit.Value, "*?[]{}") {
-			return nil, false
-		}
-		tokens = append(tokens, lit.Value)
-	}
-	return tokens, true
-}
-
-func knownSafeCommandSegment(command []string) bool {
-	if len(command) == 0 {
-		return false
-	}
-	name := commandName(command[0])
-	if runtime.GOOS == "windows" && tools.MsysProneCommandName(name) {
-		return false
-	}
-	switch name {
-	case "cat", "cd", "cut", "echo", "expr", "false", "grep", "head", "id",
-		"ls", "nl", "paste", "pwd", "rev", "seq", "stat", "tail", "tr",
-		"true", "uname", "uniq", "wc", "which", "whoami":
-		return true
-	case "base64":
-		return safeBase64Command(command[1:])
-	case "find":
-		return safeFindCommand(command[1:])
-	case "rg":
-		return safeRipgrepCommand(command[1:])
-	case "sed":
-		return safeSedCommand(command[1:])
-	case "git":
-		return safeGitCommand(command)
-	default:
-		return false
-	}
-}
-
-func safeBase64Command(args []string) bool {
-	for _, arg := range args {
-		if arg == "-o" || arg == "--output" || strings.HasPrefix(arg, "--output=") || (strings.HasPrefix(arg, "-o") && arg != "-o") {
-			return false
-		}
-	}
-	return true
-}
-
-func safeFindCommand(args []string) bool {
-	unsafe := map[string]bool{
-		"-exec": true, "-execdir": true, "-ok": true, "-okdir": true,
-		"-delete": true,
-		"-fls":    true, "-fprint": true, "-fprint0": true, "-fprintf": true,
-	}
-	for _, arg := range args {
-		if unsafe[arg] {
-			return false
-		}
-	}
-	return true
-}
-
-func safeRipgrepCommand(args []string) bool {
-	for _, arg := range args {
-		if arg == "--search-zip" || arg == "-z" || arg == "--pre" || arg == "--hostname-bin" ||
-			strings.HasPrefix(arg, "--pre=") || strings.HasPrefix(arg, "--hostname-bin=") {
-			return false
-		}
-	}
-	return true
-}
-
-func safeSedCommand(args []string) bool {
-	if len(args) < 2 || len(args) > 3 || args[0] != "-n" {
-		return false
-	}
-	return validSedPrintArg(args[1])
-}
-
-func validSedPrintArg(arg string) bool {
-	if !strings.HasSuffix(arg, "p") {
-		return false
-	}
-	body := strings.TrimSuffix(arg, "p")
-	if body == "" {
-		return false
-	}
-	for _, part := range strings.Split(body, ",") {
-		if part == "" {
-			return false
-		}
-		for _, r := range part {
-			if r < '0' || r > '9' {
-				return false
-			}
-		}
-	}
-	return true
-}
-
-func safeGitCommand(command []string) bool {
-	subIndex, subcommand, ok := gitSubcommand(command)
-	if !ok {
-		return false
-	}
-	if gitHasUnsafeGlobalOption(command[1:subIndex]) {
-		return false
-	}
-	args := command[subIndex+1:]
-	switch subcommand {
-	case "status", "log", "diff", "show":
-		return gitArgsReadOnly(args)
-	case "branch":
-		return gitArgsReadOnly(args) && gitBranchReadOnly(args)
-	default:
-		return false
-	}
-}
-
-func gitSubcommand(command []string) (int, string, bool) {
-	for index := 1; index < len(command); index++ {
-		arg := command[index]
-		if gitOptionConsumesValue(arg) {
-			index++
-			continue
-		}
-		if gitOptionHasInlineValue(arg) || arg == "--" || strings.HasPrefix(arg, "-") {
-			continue
-		}
-		switch arg {
-		case "status", "log", "diff", "show", "branch":
-			return index, arg, true
-		default:
-			return 0, "", false
-		}
-	}
-	return 0, "", false
-}
-
-func gitOptionConsumesValue(arg string) bool {
-	switch arg {
-	case "-C", "-c", "--config-env", "--exec-path", "--git-dir", "--namespace", "--super-prefix", "--work-tree":
-		return true
-	default:
-		return false
-	}
-}
-
-func gitOptionHasInlineValue(arg string) bool {
-	return strings.HasPrefix(arg, "--config-env=") ||
-		strings.HasPrefix(arg, "--exec-path=") ||
-		strings.HasPrefix(arg, "--git-dir=") ||
-		strings.HasPrefix(arg, "--namespace=") ||
-		strings.HasPrefix(arg, "--super-prefix=") ||
-		strings.HasPrefix(arg, "--work-tree=") ||
-		((strings.HasPrefix(arg, "-C") || strings.HasPrefix(arg, "-c")) && len(arg) > 2)
-}
-
-func gitHasUnsafeGlobalOption(args []string) bool {
-	for index := 0; index < len(args); index++ {
-		arg := args[index]
-		switch {
-		case strings.HasPrefix(arg, "--upload-pack"):
-			return true
-		case arg == "-C" || strings.HasPrefix(arg, "-C"):
-			return true
-		case gitOptionConsumesValue(arg):
-			index++
-		}
-	}
-	return false
-}
-
-func gitArgsReadOnly(args []string) bool {
-	for _, arg := range args {
-		switch {
-		case strings.HasPrefix(arg, "--output="), strings.HasPrefix(arg, "--exec="):
-			return false
-		case arg == "--output", arg == "--exec":
-			return false
-		}
-	}
-	return true
-}
-
-func gitBranchReadOnly(args []string) bool {
-	if len(args) == 0 {
-		return true
-	}
-	for _, arg := range args {
-		if strings.HasPrefix(arg, "--format=") {
-			continue
-		}
-		switch arg {
-		case "--list", "-l", "--show-current", "-a", "--all", "-r", "--remotes", "-v", "-vv", "--verbose":
-			continue
-		default:
-			return false
-		}
-	}
-	return true
-}
-
-func commandName(raw string) string {
-	name := strings.TrimSpace(raw)
-	if index := strings.LastIndexAny(name, `/\`); index >= 0 {
-		name = name[index+1:]
-	}
-	name = strings.ToLower(name)
-	for _, suffix := range []string{".exe", ".cmd", ".bat", ".com"} {
-		if strings.HasSuffix(name, suffix) {
-			return strings.TrimSuffix(name, suffix)
-		}
-	}
-	return name
-}
-
-func hasStringPrefix(values []string, prefix []string) bool {
-	if len(prefix) == 0 || len(prefix) > len(values) {
-		return false
-	}
-	for index := range prefix {
-		if values[index] != prefix[index] {
-			return false
-		}
-	}
-	return true
-}
-
-func equalStringSlices(left []string, right []string) bool {
-	if len(left) != len(right) {
-		return false
-	}
-	for index := range left {
-		if left[index] != right[index] {
-			return false
-		}
-	}
-	return true
-}

internal/agent/command_prefix_test.go

@@ -1,196 +0,0 @@
-package agent
-
-import (
-	"runtime"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/sandbox"
-	"github.com/Gitlawb/zero/internal/tools"
-)
-
-func TestProposedCommandPrefixUsesSafeSimpleCommands(t *testing.T) {
-	got := proposedCommandPrefix("bash", map[string]any{"command": "go test ./..."})
-	want := []string{"go", "test", "./..."}
-	if !equalStringSlices(got, want) {
-		t.Fatalf("prefix = %#v, want %#v", got, want)
-	}
-}
-
-func TestProposedCommandPrefixSupportsExecCommand(t *testing.T) {
-	got := proposedCommandPrefix("exec_command", map[string]any{"cmd": "go test ./..."})
-	want := []string{"go", "test", "./..."}
-	if !equalStringSlices(got, want) {
-		t.Fatalf("prefix = %#v, want %#v", got, want)
-	}
-}
-
-func TestProposedCommandPrefixHonorsValidatedRequestedPrefix(t *testing.T) {
-	got := proposedCommandPrefix("bash", map[string]any{
-		"command":     "git status --short",
-		"prefix_rule": []any{"git", "status"},
-	})
-	want := []string{"git", "status"}
-	if !equalStringSlices(got, want) {
-		t.Fatalf("prefix = %#v, want %#v", got, want)
-	}
-}
-
-func TestProposedCommandPrefixSupportsSegmentedCommands(t *testing.T) {
-	got := proposedCommandPrefix("bash", map[string]any{"command": "ps aux | head -5"})
-	if runtime.GOOS == "windows" {
-		// head is MSYS-prone on Windows (#458), so proposedCommandPrefix must
-		// not offer "ps aux" as a reusable prefix here: approving it would
-		// escalate the whole command, including the uncovered head segment,
-		// to bypass the sandbox unreviewed. See
-		// TestProposedCommandPrefixRejectsPrefixLeavingUnsafeTailUncovered for
-		// the platform-independent regression coverage of this behavior.
-		if got != nil {
-			t.Fatalf("expected no prefix on Windows because head is MSYS-prone, got %#v", got)
-		}
-		return
-	}
-	want := []string{"ps", "aux"}
-	if !equalStringSlices(got, want) {
-		t.Fatalf("prefix = %#v, want %#v", got, want)
-	}
-}
-
-// TestProposedCommandPrefixRejectsPrefixLeavingUnsafeTailUncovered guards
-// against proposedCommandPrefix offering to approve one segment of a
-// multi-segment command (e.g. "ps aux") while a different segment (e.g. "npm
-// install") is not known-safe. shellExecutionArgsForApproval escalates the
-// entire command once any prefix is approved, so an uncovered unsafe segment
-// would bypass the sandbox unreviewed. Uses npm, which is never known-safe on
-// any platform, so the assertion does not depend on runtime.GOOS.
-func TestProposedCommandPrefixRejectsPrefixLeavingUnsafeTailUncovered(t *testing.T) {
-	if got := proposedCommandPrefix("bash", map[string]any{"command": "ps aux && npm install"}); got != nil {
-		t.Fatalf("expected no prefix because npm segment is not known-safe, got %#v", got)
-	}
-}
-
-func TestProposedCommandPrefixHonorsRequestedPrefixAcrossSegments(t *testing.T) {
-	got := proposedCommandPrefix("bash", map[string]any{
-		"command":     "git status --short && git status --branch",
-		"prefix_rule": []any{"git", "status"},
-	})
-	want := []string{"git", "status"}
-	if !equalStringSlices(got, want) {
-		t.Fatalf("prefix = %#v, want %#v", got, want)
-	}
-}
-
-func TestProposedCommandPrefixRejectsRequestedPrefixThatDoesNotCoverSegments(t *testing.T) {
-	got := proposedCommandPrefix("bash", map[string]any{
-		"command":     "ps aux && npm install",
-		"prefix_rule": []any{"ps", "aux"},
-	})
-	if got != nil {
-		t.Fatalf("partial requested prefix should be rejected, got %#v", got)
-	}
-}
-
-func TestProposedCommandPrefixRejectsUnsafeRequestedPrefix(t *testing.T) {
-	got := proposedCommandPrefix("bash", map[string]any{
-		"command":     "git status --short",
-		"prefix_rule": []any{"git"},
-	})
-	if got != nil {
-		t.Fatalf("broad requested prefix should be rejected, got %#v", got)
-	}
-}
-
-func TestProposedCommandPrefixRejectsUnsafeShellForms(t *testing.T) {
-	cases := []string{
-		"cat < in > out",
-		"FOO=bar go test",
-		"echo $(whoami)",
-		"cat *.go",
-		"bash -lc go test",
-	}
-	for _, command := range cases {
-		t.Run(command, func(t *testing.T) {
-			if got := proposedCommandPrefix("bash", map[string]any{"command": command}); got != nil {
-				t.Fatalf("unsafe command got prefix %#v", got)
-			}
-		})
-	}
-}
-
-func TestProposedCommandPrefixRejectsUnsafeLaunchers(t *testing.T) {
-	cases := []string{
-		"find . -type f",
-		"xargs rm -rf /tmp/x",
-		"timeout 5 go test ./...",
-		"nice go test ./...",
-		"nohup go test ./...",
-		"watch ls",
-		"ssh host ls",
-		"make test",
-		"npm run dev",
-		"command git status",
-		"eval echo hi",
-		"exec echo hi",
-		"python script.py",
-		"node script.js",
-		"./script.sh --flag",
-		"/tmp/script.sh --flag",
-	}
-	for _, command := range cases {
-		t.Run(command, func(t *testing.T) {
-			if got := proposedCommandPrefix("bash", map[string]any{"command": command}); got != nil {
-				t.Fatalf("unsafe launcher got prefix %#v", got)
-			}
-		})
-	}
-}
-
-func TestMatchCommandPrefixCoversSegmentedCommandWithSafeTail(t *testing.T) {
-	engine := sandbox.NewEngine(sandbox.EngineOptions{WorkspaceRoot: t.TempDir()})
-	engine.GrantCommandPrefixForSession("bash", []string{"ps", "aux"})
-	// head is MSYS-prone on Windows (#458) and must not count as a known-safe tail.
-	command := "ps aux | head -5"
-	if runtime.GOOS == "windows" {
-		command = "ps aux | echo ok"
-	}
-
-	grant, ok, session := matchCommandPrefix("bash", map[string]any{"command": command}, Options{Sandbox: engine})
-	if !ok || !session || !equalStringSlices(grant.Prefix, []string{"ps", "aux"}) {
-		t.Fatalf("match = %#v ok=%v session=%v, want session ps aux prefix", grant, ok, session)
-	}
-}
-
-func TestKnownSafeCommandSegmentRejectsMsysProneOnWindows(t *testing.T) {
-	if runtime.GOOS != "windows" {
-		t.Skip("windows-only known-safe MSYS guard")
-	}
-	for _, command := range [][]string{{"head", "-5"}, {"cat", "file.txt"}, {"grep", "pat"}} {
-		if knownSafeCommandSegment(command) {
-			t.Fatalf("expected %q to be unsafe on Windows, got known-safe", command)
-		}
-	}
-	if !knownSafeCommandSegment([]string{"echo", "ok"}) {
-		t.Fatal("expected echo to remain known-safe on Windows")
-	}
-	if !tools.MsysProneCommandName("head") {
-		t.Fatal("expected head to be MSYS-prone")
-	}
-}
-
-func TestMatchCommandPrefixRejectsUncoveredSegment(t *testing.T) {
-	engine := sandbox.NewEngine(sandbox.EngineOptions{WorkspaceRoot: t.TempDir()})
-	engine.GrantCommandPrefixForSession("bash", []string{"ps", "aux"})
-
-	if grant, ok, session := matchCommandPrefix("bash", map[string]any{"command": "ps aux && npm install"}, Options{Sandbox: engine}); ok {
-		t.Fatalf("match = %#v session=%v, want no match because npm segment is uncovered", grant, session)
-	}
-}
-
-func TestProposedCommandPrefixRejectsRequestedUnsafeLauncherPrefix(t *testing.T) {
-	got := proposedCommandPrefix("bash", map[string]any{
-		"command":     "find . -type f",
-		"prefix_rule": []any{"find", "."},
-	})
-	if got != nil {
-		t.Fatalf("unsafe requested launcher prefix should be rejected, got %#v", got)
-	}
-}

internal/agent/compaction.go

@@ -2,7 +2,6 @@ package agent
 
 import (
 	"context"
-	"encoding/json"
 	"errors"
 	"strings"
 
@@ -21,17 +20,12 @@ import (
 // the agent loop owns the provider wiring.
 
 // defaultCompactionPreserveLast is how many trailing messages are kept verbatim
-// when the caller does not specify a preserve count. Kept in sync with the replay
-// reconstruction default in internal/sessions/replay.go.
-const defaultCompactionPreserveLast = 6
+// when the caller does not specify a preserve count.
+const defaultCompactionPreserveLast = 8
 
 // compactionTriggerRatio is the fraction of the context window at which
-// proactive compaction kicks in (top of each turn). NOTE: this is speculative
-// tuning — validate against the per-turn latency/compaction metrics (roadmap 5.2)
-// before trusting it. Measured fixed overhead is ~7.4k tokens (not the ~16k the
-// roadmap assumed), so context fills slower; if data shows 0.7 triggers redundant
-// summarization, raise it back toward 0.8.
-const compactionTriggerRatio = 0.7
+// proactive compaction kicks in (top of each turn).
+const compactionTriggerRatio = 0.8
 
 // summaryLabel prefixes the injected summary so it is unmistakable in the
 // transcript (and so tests can assert on it).
@@ -41,9 +35,7 @@ const summaryLabel = "[Summary of earlier conversation]"
 const summaryInstructions = "You are compacting a coding-assistant conversation to save context. " +
 	"Write a dense, factual summary of the conversation so far. Preserve: the user's goals and explicit constraints; " +
 	"decisions made and why; files created or modified (with paths) and key code changes; commands run and their important " +
-	"results; and anything still in progress or unresolved. Omit pleasantries. Use terse bullet points. Do not invent details. " +
-	"If the conversation already begins with an earlier summary block, treat its facts as established context and carry them " +
-	"forward into the new summary — never drop earlier information."
+	"results; and anything still in progress or unresolved. Omit pleasantries. Use terse bullet points. Do not invent details."
 
 // CompactionOptions configure a single Compact call.
 type CompactionOptions struct {
@@ -57,91 +49,24 @@ type CompactionOptions struct {
 	Summarize func(toSummarize []zeroruntime.Message) (string, error)
 }
 
-// CompactionResult is the metadata-bearing result returned by CompactMessages.
-type CompactionResult struct {
-	// Messages is the original conversation or the compacted replacement.
-	Messages []zeroruntime.Message
-	// RemovedCount is the number of original messages summarized away.
-	RemovedCount int
-	// PreservedCount is the number of original messages kept verbatim, including
-	// leading system messages and the preserved suffix.
-	PreservedCount int
-	// SummaryText is the trimmed text returned by the summarizer. The injected
-	// summary message also includes summaryLabel and any preserved structured
-	// state needed by later compactions.
-	SummaryText string
-	// Compacted reports whether Messages contains an injected summary.
-	Compacted bool
-}
-
-// imageTokenEstimate is a flat per-image token cost. Vision models tokenize an
-// image by its resolution (tiles), not its byte length, so the raw Data size is
-// a poor proxy — counting len(Data)/4 would overcount by orders of magnitude.
-// A fixed estimate keeps estimateTokens monotonic in the image count so an
-// image-heavy context still trends toward compaction instead of reading as ~0.
-// ~1k tokens is a representative mid-range cost across providers.
-const imageTokenEstimate = 1000
-
-// ApproxTextTokens estimates the token count of text WITHOUT a real tokenizer.
-// The BPE tokenizers zero targets fold a run's leading space into the following
-// token rather than emitting whitespace as its own token, so naive len/4
-// overcounts real text by ~15-20% (measured: a 24.8k-char prompt counted 5.2k
-// real tokens where len/4 said 6.2k). Counting NON-whitespace bytes / 4 tracks
-// the provider's actual count closely (validated against live usage) while
-// staying allocation- and dependency-free. zero still receives the exact count
-// back as usage on every request; this estimate is only for the pre-request
-// context budget preview and the compaction threshold.
-func ApproxTextTokens(value string) int {
-	nonSpace := 0
-	for i := 0; i < len(value); i++ {
-		switch value[i] {
-		case ' ', '\t', '\n', '\r', '\f', '\v':
-		default:
-			nonSpace++
-		}
-	}
-	return nonSpace / 4
-}
-
-// estimateTokens is a cheap, dependency-free token estimate (see ApproxTextTokens)
-// across message content plus tool call names/arguments and a flat per-image
-// cost. It deliberately uses no real tokenizer; it only needs to be monotonic
-// and roughly proportional so the loop can decide when to compact.
+// estimateTokens is a cheap, dependency-free token estimate (~4 chars/token)
+// across message content plus tool call names/arguments. It deliberately uses no
+// real tokenizer; it only needs to be monotonic and roughly proportional so the
+// loop can decide when to compact.
 func estimateTokens(messages []zeroruntime.Message) int {
 	total := 0
 	for _, message := range messages {
-		total += ApproxTextTokens(message.Content)
+		total += len(message.Content) / 4
 		for _, call := range message.ToolCalls {
-			total += ApproxTextTokens(call.Name)
-			total += ApproxTextTokens(call.Arguments)
+			total += len(call.Name) / 4
+			total += len(call.Arguments) / 4
 			total += 4 // small per-call overhead
 		}
-		total += len(message.Images) * imageTokenEstimate
 		total += 4 // per-message overhead
 	}
 	return total
 }
 
-// estimateToolDefTokens approximates the input-token cost of the tool definitions
-// sent with every request (~4 chars/token, matching estimateTokens). Compaction
-// must include them: they ride on every turn, so ignoring them under-counts the
-// real context and can let it blow past the model limit while the message-only
-// estimate still looks under threshold.
-func estimateToolDefTokens(tools []zeroruntime.ToolDefinition) int {
-	total := 0
-	for _, tool := range tools {
-		total += ApproxTextTokens(tool.Name)
-		total += ApproxTextTokens(tool.Description)
-		if len(tool.Parameters) > 0 {
-			if encoded, err := json.Marshal(tool.Parameters); err == nil {
-				total += ApproxTextTokens(string(encoded))
-			}
-		}
-		total += 4 // per-tool overhead
-	}
-	return total
-}
-
 // compactionThreshold is the estimated-token level at which proactive
 // compaction triggers for a given context window.
 func compactionThreshold(contextWindow int) int {
@@ -169,23 +94,12 @@ func compactionThreshold(contextWindow int) int {
 // Compact is pure: it performs no provider I/O. A Summarize error is returned to
 // the caller, which decides how to recover.
 func Compact(messages []zeroruntime.Message, opts CompactionOptions) ([]zeroruntime.Message, error) {
-	result, err := CompactMessages(messages, opts)
-	if err != nil {
-		return nil, err
-	}
-	return result.Messages, nil
-}
-
-// CompactMessages summarizes the oldest middle of a conversation and returns
-// both the replacement messages and UI/session-friendly metadata about what
-// changed. It uses the same compaction rules as Compact.
-func CompactMessages(messages []zeroruntime.Message, opts CompactionOptions) (CompactionResult, error) {
 	preserveLast := opts.PreserveLast
 	if preserveLast <= 0 {
 		preserveLast = defaultCompactionPreserveLast
 	}
 	if opts.Summarize == nil {
-		return CompactionResult{}, errors.New("compaction requires a Summarize function")
+		return nil, errors.New("compaction requires a Summarize function")
 	}
 
 	// Leading system messages are kept verbatim at the head.
@@ -205,36 +119,23 @@ func CompactMessages(messages []zeroruntime.Message, opts CompactionOptions) (Co
 	middle := messages[systemEnd:boundary]
 	if len(middle) == 0 {
 		// Nothing to summarize once system + preserved suffix are removed.
-		return CompactionResult{
-			Messages:       messages,
-			PreservedCount: len(messages),
-		}, nil
+		return messages, nil
 	}
 
 	summary, err := opts.Summarize(middle)
 	if err != nil {
-		return CompactionResult{}, err
+		return nil, err
 	}
 	summary = strings.TrimSpace(summary)
 
-	// Preserve structured state (active plan + loaded skills) from the elided
-	// middle verbatim, so it is not lost or paraphrased away by the prose summary.
-	content := appendPreservedState(summaryLabel+"\n"+summary, middle)
-
 	compacted := make([]zeroruntime.Message, 0, systemEnd+1+(len(messages)-boundary))
 	compacted = append(compacted, messages[:systemEnd]...)
 	compacted = append(compacted, zeroruntime.Message{
 		Role:    zeroruntime.MessageRoleUser,
-		Content: content,
+		Content: summaryLabel + "\n" + summary,
 	})
 	compacted = append(compacted, messages[boundary:]...)
-	return CompactionResult{
-		Messages:       compacted,
-		RemovedCount:   len(middle),
-		PreservedCount: len(messages) - len(middle),
-		SummaryText:    summary,
-		Compacted:      true,
-	}, nil
+	return compacted, nil
 }
 
 // safeSuffixBoundary walks the preserve boundary backward (toward systemEnd) so
@@ -285,25 +186,6 @@ func isContextLimitError(message string) bool {
 	return false
 }
 
-// isStreamTimeoutError reports whether a streamed error is a provider stream
-// idle/stall timeout (providerio surfaces these as "idle timeout after …" /
-// "no output for …" / "stream stalled"). Such a turn produced nothing, so when
-// no content was forwarded yet it can be safely re-issued on a fresh connection
-// (see the stall-retry in the agent loop). Substring + case-insensitive to
-// tolerate the prefixed "provider stream error: …" wrapping.
-func isStreamTimeoutError(message string) bool {
-	lowered := strings.ToLower(strings.TrimSpace(message))
-	if lowered == "" {
-		return false
-	}
-	for _, needle := range []string{"idle timeout after", "no output for", "stream stalled"} {
-		if strings.Contains(lowered, needle) {
-			return true
-		}
-	}
-	return false
-}
-
 // compactionState carries the per-run state the agent loop needs to compact a
 // conversation safely. It is created once per Run and is a no-op whenever
 // options.ContextWindow <= 0.
@@ -321,47 +203,6 @@ type compactionState struct {
 	// (even after compaction) could loop indefinitely; one attempt then the
 	// original error surfaces.
 	reactiveAttempted bool
-	// onUsage forwards the summarizer call's token usage for accounting/budgeting.
-	// OnText is deliberately NOT forwarded (compaction stays invisible to the user),
-	// but its token COST must still be counted so usage reports and budgets include it.
-	onUsage func(Usage)
-
-	// calibrationRatio scales the raw byte/4 token estimate toward the provider's
-	// real prompt-token count. ApproxTextTokens over-counts code-heavy content by
-	// ~15-20%, which would trip compaction early (at ~60% of true capacity). It
-	// starts at 1.0 and converges via an EMA as each turn reports actual usage, so
-	// later turns compact nearer to real capacity. Zero is treated as 1.0.
-	calibrationRatio float64
-}
-
-// calibrate folds one turn's (rawEstimate, actualPromptTokens) sample into the
-// running calibration ratio. A single sample is clamped to a sane band so an
-// outlier (a huge cache-read turn, a provider-overhead spike) can't skew the
-// estimate enough to disable or thrash compaction.
-func (state *compactionState) calibrate(rawEstimate int, actualPromptTokens int) {
-	if !state.enabled || rawEstimate <= 0 || actualPromptTokens <= 0 {
-		return
-	}
-	sample := float64(actualPromptTokens) / float64(rawEstimate)
-	if sample < 0.5 {
-		sample = 0.5
-	} else if sample > 2.0 {
-		sample = 2.0
-	}
-	if state.calibrationRatio <= 0 {
-		state.calibrationRatio = 1.0
-	}
-	const alpha = 0.3 // weight on the newest sample; smooths jitter across turns
-	state.calibrationRatio = state.calibrationRatio*(1-alpha) + sample*alpha
-}
-
-// calibratedTokens applies the learned ratio to a raw estimate. Before any sample
-// arrives (ratio unset) it returns the raw estimate unchanged.
-func (state *compactionState) calibratedTokens(raw int) int {
-	if state.calibrationRatio <= 0 {
-		return raw
-	}
-	return int(float64(raw) * state.calibrationRatio)
 }
 
 func newCompactionState(options Options) *compactionState {
@@ -369,7 +210,6 @@ func newCompactionState(options Options) *compactionState {
 		enabled:      options.ContextWindow > 0,
 		threshold:    compactionThreshold(options.ContextWindow),
 		preserveLast: options.CompactionPreserveLast,
-		onUsage:      options.OnUsage,
 	}
 }
 
@@ -381,16 +221,11 @@ func (state *compactionState) maybeCompact(
 	ctx context.Context,
 	provider Provider,
 	messages []zeroruntime.Message,
-	tools []zeroruntime.ToolDefinition,
 ) []zeroruntime.Message {
 	if !state.enabled {
 		return messages
 	}
-	// Tool definitions are part of every request's input, so count them alongside
-	// the messages; both the threshold check and the shrink check below use the
-	// same term so they stay consistent.
-	toolTokens := estimateToolDefTokens(tools)
-	size := state.calibratedTokens(estimateTokens(messages) + toolTokens)
+	size := estimateTokens(messages)
 	if size <= state.threshold {
 		return messages
 	}
@@ -401,29 +236,16 @@ func (state *compactionState) maybeCompact(
 		return messages
 	}
 
-	// CHEAP FIRST STAGE: reclaim context at zero token/latency cost by pruning
-	// the bodies of old, large tool results (the model has already acted on
-	// them). If that brings us back under threshold, skip the paid summarizer
-	// entirely and preserve recent turns verbatim.
-	if pruned, reclaimed := pruneStaleToolOutput(messages, state.preserveLast); reclaimed > 0 {
-		messages = pruned
-		size = state.calibratedTokens(estimateTokens(messages) + toolTokens)
-		if size <= state.threshold {
-			state.lowWaterMark = size
-			return messages
-		}
-	}
-
 	compacted, err := Compact(messages, CompactionOptions{
 		PreserveLast: state.preserveLast,
-		Summarize:    summarizeClosure(ctx, provider, state.onUsage),
+		Summarize:    summarizeClosure(ctx, provider),
 	})
 	if err != nil {
 		// Summarizer failed: keep the original history. The reactive path (or a
 		// later turn) can try again; we never drop messages on failure here.
 		return messages
 	}
-	newSize := state.calibratedTokens(estimateTokens(compacted) + toolTokens)
+	newSize := estimateTokens(compacted)
 	if newSize >= size {
 		// Compaction did not actually shrink anything (e.g. nothing to
 		// summarize). Leave the history untouched and don't churn next turn.
@@ -443,7 +265,6 @@ func (state *compactionState) recover(
 	ctx context.Context,
 	provider Provider,
 	messages []zeroruntime.Message,
-	tools []zeroruntime.ToolDefinition,
 	errorMessage string,
 ) (compacted []zeroruntime.Message, retried bool, err error) {
 	if !state.enabled {
@@ -460,7 +281,7 @@ func (state *compactionState) recover(
 
 	result, compactErr := Compact(messages, CompactionOptions{
 		PreserveLast: state.preserveLast,
-		Summarize:    summarizeClosure(ctx, provider, state.onUsage),
+		Summarize:    summarizeClosure(ctx, provider),
 	})
 	if compactErr != nil {
 		// A genuine compaction attempt was made (and failed): the budget is spent
@@ -477,97 +298,38 @@ func (state *compactionState) recover(
 	}
 	// Success: a real compaction shrank the history and we will retry. Consume the
 	// one-shot budget now so a provider that keeps returning context-limit errors
-	// after a successful compaction can't loop forever. Store the low-water mark in
-	// the SAME combined (messages + tool-defs) domain maybeCompact uses, so the
-	// proactive shrink-guard compares like with like.
+	// after a successful compaction can't loop forever.
 	state.reactiveAttempted = true
-	state.lowWaterMark = state.calibratedTokens(estimateTokens(result) + estimateToolDefTokens(tools))
+	state.lowWaterMark = estimateTokens(result)
 	return result, true, nil
 }
 
 // summarizeClosure builds a Summarize function backed by a focused, tool-less
-// provider call. The summary stream intentionally does NOT forward OnText (so
-// compaction stays invisible on the user-facing surface), but it DOES forward
-// OnUsage so the summarizer's token cost is still counted by usage/budgeting.
-func summarizeClosure(ctx context.Context, provider Provider, onUsage func(Usage)) func([]zeroruntime.Message) (string, error) {
+// provider call. The summary stream intentionally does NOT forward OnText /
+// OnUsage callbacks, so compaction stays invisible on the user-facing surface.
+func summarizeClosure(ctx context.Context, provider Provider) func([]zeroruntime.Message) (string, error) {
 	return func(toSummarize []zeroruntime.Message) (string, error) {
-		return summarizeWithFallback(ctx, provider, toSummarize, onUsage)
-	}
-}
-
-// summarizeWithFallback summarizes messages in a single provider call. If that
-// call fails with a context-limit error — the slice to summarize is itself too
-// large for the model's input window — it splits the slice in half, summarizes
-// each half recursively, and joins the partial summaries. This keeps compaction
-// working when the elided middle is bigger than the summarizer's own context.
-// Non-context-limit errors (and a single message that still won't fit) surface
-// to the caller unchanged.
-func summarizeWithFallback(ctx context.Context, provider Provider, messages []zeroruntime.Message, onUsage func(Usage)) (string, error) {
-	summary, err := summarizeMessagesOnce(ctx, provider, messages, onUsage)
-	if err == nil {
+		request := zeroruntime.CompletionRequest{
+			Messages: []zeroruntime.Message{
+				{Role: zeroruntime.MessageRoleSystem, Content: summaryInstructions},
+				{Role: zeroruntime.MessageRoleUser, Content: "Summarize this conversation:\n\n" + renderTranscript(toSummarize)},
+			},
+			// No tools: this is a plain text summarization call.
+		}
+		stream, err := provider.StreamCompletion(ctx, request)
+		if err != nil {
+			return "", err
+		}
+		collected := zeroruntime.CollectStream(ctx, stream)
+		if collected.Error != "" {
+			return "", errors.New(collected.Error)
+		}
+		summary := strings.TrimSpace(collected.Text)
+		if summary == "" {
+			return "", errors.New("summarizer returned no text")
+		}
 		return summary, nil
 	}
-	if len(messages) < 2 || !isContextLimitError(err.Error()) {
-		return "", err
-	}
-
-	mid := len(messages) / 2
-	left, leftErr := summarizeWithFallback(ctx, provider, messages[:mid], onUsage)
-	if leftErr != nil {
-		return "", leftErr
-	}
-	right, rightErr := summarizeWithFallback(ctx, provider, messages[mid:], onUsage)
-	if rightErr != nil {
-		return "", rightErr
-	}
-
-	// Re-summarize the two partial summaries into ONE so the persisted summary
-	// stays a single, further-summarizable unit — not an ever-growing concatenated
-	// blob that a later compaction (which can't split a single message) would fail
-	// on. If even the combined partials don't fit (extreme), fall back to the
-	// joined text: still better than failing, and each half is already compacted.
-	combined := strings.TrimSpace(left + "\n\n" + right)
-	reduced, reduceErr := summarizeMessagesOnce(ctx, provider, []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleUser, Content: combined},
-	}, onUsage)
-	if reduceErr != nil {
-		if isContextLimitError(reduceErr.Error()) {
-			// Even the two combined partials don't fit (extreme): fall back to the
-			// joined text — still better than failing, and each half is already
-			// compacted.
-			return combined, nil
-		}
-		// A non-context failure (auth/network/provider) must surface unchanged, per
-		// this function's contract — don't mask it behind the joined fallback.
-		return "", reduceErr
-	}
-	return reduced, nil
-}
-
-// summarizeMessagesOnce performs a single tool-less summarization call.
-func summarizeMessagesOnce(ctx context.Context, provider Provider, messages []zeroruntime.Message, onUsage func(Usage)) (string, error) {
-	request := zeroruntime.CompletionRequest{
-		Messages: []zeroruntime.Message{
-			{Role: zeroruntime.MessageRoleSystem, Content: summaryInstructions},
-			{Role: zeroruntime.MessageRoleUser, Content: "Summarize this conversation:\n\n" + renderTranscript(messages)},
-		},
-		// No tools: this is a plain text summarization call.
-	}
-	stream, err := provider.StreamCompletion(ctx, request)
-	if err != nil {
-		return "", err
-	}
-	// Forward OnUsage (token accounting) but not OnText (keep compaction invisible
-	// to the user); a nil onUsage is a no-op.
-	collected := zeroruntime.CollectStreamWithOptions(ctx, stream, zeroruntime.CollectOptions{OnUsage: onUsage})
-	if collected.Error != "" {
-		return "", errors.New(collected.Error)
-	}
-	summary := strings.TrimSpace(collected.Text)
-	if summary == "" {
-		return "", errors.New("summarizer returned no text")
-	}
-	return summary, nil
 }
 
 // renderTranscript flattens messages into a plain-text transcript for the

internal/agent/compaction_calibrate_test.go

@@ -1,50 +0,0 @@
-package agent
-
-import "testing"
-
-func TestCalibrationConvergesTowardActual(t *testing.T) {
-	state := &compactionState{enabled: true}
-
-	// Before any sample, the estimate is unchanged.
-	if got := state.calibratedTokens(1000); got != 1000 {
-		t.Fatalf("uncalibrated tokens = %d, want 1000", got)
-	}
-
-	// The estimator over-counts: our raw estimate is 1000 but the provider reports
-	// 850 (a 0.85 ratio). Feeding that sample repeatedly should pull the ratio
-	// toward 0.85, so a 1000-token raw estimate calibrates downward.
-	for range 20 {
-		state.calibrate(1000, 850)
-	}
-	got := state.calibratedTokens(1000)
-	if got >= 1000 || got < 820 || got > 880 {
-		t.Fatalf("calibrated tokens = %d, want ~850 after convergence", got)
-	}
-}
-
-func TestCalibrateIgnoresDegenerateSamples(t *testing.T) {
-	state := &compactionState{enabled: true}
-	state.calibrate(0, 500) // zero estimate
-	state.calibrate(500, 0) // zero actual
-	if state.calibrationRatio != 0 {
-		t.Fatalf("degenerate samples must not move the ratio, got %v", state.calibrationRatio)
-	}
-	// Disabled state never calibrates.
-	disabled := &compactionState{enabled: false}
-	disabled.calibrate(1000, 850)
-	if disabled.calibrationRatio != 0 {
-		t.Fatalf("disabled compaction must not calibrate, got %v", disabled.calibrationRatio)
-	}
-}
-
-func TestCalibrateClampsOutliers(t *testing.T) {
-	state := &compactionState{enabled: true}
-	// A wild outlier (10x) is clamped to 2.0, so one bad sample can't blow up the
-	// ratio and disable compaction.
-	for range 50 {
-		state.calibrate(1000, 10000)
-	}
-	if state.calibrationRatio > 2.0 {
-		t.Fatalf("ratio should be clamped at 2.0, got %v", state.calibrationRatio)
-	}
-}

internal/agent/compaction_metadata_test.go

@@ -1,126 +0,0 @@
-package agent
-
-import (
-	"errors"
-	"reflect"
-	"strings"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-func TestCompactMessagesReturnsMetadataForManualCompaction(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleSystem, Content: "system prompt"},
-		{Role: zeroruntime.MessageRoleUser, Content: "first question"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "first answer"},
-		{Role: zeroruntime.MessageRoleUser, Content: "second question"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "recent answer"},
-		{Role: zeroruntime.MessageRoleUser, Content: "latest question"},
-	}
-
-	var captured []zeroruntime.Message
-	result, err := CompactMessages(messages, CompactionOptions{
-		PreserveLast: 2,
-		Summarize: func(toSummarize []zeroruntime.Message) (string, error) {
-			captured = append([]zeroruntime.Message(nil), toSummarize...)
-			return "  manual summary  ", nil
-		},
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	if !result.Compacted {
-		t.Fatal("expected compaction to be reported")
-	}
-	if result.RemovedCount != 3 {
-		t.Fatalf("RemovedCount = %d, want 3", result.RemovedCount)
-	}
-	if result.PreservedCount != 3 {
-		t.Fatalf("PreservedCount = %d, want 3", result.PreservedCount)
-	}
-	if result.SummaryText != "manual summary" {
-		t.Fatalf("SummaryText = %q, want trimmed summary", result.SummaryText)
-	}
-	if len(captured) != 3 || captured[0].Content != "first question" || captured[2].Content != "second question" {
-		t.Fatalf("summarized middle = %#v, want the three non-preserved non-system messages", captured)
-	}
-	if len(result.Messages) != 4 {
-		t.Fatalf("compacted message count = %d, want 4", len(result.Messages))
-	}
-	if result.Messages[0].Content != "system prompt" {
-		t.Fatalf("system message was not preserved at head: %#v", result.Messages)
-	}
-	if result.Messages[1].Role != zeroruntime.MessageRoleUser {
-		t.Fatalf("summary message role = %s, want user", result.Messages[1].Role)
-	}
-	if !strings.Contains(result.Messages[1].Content, summaryLabel) || !strings.Contains(result.Messages[1].Content, "manual summary") {
-		t.Fatalf("summary message did not include label and body: %q", result.Messages[1].Content)
-	}
-	if result.Messages[2].Content != "recent answer" || result.Messages[3].Content != "latest question" {
-		t.Fatalf("preserved suffix changed: %#v", result.Messages[2:])
-	}
-}
-
-func TestCompactMessagesNoopReturnsUncompactedMetadata(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleSystem, Content: "system"},
-		{Role: zeroruntime.MessageRoleUser, Content: "hi"},
-	}
-	called := false
-
-	result, err := CompactMessages(messages, CompactionOptions{
-		PreserveLast: 8,
-		Summarize: func([]zeroruntime.Message) (string, error) {
-			called = true
-			return "summary", nil
-		},
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	if called {
-		t.Fatal("Summarize should not be called for a no-op compaction")
-	}
-	if result.Compacted {
-		t.Fatal("Compacted = true for a no-op")
-	}
-	if result.RemovedCount != 0 {
-		t.Fatalf("RemovedCount = %d, want 0", result.RemovedCount)
-	}
-	if result.PreservedCount != len(messages) {
-		t.Fatalf("PreservedCount = %d, want %d", result.PreservedCount, len(messages))
-	}
-	if result.SummaryText != "" {
-		t.Fatalf("SummaryText = %q, want empty", result.SummaryText)
-	}
-	if !reflect.DeepEqual(result.Messages, messages) {
-		t.Fatalf("Messages changed on no-op: %#v", result.Messages)
-	}
-}
-
-func TestCompactMessagesPropagatesSummarizeError(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleSystem, Content: "system"},
-		{Role: zeroruntime.MessageRoleUser, Content: "first question"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "first answer"},
-		{Role: zeroruntime.MessageRoleUser, Content: "second question"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "recent answer"},
-		{Role: zeroruntime.MessageRoleUser, Content: "latest question"},
-	}
-
-	_, err := CompactMessages(messages, CompactionOptions{
-		PreserveLast: 2,
-		Summarize: func([]zeroruntime.Message) (string, error) {
-			return "", errors.New("summarizer unavailable")
-		},
-	})
-	if err == nil {
-		t.Fatal("expected summarizer error")
-	}
-	if !strings.Contains(err.Error(), "summarizer unavailable") {
-		t.Fatalf("unexpected error: %v", err)
-	}
-}

internal/agent/compaction_preserve.go

@@ -1,683 +0,0 @@
-package agent
-
-import (
-	"encoding/json"
-	"strings"
-	"unicode/utf8"
-
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-// Compaction preservation.
-//
-// A plain prose summary loses structured state the model needs to keep working:
-// the active plan, loaded deferred-tool schemas, loaded skills, and project
-// instruction blocks. When those turns fall into the elided middle, Compact
-// appends that state to the injected summary as JSON so it survives exactly
-// rather than being paraphrased away.
-
-const (
-	toolNameUpdatePlan = "update_plan"
-	toolNameToolSearch = "tool_search"
-	toolNameSkill      = "skill"
-	toolNameWriteFile  = "write_file"
-	toolNameEditFile   = "edit_file"
-)
-
-// maxRecentEdits caps how many edited files are carried across a compaction, and
-// maxEditNoteBytes caps each edit's one-line note, so a long editing run can't
-// bloat the preserved-state block.
-const (
-	maxRecentEdits   = 20
-	maxEditNoteBytes = 160
-)
-
-const (
-	projectInstructionsHeadingPrefix = "# "
-	projectInstructionsHeadingMarker = " instructions for "
-	projectInstructionsOpenTag       = "<INSTRUCTIONS>"
-	projectInstructionsCloseTag      = "</INSTRUCTIONS>"
-)
-
-// preservedStateLabel heads the preserved-state block. Keep this label stable so
-// summaries created by earlier builds remain parseable; the JSON body may carry
-// more fields than the historical label names.
-//
-// The block body is a single line of JSON (see formatPreservedState): JSON
-// escapes everything, so markdown headings, code fences, or quotes round-trip
-// losslessly across repeated compactions.
-const preservedStateLabel = "## Preserved state (active plan + loaded skills; carried across compaction)"
-
-// maxPreservedSkillBytes caps how much of each loaded skill body is carried
-// across a compaction, so a large skill can't defeat the compaction it is part
-// of. The skill's name and head survive; the model can re-load it in full if it
-// needs the rest.
-const maxPreservedSkillBytes = 2 << 10 // 2 KiB
-
-// extractLatestPlan returns a formatted view of the most recent update_plan tool
-// call in messages, so an in-progress plan survives when its turns are elided by
-// compaction. Returns "" when no plan was issued.
-func extractLatestPlan(messages []zeroruntime.Message) string {
-	for i := len(messages) - 1; i >= 0; i-- {
-		calls := messages[i].ToolCalls
-		for j := len(calls) - 1; j >= 0; j-- {
-			if calls[j].Name != toolNameUpdatePlan {
-				continue
-			}
-			if formatted := formatPlanArguments(calls[j].Arguments); formatted != "" {
-				return formatted
-			}
-		}
-	}
-	return ""
-}
-
-// formatPlanArguments renders an update_plan tool call's JSON arguments
-// ({"plan":[{content,status,...}]}) as terse status-tagged bullet lines. Returns
-// "" on malformed arguments or an empty plan.
-func formatPlanArguments(arguments string) string {
-	var parsed struct {
-		Plan []struct {
-			Content string `json:"content"`
-			Step    string `json:"step"`
-			Status  string `json:"status"`
-			Notes   string `json:"notes"`
-		} `json:"plan"`
-	}
-	if err := json.Unmarshal([]byte(strings.TrimSpace(arguments)), &parsed); err != nil {
-		return ""
-	}
-	lines := make([]string, 0, len(parsed.Plan))
-	for _, item := range parsed.Plan {
-		content := strings.TrimSpace(item.Content)
-		if content == "" {
-			content = strings.TrimSpace(item.Step)
-		}
-		if content == "" {
-			continue
-		}
-		status := strings.TrimSpace(item.Status)
-		if status == "" {
-			status = "pending"
-		}
-		line := "- [" + status + "] " + content
-		if notes := strings.TrimSpace(item.Notes); notes != "" {
-			line += "\n  Notes: " + notes
-		}
-		lines = append(lines, line)
-	}
-	return strings.Join(lines, "\n")
-}
-
-// skillEntry is a named preserved body. It began as loaded-skill state and is
-// reused for loaded tools, project instruction blocks, and recent file edits.
-type skillEntry struct {
-	name string
-	body string
-}
-
-// recentEdits returns the files mutated by write_file/edit_file calls in messages
-// — latest note per path, in last-seen order — as skillEntry{name: path, body:
-// note}. After compaction elides the editing turns, this tells the model WHAT it
-// changed in each file (from the tool's result) so it need not re-read to
-// rediscover its own footprint. Capped at maxRecentEdits paths.
-//
-// Ordering is by LAST edit, not first: re-editing a file moves it to the newest
-// position so the tail cap below keeps the file the model most recently touched
-// rather than an earlier, now-stale entry.
-func recentEdits(messages []zeroruntime.Message) []skillEntry {
-	pathByID := map[string]string{}
-	sequence := make([]string, 0)
-	for _, message := range messages {
-		for _, call := range message.ToolCalls {
-			if call.Name != toolNameWriteFile && call.Name != toolNameEditFile {
-				continue
-			}
-			path := editPathFromArguments(call.Arguments)
-			if path == "" {
-				continue
-			}
-			if call.ID != "" {
-				pathByID[call.ID] = path
-			}
-			sequence = append(sequence, path)
-		}
-	}
-	order := lastSeenOrder(sequence)
-	if len(order) == 0 {
-		return nil
-	}
-
-	noteByPath := map[string]string{}
-	for _, message := range messages {
-		if message.Role != zeroruntime.MessageRoleTool || message.ToolCallID == "" {
-			continue
-		}
-		if path, ok := pathByID[message.ToolCallID]; ok {
-			noteByPath[path] = editNote(message.Content)
-		}
-	}
-
-	// Keep the most recent maxRecentEdits paths (the tail of last-seen order).
-	if len(order) > maxRecentEdits {
-		order = order[len(order)-maxRecentEdits:]
-	}
-	entries := make([]skillEntry, 0, len(order))
-	for _, path := range order {
-		entries = append(entries, skillEntry{name: path, body: noteByPath[path]})
-	}
-	return entries
-}
-
-// lastSeenOrder dedupes paths keeping each at its LAST occurrence, preserving
-// chronological order otherwise. A re-edited path therefore lands at the newest
-// position rather than staying pinned to where it first appeared.
-func lastSeenOrder(paths []string) []string {
-	lastIdx := make(map[string]int, len(paths))
-	for i, p := range paths {
-		lastIdx[p] = i
-	}
-	order := make([]string, 0, len(lastIdx))
-	for i, p := range paths {
-		if lastIdx[p] == i {
-			order = append(order, p)
-		}
-	}
-	return order
-}
-
-// editPathFromArguments pulls the target file path from a write_file/edit_file
-// call's JSON arguments (path/file/file_path/filename aliases). Returns "" on
-// malformed arguments or no path.
-func editPathFromArguments(arguments string) string {
-	var parsed struct {
-		Path     string `json:"path"`
-		File     string `json:"file"`
-		FilePath string `json:"file_path"`
-		Filename string `json:"filename"`
-	}
-	if err := json.Unmarshal([]byte(strings.TrimSpace(arguments)), &parsed); err != nil {
-		return ""
-	}
-	for _, candidate := range []string{parsed.Path, parsed.File, parsed.FilePath, parsed.Filename} {
-		if s := strings.TrimSpace(candidate); s != "" {
-			return s
-		}
-	}
-	return ""
-}
-
-// editNote reduces an edit tool's result to a single short line (its first
-// non-empty line, byte-capped on a rune boundary) for the preserved summary.
-func editNote(content string) string {
-	for _, line := range strings.Split(content, "\n") {
-		s := strings.TrimSpace(line)
-		if s == "" {
-			continue
-		}
-		if len(s) > maxEditNoteBytes {
-			limit := maxEditNoteBytes
-			for limit > 0 && !utf8.RuneStart(s[limit]) {
-				limit--
-			}
-			s = s[:limit] + "…"
-		}
-		return s
-	}
-	return ""
-}
-
-// loadedSkills returns the skills loaded via the skill tool in messages — the
-// latest body per name, in first-seen order — matching each skill tool call to
-// its tool result by id.
-func loadedSkills(messages []zeroruntime.Message) []skillEntry {
-	nameByID := map[string]string{}
-	for _, message := range messages {
-		for _, call := range message.ToolCalls {
-			if call.Name == toolNameSkill && call.ID != "" {
-				nameByID[call.ID] = skillNameFromArguments(call.Arguments)
-			}
-		}
-	}
-	if len(nameByID) == 0 {
-		return nil
-	}
-
-	bodyByName := map[string]string{}
-	nameOrder := make([]string, 0, len(nameByID))
-	for _, message := range messages {
-		if message.Role != zeroruntime.MessageRoleTool || message.ToolCallID == "" {
-			continue
-		}
-		name, ok := nameByID[message.ToolCallID]
-		if !ok {
-			continue
-		}
-		if name == "" {
-			name = "(unnamed)"
-		}
-		body := strings.TrimSpace(message.Content)
-		if body == "" {
-			continue
-		}
-		if _, seen := bodyByName[name]; !seen {
-			nameOrder = append(nameOrder, name)
-		}
-		bodyByName[name] = capBody(body)
-	}
-
-	entries := make([]skillEntry, 0, len(nameOrder))
-	for _, name := range nameOrder {
-		entries = append(entries, skillEntry{name: name, body: bodyByName[name]})
-	}
-	return entries
-}
-
-// loadedToolSchemas returns tool_search-loaded schemas from their normal tool
-// result text. ToolResult.Meta is not part of zeroruntime.Message history, so the
-// rendered "Loaded N tools" output is the durable transcript format.
-func loadedToolSchemas(messages []zeroruntime.Message) []skillEntry {
-	toolSearchIDs := map[string]bool{}
-	for _, message := range messages {
-		for _, call := range message.ToolCalls {
-			if call.Name == toolNameToolSearch && call.ID != "" {
-				toolSearchIDs[call.ID] = true
-			}
-		}
-	}
-	if len(toolSearchIDs) == 0 {
-		return nil
-	}
-
-	bodyByName := map[string]string{}
-	nameOrder := make([]string, 0)
-	for _, message := range messages {
-		if message.Role != zeroruntime.MessageRoleTool || !toolSearchIDs[message.ToolCallID] {
-			continue
-		}
-		for _, entry := range loadedToolEntriesFromOutput(message.Content) {
-			if _, seen := bodyByName[entry.name]; !seen {
-				nameOrder = append(nameOrder, entry.name)
-			}
-			bodyByName[entry.name] = entry.body
-		}
-	}
-
-	entries := make([]skillEntry, 0, len(nameOrder))
-	for _, name := range nameOrder {
-		entries = append(entries, skillEntry{name: name, body: bodyByName[name]})
-	}
-	return entries
-}
-
-func loadedToolEntriesFromOutput(output string) []skillEntry {
-	output = strings.TrimSpace(output)
-	if !strings.HasPrefix(output, "Loaded ") || !strings.Contains(output, "Full schemas follow") {
-		return nil
-	}
-	lines := strings.Split(output, "\n")
-	var entries []skillEntry
-	for i := 0; i < len(lines); i++ {
-		name, ok := strings.CutPrefix(strings.TrimSpace(lines[i]), "## ")
-		if !ok {
-			continue
-		}
-		name = strings.TrimSpace(name)
-		if name == "" {
-			continue
-		}
-		start := i
-		end := len(lines)
-		for j := i + 1; j < len(lines); j++ {
-			if strings.HasPrefix(strings.TrimSpace(lines[j]), "## ") {
-				end = j
-				break
-			}
-		}
-		entries = append(entries, skillEntry{name: name, body: capBody(strings.TrimSpace(strings.Join(lines[start:end], "\n")))})
-		i = end - 1
-	}
-	return entries
-}
-
-func projectInstructionEntries(messages []zeroruntime.Message) []skillEntry {
-	bodyBySource := map[string]string{}
-	sourceOrder := make([]string, 0)
-	for _, message := range messages {
-		if message.Role != zeroruntime.MessageRoleUser {
-			continue
-		}
-		source, body := projectInstructionBlock(message.Content)
-		if body == "" {
-			continue
-		}
-		if _, seen := bodyBySource[source]; !seen {
-			sourceOrder = append(sourceOrder, source)
-		}
-		bodyBySource[source] = body
-	}
-
-	entries := make([]skillEntry, 0, len(sourceOrder))
-	for _, source := range sourceOrder {
-		entries = append(entries, skillEntry{name: source, body: bodyBySource[source]})
-	}
-	return entries
-}
-
-func projectInstructionBlock(content string) (string, string) {
-	content = strings.TrimSpace(content)
-	if !strings.HasPrefix(content, projectInstructionsHeadingPrefix) {
-		return "", ""
-	}
-	firstLineEnd := strings.IndexByte(content, '\n')
-	if firstLineEnd < 0 {
-		return "", ""
-	}
-	heading := strings.TrimSpace(content[:firstLineEnd])
-	if !strings.Contains(heading, projectInstructionsHeadingMarker) {
-		return "", ""
-	}
-	open := strings.Index(content, projectInstructionsOpenTag)
-	close := strings.Index(content, projectInstructionsCloseTag)
-	if open < 0 || close < open {
-		return "", ""
-	}
-	close += len(projectInstructionsCloseTag)
-
-	source := strings.TrimPrefix(heading, "# ")
-	body := strings.TrimSpace(heading + "\n\n" + strings.TrimSpace(content[open:close]))
-	return source, body
-}
-
-// skillNameFromArguments pulls the "name" field from a skill tool call's JSON
-// arguments ({"name":"..."}). Returns "" on malformed arguments.
-func skillNameFromArguments(arguments string) string {
-	var parsed struct {
-		Name  string `json:"name"`
-		Skill string `json:"skill"`
-	}
-	if err := json.Unmarshal([]byte(strings.TrimSpace(arguments)), &parsed); err != nil {
-		return ""
-	}
-	if name := strings.TrimSpace(parsed.Name); name != "" {
-		return name
-	}
-	return strings.TrimSpace(parsed.Skill)
-}
-
-// truncationNote is appended to a capped skill body. Its length is reserved
-// within the byte budget so the result never exceeds maxPreservedSkillBytes.
-const truncationNote = "\n… (truncated; re-load the skill for the full body)"
-
-// capBody truncates an over-long skill body to maxPreservedSkillBytes BYTES,
-// cutting on a UTF-8 rune boundary (never splitting a multibyte rune) and
-// reserving room for the note so the result stays within the byte budget. The
-// note is added only when the body is actually truncated.
-func capBody(body string) string {
-	if len(body) <= maxPreservedSkillBytes {
-		return body
-	}
-	limit := maxPreservedSkillBytes - len(truncationNote)
-	if limit < 0 {
-		limit = 0
-	}
-	// Walk back to the start of a rune so a multibyte sequence is never split.
-	for limit > 0 && !utf8.RuneStart(body[limit]) {
-		limit--
-	}
-	return body[:limit] + truncationNote
-}
-
-// preservedState is the JSON shape of the carried-across-compaction block.
-type preservedState struct {
-	Plan                string                 `json:"plan,omitempty"`
-	RecentEdits         []preservedEdit        `json:"recent_edits,omitempty"`
-	Tools               []preservedTool        `json:"tools,omitempty"`
-	Skills              []preservedSkill       `json:"skills,omitempty"`
-	ProjectInstructions []preservedInstruction `json:"project_instructions,omitempty"`
-}
-
-type preservedEdit struct {
-	Path string `json:"path"`
-	Note string `json:"note,omitempty"`
-}
-
-type preservedTool struct {
-	Name string `json:"name"`
-	Body string `json:"body"`
-}
-
-type preservedSkill struct {
-	Name string `json:"name"`
-	Body string `json:"body"`
-}
-
-type preservedInstruction struct {
-	Source string `json:"source"`
-	Body   string `json:"body"`
-}
-
-// appendPreservedState appends active structured state to a compaction summary
-// as a single JSON block. middle is the slice being summarized away.
-//
-// It is robust across repeated compactions: after the first compaction the state
-// may live only inside the injected summary message, which on a later compaction
-// lands in middle with no real tool calls left to extract. Fresh tool calls and
-// instruction blocks override the carried-forward copy by name/source.
-func appendPreservedState(summary string, middle []zeroruntime.Message) string {
-	priorState := parsePreservedStateBlock(latestSummaryContent(middle))
-
-	// Plan: a fresh update_plan in middle is authoritative; otherwise carry the
-	// plan preserved by an earlier compaction.
-	plan := extractLatestPlan(middle)
-	if plan == "" {
-		plan = priorState.Plan
-	}
-
-	// Recent edits: merge edits preserved earlier with fresh write_file/edit_file
-	// results in middle (newer note per path wins AND moves to the newest
-	// position), so the tail cap keeps the files the model most recently touched
-	// across repeated compactions rather than dropping a just-re-edited file.
-	edits := capRecentEdits(mergeRecentEdits(preservedEditsToEntries(priorState.RecentEdits), recentEdits(middle)))
-
-	// Tools: preserve deferred tool_search schemas from the transcript. Fresh
-	// loads override older carried copies by name.
-	tools := mergeSkillEntries(preservedToolsToEntries(priorState.Tools), loadedToolSchemas(middle))
-
-	// Skills: merge skills preserved earlier (older) with fresh loads (newer wins
-	// per name), so a loaded skill survives repeated compactions.
-	skills := mergeSkillEntries(preservedSkillsToEntries(priorState.Skills), loadedSkills(middle))
-
-	instructions := mergeSkillEntries(
-		preservedInstructionsToEntries(priorState.ProjectInstructions),
-		projectInstructionEntries(middle),
-	)
-
-	if block := formatPreservedState(plan, edits, tools, skills, instructions); block != "" {
-		summary += "\n\n" + block
-	}
-	return summary
-}
-
-// mergeRecentEdits overlays fresh edits onto edits preserved by an earlier
-// compaction. Unlike mergeSkillEntries (which keeps refreshed entries in their
-// original slot), a path touched again by a fresh edit MOVES to the newest
-// position and takes the fresh note, so capRecentEdits — which keeps only the
-// most-recent tail — never drops a file the model just re-edited. Paths present
-// only in the older set keep their relative order, ahead of the fresh ones.
-func mergeRecentEdits(older, newer []skillEntry) []skillEntry {
-	if len(newer) == 0 {
-		return older
-	}
-	freshBody := make(map[string]string, len(newer))
-	freshOrder := make([]string, 0, len(newer))
-	for _, e := range newer {
-		if _, ok := freshBody[e.name]; !ok {
-			freshOrder = append(freshOrder, e.name)
-		}
-		freshBody[e.name] = e.body
-	}
-	merged := make([]skillEntry, 0, len(older)+len(freshOrder))
-	// Older entries that were NOT re-edited keep their position, oldest first.
-	for _, e := range older {
-		if _, refreshed := freshBody[e.name]; refreshed {
-			continue // re-appended at its newest position below
-		}
-		merged = append(merged, e)
-	}
-	// Fresh edits follow, in last-seen order, each carrying the fresh note — so
-	// the most recently touched files sit at the tail the cap preserves.
-	for _, name := range freshOrder {
-		merged = append(merged, skillEntry{name: name, body: freshBody[name]})
-	}
-	return merged
-}
-
-// capRecentEdits bounds the preserved edit list to the most recent maxRecentEdits
-// entries after a merge, so repeated compactions can't grow it without limit.
-func capRecentEdits(entries []skillEntry) []skillEntry {
-	if len(entries) > maxRecentEdits {
-		return entries[len(entries)-maxRecentEdits:]
-	}
-	return entries
-}
-
-func preservedEditsToEntries(edits []preservedEdit) []skillEntry {
-	entries := make([]skillEntry, 0, len(edits))
-	for _, e := range edits {
-		if e.Path == "" {
-			continue
-		}
-		entries = append(entries, skillEntry{name: e.Path, body: e.Note})
-	}
-	return entries
-}
-
-// formatPreservedState renders state as the labelled, single-line
-// JSON block. Returns "" when there is nothing to preserve.
-func formatPreservedState(plan string, edits, tools, skills, instructions []skillEntry) string {
-	if plan == "" && len(edits) == 0 && len(tools) == 0 && len(skills) == 0 && len(instructions) == 0 {
-		return ""
-	}
-	state := preservedState{Plan: plan}
-	for _, e := range edits {
-		state.RecentEdits = append(state.RecentEdits, preservedEdit{Path: e.name, Note: e.body})
-	}
-	for _, t := range tools {
-		state.Tools = append(state.Tools, preservedTool{Name: t.name, Body: t.body})
-	}
-	for _, s := range skills {
-		state.Skills = append(state.Skills, preservedSkill{Name: s.name, Body: s.body})
-	}
-	for _, i := range instructions {
-		state.ProjectInstructions = append(state.ProjectInstructions, preservedInstruction{Source: i.name, Body: i.body})
-	}
-	encoded, err := json.Marshal(state)
-	if err != nil {
-		return ""
-	}
-	return preservedStateLabel + "\n" + string(encoded)
-}
-
-// parsePreservedState recovers the plan + skills from a prior summary's preserved
-// block. JSON escaping makes this lossless even when a skill body contains
-// markdown headings, code fences, or quotes. Returns ("", nil) when absent or
-// malformed.
-func parsePreservedState(summaryContent string) (string, []skillEntry) {
-	state := parsePreservedStateBlock(summaryContent)
-	return state.Plan, preservedSkillsToEntries(state.Skills)
-}
-
-func parsePreservedStateBlock(summaryContent string) preservedState {
-	idx := strings.LastIndex(summaryContent, preservedStateLabel)
-	if idx < 0 {
-		return preservedState{}
-	}
-	rest := strings.TrimPrefix(summaryContent[idx+len(preservedStateLabel):], "\n")
-	// The JSON is a single line (json.Marshal escapes newlines).
-	if nl := strings.IndexByte(rest, '\n'); nl >= 0 {
-		rest = rest[:nl]
-	}
-	rest = strings.TrimSpace(rest)
-	if rest == "" {
-		return preservedState{}
-	}
-	var state preservedState
-	if err := json.Unmarshal([]byte(rest), &state); err != nil {
-		return preservedState{}
-	}
-	return state
-}
-
-func preservedToolsToEntries(tools []preservedTool) []skillEntry {
-	entries := make([]skillEntry, 0, len(tools))
-	for _, t := range tools {
-		if t.Name == "" {
-			continue
-		}
-		entries = append(entries, skillEntry{name: t.Name, body: t.Body})
-	}
-	return entries
-}
-
-func preservedSkillsToEntries(skills []preservedSkill) []skillEntry {
-	entries := make([]skillEntry, 0, len(skills))
-	for _, s := range skills {
-		if s.Name == "" {
-			continue
-		}
-		entries = append(entries, skillEntry{name: s.Name, body: s.Body})
-	}
-	return entries
-}
-
-func preservedInstructionsToEntries(instructions []preservedInstruction) []skillEntry {
-	entries := make([]skillEntry, 0, len(instructions))
-	for _, i := range instructions {
-		if i.Source == "" {
-			continue
-		}
-		entries = append(entries, skillEntry{name: i.Source, body: i.Body})
-	}
-	return entries
-}
-
-// latestSummaryContent returns the content of the most recent injected summary
-// message in messages (a user message beginning with summaryLabel), or "".
-func latestSummaryContent(messages []zeroruntime.Message) string {
-	for i := len(messages) - 1; i >= 0; i-- {
-		m := messages[i]
-		if m.Role == zeroruntime.MessageRoleUser && strings.HasPrefix(strings.TrimSpace(m.Content), summaryLabel) {
-			return m.Content
-		}
-	}
-	return ""
-}
-
-// mergeSkillEntries overlays newer skill loads onto older preserved entries by
-// name (newer body wins), keeping the older order and appending genuinely-new
-// skills after.
-func mergeSkillEntries(older, newer []skillEntry) []skillEntry {
-	if len(newer) == 0 {
-		return older
-	}
-	newBody := make(map[string]string, len(newer))
-	for _, e := range newer {
-		newBody[e.name] = e.body
-	}
-	merged := make([]skillEntry, 0, len(older)+len(newer))
-	seen := make(map[string]bool, len(older))
-	for _, e := range older {
-		if b, ok := newBody[e.name]; ok {
-			e.body = b
-		}
-		merged = append(merged, e)
-		seen[e.name] = true
-	}
-	for _, e := range newer {
-		if !seen[e.name] {
-			merged = append(merged, e)
-		}
-	}
-	return merged
-}

internal/agent/compaction_preserve_test.go

@@ -1,365 +0,0 @@
-package agent
-
-import (
-	"strings"
-	"testing"
-	"unicode/utf8"
-
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-// stateConversation is a long enough conversation that Compact elides a middle
-// containing an update_plan call and a loaded skill (call + result).
-func stateConversation() []zeroruntime.Message {
-	return []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleSystem, Content: "system"},
-		{Role: zeroruntime.MessageRoleUser, Content: "build the thing"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "planning", ToolCalls: []zeroruntime.ToolCall{
-			{ID: "p1", Name: "update_plan", Arguments: `{"plan":[{"content":"write code","status":"in_progress"},{"content":"add tests","status":"pending"}]}`},
-		}},
-		{Role: zeroruntime.MessageRoleTool, Content: "plan updated", ToolCallID: "p1"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "loading skill", ToolCalls: []zeroruntime.ToolCall{
-			{ID: "s1", Name: "skill", Arguments: `{"name":"deploy"}`},
-		}},
-		{Role: zeroruntime.MessageRoleTool, Content: "Deploy skill: run `make deploy` then tag the release.", ToolCallID: "s1"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "done step 1"},
-		{Role: zeroruntime.MessageRoleUser, Content: "continue"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "continuing"},
-	}
-}
-
-func compactStateConversation(t *testing.T, messages []zeroruntime.Message) string {
-	t.Helper()
-	compacted, err := Compact(messages, CompactionOptions{
-		PreserveLast: 2,
-		Summarize:    func([]zeroruntime.Message) (string, error) { return "SUMMARY", nil },
-	})
-	if err != nil {
-		t.Fatalf("Compact returned error: %v", err)
-	}
-	// [system, summaryUserMsg, ...suffix] — the summary is the message after system.
-	if len(compacted) < 2 || compacted[1].Role != zeroruntime.MessageRoleUser {
-		t.Fatalf("unexpected compacted shape: %#v", compacted)
-	}
-	if !strings.Contains(compacted[1].Content, summaryLabel) {
-		t.Fatalf("summary message missing label: %q", compacted[1].Content)
-	}
-	return compacted[1].Content
-}
-
-func TestCompactPreservesActivePlan(t *testing.T) {
-	summary := compactStateConversation(t, stateConversation())
-	if !strings.Contains(summary, preservedStateLabel) {
-		t.Fatalf("expected preserved-state block, got %q", summary)
-	}
-	for _, want := range []string{"- [in_progress] write code", "- [pending] add tests"} {
-		if !strings.Contains(summary, want) {
-			t.Fatalf("plan item %q not preserved in %q", want, summary)
-		}
-	}
-}
-
-func TestCompactPreservesLoadedSkills(t *testing.T) {
-	summary := compactStateConversation(t, stateConversation())
-	if !strings.Contains(summary, preservedStateLabel) {
-		t.Fatalf("expected preserved-state block, got %q", summary)
-	}
-	if !strings.Contains(summary, `"name":"deploy"`) || !strings.Contains(summary, "make deploy") {
-		t.Fatalf("skill name/body not preserved in %q", summary)
-	}
-}
-
-func TestCompactPreservesLoadedToolSearchSchemas(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleSystem, Content: "system"},
-		{Role: zeroruntime.MessageRoleUser, Content: "load weather tool"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "loading", ToolCalls: []zeroruntime.ToolCall{
-			{ID: "t1", Name: "tool_search", Arguments: `{"query":"select:weather_lookup"}`},
-		}},
-		{Role: zeroruntime.MessageRoleTool, ToolCallID: "t1", Content: "Loaded 1 tool. Full schemas follow; call them on the next turn.\n\n## weather_lookup\nLook up weather.\ninput schema:\n{\n  \"type\": \"object\"\n}"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "ready"},
-		{Role: zeroruntime.MessageRoleUser, Content: "continue"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "continuing"},
-	}
-	summary := compactStateConversation(t, messages)
-	if !strings.Contains(summary, `"name":"weather_lookup"`) || !strings.Contains(summary, "input schema") {
-		t.Fatalf("loaded tool schema not preserved in %q", summary)
-	}
-}
-
-func TestCompactPreservesProjectInstructions(t *testing.T) {
-	projectInstructions := "# AGENTS.md instructions for D:\\repo\n\n<INSTRUCTIONS>\nUse `go test ./internal/agent` for agent changes.\nDo not touch TUI code.\n</INSTRUCTIONS>\n\n<environment_context>\nignored runtime context\n</environment_context>"
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleSystem, Content: "system"},
-		{Role: zeroruntime.MessageRoleUser, Content: projectInstructions},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "ack"},
-		{Role: zeroruntime.MessageRoleUser, Content: "work on compaction"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "working"},
-		{Role: zeroruntime.MessageRoleUser, Content: "continue"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "continuing"},
-	}
-	summary := compactStateConversation(t, messages)
-	state := parsePreservedStateBlock(summary)
-	if len(state.ProjectInstructions) != 1 {
-		t.Fatalf("expected one preserved project instruction block, got %#v", state.ProjectInstructions)
-	}
-	body := state.ProjectInstructions[0].Body
-	if state.ProjectInstructions[0].Source != "AGENTS.md instructions for D:\\repo" ||
-		!strings.Contains(body, "# AGENTS.md instructions for D:\\repo") ||
-		!strings.Contains(body, "Do not touch TUI code.") ||
-		strings.Contains(body, "ignored runtime context") {
-		t.Fatalf("project instructions not preserved cleanly in %#v", state.ProjectInstructions[0])
-	}
-}
-
-func TestProjectInstructionBlockAcceptsProjectGuidelineFilename(t *testing.T) {
-	source, body := projectInstructionBlock("# ZERO.md instructions for /repo\n\n<INSTRUCTIONS>\nPrefer Go commands.\n</INSTRUCTIONS>")
-	if source != "ZERO.md instructions for /repo" || !strings.Contains(body, "Prefer Go commands.") {
-		t.Fatalf("expected ZERO.md instruction block to parse, got source=%q body=%q", source, body)
-	}
-}
-
-func TestCompactWithoutStateHasNoPreserveSections(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleSystem, Content: "system"},
-		{Role: zeroruntime.MessageRoleUser, Content: "hello"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "hi there"},
-		{Role: zeroruntime.MessageRoleUser, Content: "tell me more"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "sure"},
-		{Role: zeroruntime.MessageRoleUser, Content: "and more"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "ok"},
-	}
-	summary := compactStateConversation(t, messages)
-	if strings.Contains(summary, preservedStateLabel) {
-		t.Fatalf("did not expect a preserved-state block without plan/skill: %q", summary)
-	}
-}
-
-func TestCompactCarriesPreservedStateAcrossRepeatedCompaction(t *testing.T) {
-	// First compaction: real update_plan + skill load in the elided middle.
-	first, err := Compact(stateConversation(), CompactionOptions{
-		PreserveLast: 2,
-		Summarize:    func([]zeroruntime.Message) (string, error) { return "FIRST SUMMARY", nil },
-	})
-	if err != nil {
-		t.Fatalf("first Compact: %v", err)
-	}
-
-	// Grow the history so the first summary (which holds the preserved sections,
-	// but no real tool calls) falls into the SECOND compaction's middle.
-	second := append([]zeroruntime.Message{}, first...)
-	second = append(second,
-		zeroruntime.Message{Role: zeroruntime.MessageRoleUser, Content: "what next"},
-		zeroruntime.Message{Role: zeroruntime.MessageRoleAssistant, Content: "continuing"},
-		zeroruntime.Message{Role: zeroruntime.MessageRoleUser, Content: "keep going"},
-		zeroruntime.Message{Role: zeroruntime.MessageRoleAssistant, Content: "done"},
-	)
-
-	// The second summarizer deliberately DROPS the preserved sections.
-	out, err := Compact(second, CompactionOptions{
-		PreserveLast: 2,
-		Summarize:    func([]zeroruntime.Message) (string, error) { return "SECOND SUMMARY with no preserved sections", nil },
-	})
-	if err != nil {
-		t.Fatalf("second Compact: %v", err)
-	}
-	if len(out) < 2 || out[1].Role != zeroruntime.MessageRoleUser {
-		t.Fatalf("unexpected compacted shape: %#v", out)
-	}
-	newSummary := out[1].Content
-	// Plan and skill must survive even though the summarizer didn't copy them.
-	if !strings.Contains(newSummary, preservedStateLabel) || !strings.Contains(newSummary, "write code") {
-		t.Fatalf("active plan lost on the second compaction: %q", newSummary)
-	}
-	if !strings.Contains(newSummary, `"name":"deploy"`) || !strings.Contains(newSummary, "make deploy") {
-		t.Fatalf("loaded skill lost on the second compaction: %q", newSummary)
-	}
-}
-
-func TestCompactCarriesLoadedToolsAndProjectInstructionsAcrossRepeatedCompaction(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleSystem, Content: "system"},
-		{Role: zeroruntime.MessageRoleUser, Content: "# AGENTS.md instructions for /repo\n\n<INSTRUCTIONS>\nStay in internal/agent.\n</INSTRUCTIONS>"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "loading", ToolCalls: []zeroruntime.ToolCall{
-			{ID: "t1", Name: "tool_search", Arguments: `{"query":"select:weather_lookup"}`},
-		}},
-		{Role: zeroruntime.MessageRoleTool, ToolCallID: "t1", Content: "Loaded 1 tool. Full schemas follow; call them on the next turn.\n\n## weather_lookup\nLook up weather.\ninput schema:\n{\n  \"type\": \"object\"\n}"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "ready"},
-		{Role: zeroruntime.MessageRoleUser, Content: "continue"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "continuing"},
-	}
-
-	first, err := Compact(messages, CompactionOptions{
-		PreserveLast: 2,
-		Summarize:    func([]zeroruntime.Message) (string, error) { return "FIRST SUMMARY", nil },
-	})
-	if err != nil {
-		t.Fatalf("first Compact: %v", err)
-	}
-	second := append(append([]zeroruntime.Message{}, first...),
-		zeroruntime.Message{Role: zeroruntime.MessageRoleUser, Content: "more"},
-		zeroruntime.Message{Role: zeroruntime.MessageRoleAssistant, Content: "ok"},
-		zeroruntime.Message{Role: zeroruntime.MessageRoleUser, Content: "again"},
-		zeroruntime.Message{Role: zeroruntime.MessageRoleAssistant, Content: "fine"},
-	)
-
-	out, err := Compact(second, CompactionOptions{
-		PreserveLast: 2,
-		Summarize:    func([]zeroruntime.Message) (string, error) { return "SECOND SUMMARY", nil },
-	})
-	if err != nil {
-		t.Fatalf("second Compact: %v", err)
-	}
-	state := parsePreservedStateBlock(out[1].Content)
-	if len(state.Tools) != 1 || state.Tools[0].Name != "weather_lookup" || !strings.Contains(state.Tools[0].Body, "input schema") {
-		t.Fatalf("loaded tool state was not carried forward: %#v", state.Tools)
-	}
-	if len(state.ProjectInstructions) != 1 ||
-		state.ProjectInstructions[0].Source != "AGENTS.md instructions for /repo" ||
-		!strings.Contains(state.ProjectInstructions[0].Body, "Stay in internal/agent.") {
-		t.Fatalf("project instructions were not carried forward: %#v", state.ProjectInstructions)
-	}
-}
-
-// TestCompactPreservesSkillBodyWithMarkdownHeadings is CodeRabbit's regression:
-// a verbatim skill body containing ## / ### headings (and a code fence) must
-// round-trip across TWO compactions without truncation or bogus extra skills —
-// which the old markdown-delimited format could not guarantee.
-func TestCompactPreservesSkillBodyWithMarkdownHeadings(t *testing.T) {
-	body := "## Usage\nrun it\n### Examples\n```\nzero do\n```\n## Notes\ndone"
-	conv := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleSystem, Content: "system"},
-		{Role: zeroruntime.MessageRoleUser, Content: "load a skill"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "loading", ToolCalls: []zeroruntime.ToolCall{
-			{ID: "s1", Name: "skill", Arguments: `{"name":"guide"}`},
-		}},
-		{Role: zeroruntime.MessageRoleTool, Content: body, ToolCallID: "s1"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "done step 1"},
-		{Role: zeroruntime.MessageRoleUser, Content: "continue"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "continuing"},
-	}
-
-	mustContainBody := func(label string, messages []zeroruntime.Message) []zeroruntime.Message {
-		out, err := Compact(messages, CompactionOptions{
-			PreserveLast: 2,
-			Summarize:    func([]zeroruntime.Message) (string, error) { return "SUMMARY", nil },
-		})
-		if err != nil {
-			t.Fatalf("%s Compact: %v", label, err)
-		}
-		if len(out) < 2 || out[1].Role != zeroruntime.MessageRoleUser {
-			t.Fatalf("%s: unexpected compacted shape: %#v", label, out)
-		}
-		_, skills := parsePreservedState(out[1].Content)
-		if len(skills) != 1 || skills[0].name != "guide" || skills[0].body != body {
-			t.Fatalf("%s: skill body not round-tripped intact: %#v", label, skills)
-		}
-		return out
-	}
-
-	first := mustContainBody("first", conv)
-	// Second compaction with NO fresh tool calls and a summarizer that drops it.
-	second := append(append([]zeroruntime.Message{}, first...),
-		zeroruntime.Message{Role: zeroruntime.MessageRoleUser, Content: "more"},
-		zeroruntime.Message{Role: zeroruntime.MessageRoleAssistant, Content: "ok"},
-		zeroruntime.Message{Role: zeroruntime.MessageRoleUser, Content: "again"},
-		zeroruntime.Message{Role: zeroruntime.MessageRoleAssistant, Content: "fine"},
-	)
-	mustContainBody("second", second)
-}
-
-func TestExtractLatestPlanReturnsMostRecent(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleAssistant, ToolCalls: []zeroruntime.ToolCall{
-			{ID: "a", Name: "update_plan", Arguments: `{"plan":[{"content":"old step","status":"completed"}]}`},
-		}},
-		{Role: zeroruntime.MessageRoleAssistant, ToolCalls: []zeroruntime.ToolCall{
-			{ID: "b", Name: "update_plan", Arguments: `{"plan":[{"content":"new step","status":"in_progress"}]}`},
-		}},
-	}
-	got := extractLatestPlan(messages)
-	if !strings.Contains(got, "new step") || strings.Contains(got, "old step") {
-		t.Fatalf("extractLatestPlan should return the most recent plan, got %q", got)
-	}
-}
-
-func TestFormatPlanArgumentsAcceptsStepAlias(t *testing.T) {
-	got := formatPlanArguments(`{"plan":[{"step":"write failing test","status":"in_progress"},{"content":"keep existing shape","status":"pending"}]}`)
-	for _, want := range []string{"- [in_progress] write failing test", "- [pending] keep existing shape"} {
-		if !strings.Contains(got, want) {
-			t.Fatalf("expected %q in formatted plan, got %q", want, got)
-		}
-	}
-}
-
-func TestFormatPlanArgumentsPreservesNotes(t *testing.T) {
-	got := formatPlanArguments(`{"plan":[{"content":"finish preservation","status":"in_progress","notes":"keep TUI untouched"}]}`)
-	if !strings.Contains(got, "- [in_progress] finish preservation") || !strings.Contains(got, "Notes: keep TUI untouched") {
-		t.Fatalf("expected plan content and notes to be preserved, got %q", got)
-	}
-}
-
-func TestCapBodyShortBodyUnchanged(t *testing.T) {
-	body := "short skill body"
-	if got := capBody(body); got != body {
-		t.Fatalf("capBody changed a short body: %q", got)
-	}
-	if strings.Contains(capBody(body), "truncated") {
-		t.Fatalf("note added without truncation")
-	}
-}
-
-func TestCapBodyRespectsByteBudgetForMultibyte(t *testing.T) {
-	// Each '世' is 3 bytes; build a body well over the byte budget.
-	body := strings.Repeat("世", maxPreservedSkillBytes)
-	got := capBody(body)
-	if len(got) > maxPreservedSkillBytes {
-		t.Fatalf("capBody returned %d bytes, want <= %d (byte budget)", len(got), maxPreservedSkillBytes)
-	}
-	if !strings.Contains(got, "truncated") {
-		t.Fatalf("expected truncation note on an over-budget body")
-	}
-	if !utf8.ValidString(got) {
-		t.Fatalf("capBody split a multibyte rune (invalid UTF-8)")
-	}
-}
-
-func TestCapBodyNoteOnlyWhenTruncated(t *testing.T) {
-	// A body whose RUNE count is under the cap but BYTE length is over it must
-	// still be byte-capped (the old rune-based check mishandled this case).
-	body := strings.Repeat("世", (maxPreservedSkillBytes/3)+100)
-	if len(body) <= maxPreservedSkillBytes {
-		t.Fatalf("test setup: body should exceed the byte budget, got %d", len(body))
-	}
-	got := capBody(body)
-	if len(got) > maxPreservedSkillBytes {
-		t.Fatalf("capBody returned %d bytes, want <= %d", len(got), maxPreservedSkillBytes)
-	}
-	if !strings.Contains(got, "truncated") || !utf8.ValidString(got) {
-		t.Fatalf("expected a valid, truncated body, got %q", got)
-	}
-}
-
-func TestLoadedSkillsSkipsCallsWithoutResult(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleAssistant, ToolCalls: []zeroruntime.ToolCall{
-			{ID: "s1", Name: "skill", Arguments: `{"name":"ghost"}`}, // no matching tool result
-		}},
-	}
-	if got := loadedSkills(messages); len(got) != 0 {
-		t.Fatalf("expected no skills without a result body, got %#v", got)
-	}
-}
-
-func TestLoadedSkillsAcceptsSkillArgumentAlias(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleAssistant, ToolCalls: []zeroruntime.ToolCall{
-			{ID: "s1", Name: "skill", Arguments: `{"skill":"deploy"}`},
-		}},
-		{Role: zeroruntime.MessageRoleTool, ToolCallID: "s1", Content: "deploy instructions"},
-	}
-	got := loadedSkills(messages)
-	if len(got) != 1 || got[0].name != "deploy" || got[0].body != "deploy instructions" {
-		t.Fatalf("loadedSkills should honor skill alias, got %#v", got)
-	}
-}

internal/agent/compaction_recent_edits_test.go

@@ -1,121 +0,0 @@
-package agent
-
-import (
-	"fmt"
-	"strings"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-// recentEdits extracts each mutated file's path and a one-line note from the
-// matching tool result, latest note per path in last-seen order.
-func TestRecentEditsExtractsPathsAndNotes(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleAssistant, ToolCalls: []zeroruntime.ToolCall{
-			{ID: "e1", Name: "write_file", Arguments: `{"path":"internal/foo.go","content":"package foo"}`},
-		}},
-		{Role: zeroruntime.MessageRoleTool, ToolCallID: "e1", Content: "Wrote internal/foo.go (12 lines)"},
-		{Role: zeroruntime.MessageRoleAssistant, ToolCalls: []zeroruntime.ToolCall{
-			{ID: "e2", Name: "edit_file", Arguments: `{"path":"internal/bar.go","old_string":"a","new_string":"b"}`},
-		}},
-		{Role: zeroruntime.MessageRoleTool, ToolCallID: "e2", Content: "Applied edit to internal/bar.go"},
-	}
-
-	edits := recentEdits(messages)
-	if len(edits) != 2 {
-		t.Fatalf("expected 2 edited files, got %d: %#v", len(edits), edits)
-	}
-	if edits[0].name != "internal/foo.go" || !strings.Contains(edits[0].body, "12 lines") {
-		t.Fatalf("first edit = %#v, want foo.go with its note", edits[0])
-	}
-	if edits[1].name != "internal/bar.go" || !strings.Contains(edits[1].body, "Applied edit") {
-		t.Fatalf("second edit = %#v, want bar.go with its note", edits[1])
-	}
-}
-
-// After compaction elides the editing turns, the preserved-state block still
-// names the edited files and what changed, so the model needn't re-read them.
-func TestCompactionPreservesRecentEdits(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleSystem, Content: "system"},
-		{Role: zeroruntime.MessageRoleUser, Content: "add a flag"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "editing", ToolCalls: []zeroruntime.ToolCall{
-			{ID: "e1", Name: "write_file", Arguments: `{"path":"cmd/main.go","content":"..."}`},
-		}},
-		{Role: zeroruntime.MessageRoleTool, ToolCallID: "e1", Content: "Wrote cmd/main.go (adds --version flag)"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "done"},
-		{Role: zeroruntime.MessageRoleUser, Content: "continue"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "continuing"},
-	}
-	summary := compactStateConversation(t, messages)
-
-	state := parsePreservedStateBlock(summary)
-	if len(state.RecentEdits) != 1 {
-		t.Fatalf("expected 1 preserved edit, got %#v", state.RecentEdits)
-	}
-	if state.RecentEdits[0].Path != "cmd/main.go" || !strings.Contains(state.RecentEdits[0].Note, "--version") {
-		t.Fatalf("preserved edit = %#v, want cmd/main.go + its note", state.RecentEdits[0])
-	}
-}
-
-// A fresh edit note for a path overrides the one carried from an earlier
-// compaction (newer wins) and moves the path to the newest position rather than
-// duplicating it or leaving it pinned to its old slot.
-func TestRecentEditsMergeNewerWins(t *testing.T) {
-	prior := preservedState{RecentEdits: []preservedEdit{{Path: "a.go", Note: "old note"}, {Path: "z.go", Note: "z"}}}
-	older := preservedEditsToEntries(prior.RecentEdits)
-	newer := []skillEntry{{name: "a.go", body: "new note"}, {name: "b.go", body: "added"}}
-
-	merged := mergeRecentEdits(older, newer)
-	// z.go (untouched) keeps its lead; a.go is re-edited so it takes the new note
-	// and moves behind z.go, ahead of the brand-new b.go: [z.go, a.go, b.go].
-	if len(merged) != 3 {
-		t.Fatalf("expected z.go + refreshed a.go + b.go, got %#v", merged)
-	}
-	if merged[0].name != "z.go" {
-		t.Fatalf("untouched z.go should stay first, got %#v", merged)
-	}
-	if merged[1].name != "a.go" || merged[1].body != "new note" {
-		t.Fatalf("re-edited a.go should move to the newest edits with the new note, got %#v", merged[1])
-	}
-	if merged[2].name != "b.go" {
-		t.Fatalf("brand-new b.go should be last, got %#v", merged[2])
-	}
-}
-
-// Regression: once more than maxRecentEdits distinct files are tracked, re-editing
-// an early file must move it into the capped tail rather than leaving it to be
-// dropped — otherwise the next compaction omits exactly the file the model most
-// recently touched.
-func TestRecentEditsCapKeepsReeditedFile(t *testing.T) {
-	// Earlier compaction preserved exactly maxRecentEdits files, f0 the oldest.
-	older := make([]skillEntry, 0, maxRecentEdits)
-	for i := 0; i < maxRecentEdits; i++ {
-		older = append(older, skillEntry{name: fmt.Sprintf("f%d.go", i), body: "old"})
-	}
-	// Fresh window re-edits the OLDEST file (f0) and adds a brand-new file, so the
-	// merged list exceeds the cap and something must be dropped.
-	newer := []skillEntry{{name: "f0.go", body: "re-edited"}, {name: "fresh.go", body: "new"}}
-
-	merged := capRecentEdits(mergeRecentEdits(older, newer))
-	if len(merged) != maxRecentEdits {
-		t.Fatalf("cap should keep %d entries, got %d", maxRecentEdits, len(merged))
-	}
-
-	var f0 *skillEntry
-	for i := range merged {
-		if merged[i].name == "f0.go" {
-			f0 = &merged[i]
-		}
-		if merged[i].name == "f1.go" {
-			t.Fatalf("f1.go (now the least-recently touched) should have been dropped, got %#v", merged)
-		}
-	}
-	if f0 == nil {
-		t.Fatalf("re-edited f0.go must survive the cap, got %#v", merged)
-	}
-	if f0.body != "re-edited" {
-		t.Fatalf("surviving f0.go should carry the fresh note, got %q", f0.body)
-	}
-}

internal/agent/compaction_summarizer_test.go

@@ -1,179 +0,0 @@
-package agent
-
-import (
-	"context"
-	"fmt"
-	"regexp"
-	"strings"
-	"sync/atomic"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-var markerPattern = regexp.MustCompile(`msg-\d+`)
-
-// sizeLimitedSummarizer returns a context-limit error when the rendered
-// transcript carries more than maxMarkers messages, and otherwise "summarizes"
-// by echoing the message markers it saw — so a successful summary records exactly
-// which messages it covered.
-type sizeLimitedSummarizer struct {
-	maxMarkers int
-	calls      int32
-}
-
-func (p *sizeLimitedSummarizer) StreamCompletion(_ context.Context, request zeroruntime.CompletionRequest) (<-chan zeroruntime.StreamEvent, error) {
-	atomic.AddInt32(&p.calls, 1)
-	text := request.Messages[len(request.Messages)-1].Content
-	markers := markerPattern.FindAllString(text, -1)
-	ch := make(chan zeroruntime.StreamEvent, 2)
-	if len(markers) > p.maxMarkers {
-		ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventError, Error: "context length exceeded"}
-		close(ch)
-		return ch, nil
-	}
-	ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventText, Content: strings.Join(markers, " ")}
-	ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventDone}
-	close(ch)
-	return ch, nil
-}
-
-type errorSummarizer struct {
-	message string
-	calls   int32
-}
-
-func (p *errorSummarizer) StreamCompletion(_ context.Context, _ zeroruntime.CompletionRequest) (<-chan zeroruntime.StreamEvent, error) {
-	atomic.AddInt32(&p.calls, 1)
-	ch := make(chan zeroruntime.StreamEvent, 1)
-	ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventError, Error: p.message}
-	close(ch)
-	return ch, nil
-}
-
-// compressingSummarizer fails on more than maxMarkers messages but returns a
-// SHORT marker-free summary, so two partial summaries combine into something that
-// fits — modelling real summarization that shrinks its input.
-type compressingSummarizer struct {
-	maxMarkers int
-	calls      int32
-}
-
-func (p *compressingSummarizer) StreamCompletion(_ context.Context, request zeroruntime.CompletionRequest) (<-chan zeroruntime.StreamEvent, error) {
-	atomic.AddInt32(&p.calls, 1)
-	text := request.Messages[len(request.Messages)-1].Content
-	ch := make(chan zeroruntime.StreamEvent, 2)
-	if len(markerPattern.FindAllString(text, -1)) > p.maxMarkers {
-		ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventError, Error: "context length exceeded"}
-		close(ch)
-		return ch, nil
-	}
-	ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventText, Content: "S"}
-	ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventDone}
-	close(ch)
-	return ch, nil
-}
-
-func TestSummarizeWithFallbackReSummarizesPartialsIntoOne(t *testing.T) {
-	messages := make([]zeroruntime.Message, 4)
-	for i := range messages {
-		messages[i] = zeroruntime.Message{Role: zeroruntime.MessageRoleUser, Content: fmt.Sprintf("msg-%d body", i)}
-	}
-	provider := &compressingSummarizer{maxMarkers: 2}
-
-	summary, err := summarizeWithFallback(context.Background(), provider, messages, nil)
-	if err != nil {
-		t.Fatalf("summarizeWithFallback failed: %v", err)
-	}
-	// The two chunk summaries ("S" / "S") are re-summarized into ONE unit, not
-	// returned as the joined "S\n\nS" blob — so a later compaction can shrink it.
-	if strings.Contains(summary, "\n\n") {
-		t.Fatalf("expected a single re-summarized result, got a joined blob: %q", summary)
-	}
-	if summary != "S" {
-		t.Fatalf("summary = %q, want the reduced %q", summary, "S")
-	}
-}
-
-func TestSummarizeWithFallbackChunksOnContextLimit(t *testing.T) {
-	const n = 8
-	messages := make([]zeroruntime.Message, n)
-	for i := range messages {
-		messages[i] = zeroruntime.Message{Role: zeroruntime.MessageRoleUser, Content: fmt.Sprintf("msg-%d some content", i)}
-	}
-	// The summarizer can only handle 2 messages per call, so the 8-message slice
-	// must be split recursively until each chunk fits.
-	provider := &sizeLimitedSummarizer{maxMarkers: 2}
-
-	summary, err := summarizeWithFallback(context.Background(), provider, messages, nil)
-	if err != nil {
-		t.Fatalf("summarizeWithFallback failed: %v", err)
-	}
-	for i := 0; i < n; i++ {
-		if !strings.Contains(summary, fmt.Sprintf("msg-%d", i)) {
-			t.Fatalf("combined summary missing msg-%d: %q", i, summary)
-		}
-	}
-	if got := atomic.LoadInt32(&provider.calls); got < 2 {
-		t.Fatalf("expected multiple calls from splitting, got %d", got)
-	}
-}
-
-func TestSummarizeWithFallbackPropagatesNonContextErrors(t *testing.T) {
-	provider := &errorSummarizer{message: "auth error: invalid key"}
-	_, err := summarizeWithFallback(context.Background(), provider, []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleUser, Content: "msg-0"},
-		{Role: zeroruntime.MessageRoleUser, Content: "msg-1"},
-	}, nil)
-	if err == nil {
-		t.Fatal("expected a non-context-limit error to propagate")
-	}
-	if got := atomic.LoadInt32(&provider.calls); got != 1 {
-		t.Fatalf("a non-context error must not trigger splitting/retry, calls=%d", got)
-	}
-}
-
-func TestSummarizeWithFallbackSingleMessageContextLimitSurfaces(t *testing.T) {
-	// A single message that still won't fit can't be split further → error surfaces.
-	provider := &sizeLimitedSummarizer{maxMarkers: 0}
-	_, err := summarizeWithFallback(context.Background(), provider, []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleUser, Content: "msg-0 too big"},
-	}, nil)
-	if err == nil {
-		t.Fatal("expected the context-limit error to surface for an unsplittable single message")
-	}
-}
-
-// usageReportingSummarizer emits a usage event so a test can assert the
-// summarizer's token cost is forwarded to OnUsage.
-type usageReportingSummarizer struct{}
-
-func (usageReportingSummarizer) StreamCompletion(_ context.Context, _ zeroruntime.CompletionRequest) (<-chan zeroruntime.StreamEvent, error) {
-	ch := make(chan zeroruntime.StreamEvent, 3)
-	ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventText, Content: "summary"}
-	ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventUsage, Usage: zeroruntime.Usage{PromptTokens: 100, CompletionTokens: 20}}
-	ch <- zeroruntime.StreamEvent{Type: zeroruntime.StreamEventDone}
-	return ch, nil
-}
-
-func TestSummarizeForwardsUsageButNotText(t *testing.T) {
-	// Compaction must stay invisible to the user (no OnText), but its token cost
-	// MUST be counted, so OnUsage has to fire for the summarizer call.
-	var got zeroruntime.Usage
-	var calls int
-	summary, err := summarizeWithFallback(context.Background(), usageReportingSummarizer{}, []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleUser, Content: "hello"},
-	}, func(u zeroruntime.Usage) { calls++; got = u })
-	if err != nil {
-		t.Fatalf("summarize failed: %v", err)
-	}
-	if summary != "summary" {
-		t.Fatalf("unexpected summary: %q", summary)
-	}
-	if calls != 1 {
-		t.Fatalf("expected OnUsage to fire once, got %d", calls)
-	}
-	if got.PromptTokens != 100 || got.CompletionTokens != 20 {
-		t.Fatalf("unexpected forwarded usage: %#v", got)
-	}
-}

internal/agent/compaction_test.go

@@ -152,59 +152,6 @@ func TestCompactPropagatesSummarizeError(t *testing.T) {
 	}
 }
 
-// scriptedSummaryProvider returns, per StreamCompletion call, either an error
-// event (when the script entry is non-empty) or a success text, so a test can
-// drive the recursive summarizeWithFallback split/reduce path deterministically.
-type scriptedSummaryProvider struct {
-	calls   int
-	scripts []string
-}
-
-func (p *scriptedSummaryProvider) StreamCompletion(_ context.Context, _ zeroruntime.CompletionRequest) (<-chan zeroruntime.StreamEvent, error) {
-	i := p.calls
-	p.calls++
-	if i < len(p.scripts) && p.scripts[i] != "" {
-		return streamEvents([]zeroruntime.StreamEvent{{Type: zeroruntime.StreamEventError, Error: p.scripts[i]}}), nil
-	}
-	return streamEvents([]zeroruntime.StreamEvent{
-		{Type: zeroruntime.StreamEventText, Content: "PARTIAL"},
-		{Type: zeroruntime.StreamEventDone},
-	}), nil
-}
-
-func TestSummarizeWithFallbackPropagatesNonContextReduceError(t *testing.T) {
-	msgs := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleUser, Content: "a"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "b"},
-	}
-	ctxLimit := "This model's maximum context length is 1000 tokens. Please reduce the length of the messages."
-	// call 1 (full slice): context-limit → split; calls 2,3 (halves): succeed;
-	// call 4 (reduce of the two partials): a NON-context provider error that must
-	// surface unchanged rather than be masked by the joined-text fallback.
-	provider := &scriptedSummaryProvider{scripts: []string{ctxLimit, "", "", "provider request failed: 401 unauthorized"}}
-	if _, err := summarizeWithFallback(context.Background(), provider, msgs, func(Usage) {}); err == nil || !strings.Contains(err.Error(), "401") {
-		t.Fatalf("a non-context reduce error must surface, got %v", err)
-	}
-}
-
-func TestSummarizeWithFallbackUsesJoinedTextOnContextLimitReduce(t *testing.T) {
-	msgs := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleUser, Content: "a"},
-		{Role: zeroruntime.MessageRoleAssistant, Content: "b"},
-	}
-	ctxLimit := "This model's maximum context length is 1000 tokens. Please reduce the length of the messages."
-	// Same split, but the reduce also hits a context-limit: fall back to the joined
-	// partial summaries instead of failing.
-	provider := &scriptedSummaryProvider{scripts: []string{ctxLimit, "", "", ctxLimit}}
-	got, err := summarizeWithFallback(context.Background(), provider, msgs, func(Usage) {})
-	if err != nil {
-		t.Fatalf("a context-limit reduce must fall back to joined text, got error %v", err)
-	}
-	if !strings.Contains(got, "PARTIAL") {
-		t.Fatalf("joined fallback should contain the partial summaries, got %q", got)
-	}
-}
-
 // --- estimateTokens tests -------------------------------------------------
 
 func TestEstimateTokensMonotonic(t *testing.T) {
@@ -233,54 +180,6 @@ func TestEstimateTokensCountsToolCallsAndResults(t *testing.T) {
 	}
 }
 
-func TestEstimateTokensCountsImages(t *testing.T) {
-	// A tiny image carries far more model tokens than its text content suggests;
-	// the estimate must rise with each image so an image-heavy context still
-	// trends toward compaction instead of reading as ~0.
-	plain := []zeroruntime.Message{{Role: zeroruntime.MessageRoleUser, Content: "look"}}
-	withImage := []zeroruntime.Message{{
-		Role:    zeroruntime.MessageRoleUser,
-		Content: "look",
-		Images:  []zeroruntime.ImageBlock{{MediaType: "image/png", Data: []byte("tiny")}},
-	}}
-	twoImages := []zeroruntime.Message{{
-		Role:    zeroruntime.MessageRoleUser,
-		Content: "look",
-		Images: []zeroruntime.ImageBlock{
-			{MediaType: "image/png", Data: []byte("tiny")},
-			{MediaType: "image/png", Data: []byte("tiny")},
-		},
-	}}
-	if estimateTokens(withImage) <= estimateTokens(plain) {
-		t.Fatal("expected an image to increase the token estimate")
-	}
-	if estimateTokens(twoImages) <= estimateTokens(withImage) {
-		t.Fatal("expected the estimate to grow with image count")
-	}
-}
-
-func TestEstimateToolDefTokensCountsDefinitions(t *testing.T) {
-	if got := estimateToolDefTokens(nil); got != 0 {
-		t.Fatalf("no tools should estimate 0 tokens, got %d", got)
-	}
-	one := []zeroruntime.ToolDefinition{{
-		Name:        "read_file",
-		Description: "Read a file from the workspace and return its contents.",
-		Parameters:  map[string]any{"type": "object", "properties": map[string]any{"path": map[string]any{"type": "string"}}},
-	}}
-	if estimateToolDefTokens(one) <= 0 {
-		t.Fatal("a tool definition (name + description + schema) should estimate > 0 tokens")
-	}
-	two := append(append([]zeroruntime.ToolDefinition{}, one...), zeroruntime.ToolDefinition{
-		Name:        "write_file",
-		Description: "Write contents to a file in the workspace.",
-		Parameters:  map[string]any{"type": "object"},
-	})
-	if estimateToolDefTokens(two) <= estimateToolDefTokens(one) {
-		t.Fatal("the estimate must grow with more tool definitions")
-	}
-}
-
 // --- Loop integration: provider mocks -------------------------------------
 
 // summarizeRecordingProvider returns scripted turns for the main agent loop
@@ -510,9 +409,9 @@ func TestRunReactiveRetryDoesNotDoubleEmitText(t *testing.T) {
 	if provider.summarizeCalls == 0 {
 		t.Fatal("expected reactive compaction to run")
 	}
-	// The retried turn's text must NOT be re-streamed to OnText. Before the fix it
-	// was emitted on both the original (mid-stream) collect AND the retry collect,
-	// double-emitting the retried response.
+	// The retried turn's text must be streamed to OnText at most once. Before the
+	// fix it was emitted on both the original (mid-stream) collect AND the retry
+	// collect, double-emitting the retried response.
 	joined := strings.Join(deltas, "")
 	if got := strings.Count(joined, "recovered"); got != 0 {
 		t.Fatalf("retried-turn text must NOT be re-streamed to OnText, saw %d occurrences in %q", got, joined)
@@ -592,7 +491,7 @@ func TestRecoverNoopDoesNotConsumeReactiveBudget(t *testing.T) {
 		{Role: zeroruntime.MessageRoleSystem, Content: "sys"},
 		{Role: zeroruntime.MessageRoleUser, Content: "hi"},
 	}
-	_, retried, err := st.recover(context.Background(), provider, tiny, nil, "context length exceeded")
+	_, retried, err := st.recover(context.Background(), provider, tiny, "context length exceeded")
 	if err != nil {
 		t.Fatalf("unexpected error from no-op recover: %v", err)
 	}
@@ -612,7 +511,7 @@ func TestRecoverNoopDoesNotConsumeReactiveBudget(t *testing.T) {
 		{Role: zeroruntime.MessageRoleAssistant, Content: "a2"},
 		{Role: zeroruntime.MessageRoleUser, Content: "u3"},
 	}
-	compacted, retried, err := st.recover(context.Background(), provider, big, nil, "context length exceeded")
+	compacted, retried, err := st.recover(context.Background(), provider, big, "context length exceeded")
 	if err != nil {
 		t.Fatalf("unexpected error from second recover: %v", err)
 	}
@@ -635,7 +534,7 @@ func TestRecoverDisabledIsNoop(t *testing.T) {
 	// context-limit error string.
 	got, retried, err := st.recover(context.Background(), &mockProvider{turns: [][]zeroruntime.StreamEvent{{
 		{Type: zeroruntime.StreamEventText, Content: "should not be called"}, {Type: zeroruntime.StreamEventDone},
-	}}}, msgs, nil, "context length exceeded")
+	}}}, msgs, "context length exceeded")
 	_ = called
 	if retried || err != nil || len(got) != 1 {
 		t.Fatalf("disabled recover must be a no-op, got retried=%v err=%v len=%d", retried, err, len(got))

internal/agent/completion_gate_test.go

@@ -1,342 +0,0 @@
-package agent
-
-import (
-	"context"
-	"errors"
-	"strings"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/tools"
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-// someRequestContains reports whether any message sent to the provider across all
-// turns contains substr — used to assert a continue nudge was actually injected.
-func someRequestContains(requests []zeroruntime.CompletionRequest, substr string) bool {
-	for _, req := range requests {
-		for _, msg := range req.Messages {
-			if strings.Contains(msg.Content, substr) {
-				return true
-			}
-		}
-	}
-	return false
-}
-
-// planTurn is a turn that calls update_plan with the given item statuses (reusing
-// the package's shared toolTurn helper).
-func planTurn(statuses ...string) []zeroruntime.StreamEvent {
-	items := make([]string, len(statuses))
-	for i, s := range statuses {
-		items[i] = `{"content":"step ` + s + `","status":"` + s + `"}`
-	}
-	return toolTurn("plan", "update_plan", `{"plan":[`+strings.Join(items, ",")+`]}`)
-}
-
-// BUG #1 regression: a no-tool-call turn that ends mid-step while plan items are
-// still pending must NOT be accepted as success. The loop must re-prompt (bounded)
-// and, if the model keeps stalling, finalize as INCOMPLETE — never success.
-func TestCompletionGatePendingPlanContinuesThenIncomplete(t *testing.T) {
-	// Mirrors the git-multibranch failure: plan with pending steps, then the model
-	// keeps emitting "…Let me check the SSH configuration:" without acting.
-	cue := "Now I need to configure the SSH server. Let me check the current SSH configuration:"
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewUpdatePlanTool())
-
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		planTurn("completed", "pending", "pending"),
-		textTurn(cue), textTurn(cue), textTurn(cue), textTurn(cue), textTurn(cue),
-	}}
-
-	result, err := Run(context.Background(), "set up a git server", provider, Options{
-		Registry:                registry,
-		MaxTurns:                10,
-		RequireCompletionSignal: true,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if !result.Incomplete {
-		t.Fatalf("expected Incomplete=true (model stalled with pending plan), got false; final=%q turns=%d", result.FinalAnswer, result.Turns)
-	}
-	// 1 plan turn + maxContinueNudges(3) nudged turns + 1 final stalling turn = 5.
-	// Critically it did NOT stop at the first text turn (request 2) as success.
-	if len(provider.requests) != 1+maxContinueNudges+1 {
-		t.Fatalf("expected %d provider turns (1 plan + %d nudges + 1 final), got %d",
-			1+maxContinueNudges+1, maxContinueNudges, len(provider.requests))
-	}
-	if !someRequestContains(provider.requests, continueNudgeMarker) {
-		t.Fatalf("expected a continue nudge (%q) to be injected into the conversation", continueNudgeMarker)
-	}
-}
-
-// A genuinely-complete single-turn answer (no plan, no continuation cue) must
-// still finalize as success — the gate must not break short/read-only tasks.
-func TestCompletionGateAcceptsGenuineCompletion(t *testing.T) {
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		textTurn("The file contains 42 lines."),
-	}}
-
-	result, err := Run(context.Background(), "count the lines", provider, Options{
-		Registry:                tools.NewRegistry(),
-		MaxTurns:                10,
-		RequireCompletionSignal: true,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.Incomplete {
-		t.Fatalf("genuine completion wrongly marked Incomplete; final=%q", result.FinalAnswer)
-	}
-	if result.FinalAnswer != "The file contains 42 lines." {
-		t.Fatalf("final answer = %q, want the completed answer", result.FinalAnswer)
-	}
-	if len(provider.requests) != 1 {
-		t.Fatalf("expected exactly 1 turn (no spurious re-prompt), got %d", len(provider.requests))
-	}
-}
-
-// A continuation-cue turn triggers a re-prompt, but once the model actually
-// finishes (clean answer, no cue, no pending plan) the run exits as success — the
-// nudge gives the model a path to a legitimate completion.
-func TestCompletionGateContinuesOnCueThenSucceeds(t *testing.T) {
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		textTurn("Let me read the file:"),
-		textTurn("Done. The file has 42 lines."),
-	}}
-
-	result, err := Run(context.Background(), "count the lines", provider, Options{
-		Registry:                tools.NewRegistry(),
-		MaxTurns:                10,
-		RequireCompletionSignal: true,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.Incomplete {
-		t.Fatalf("run wrongly marked Incomplete after the model completed; final=%q", result.FinalAnswer)
-	}
-	if result.FinalAnswer != "Done. The file has 42 lines." {
-		t.Fatalf("final answer = %q, want the completed answer", result.FinalAnswer)
-	}
-	if len(provider.requests) != 2 {
-		t.Fatalf("expected 2 turns (1 cue re-prompted + 1 completion), got %d", len(provider.requests))
-	}
-	if !someRequestContains(provider.requests, continueNudgeMarker) {
-		t.Fatalf("expected a continue nudge after the continuation-cue turn")
-	}
-}
-
-// Issue #666: the plan-aware gate applies even when RequireCompletionSignal is
-// off (interactive/TUI). A model that creates a plan then stalls on a
-// continuation-cue text turn must be re-prompted and eventually INCOMPLETE, not
-// accepted as success while the plan panel stays on step one.
-func TestPlanPendingGateAppliesWithoutRequireCompletionSignal(t *testing.T) {
-	cue := "Now I need to configure the SSH server. Let me check the current SSH configuration:"
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewUpdatePlanTool())
-
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		planTurn("in_progress", "pending", "pending"),
-		textTurn(cue), textTurn(cue), textTurn(cue), textTurn(cue), textTurn(cue),
-	}}
-
-	result, err := Run(context.Background(), "set up a git server", provider, Options{
-		Registry: registry,
-		MaxTurns: 10,
-		// RequireCompletionSignal deliberately off — interactive/TUI default.
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if !result.Incomplete {
-		t.Fatalf("expected Incomplete=true (plan stalled without headless gate), got false; final=%q turns=%d", result.FinalAnswer, result.Turns)
-	}
-	if len(provider.requests) != 1+maxContinueNudges+1 {
-		t.Fatalf("expected %d provider turns (1 plan + %d nudges + 1 final), got %d",
-			1+maxContinueNudges+1, maxContinueNudges, len(provider.requests))
-	}
-	if !someRequestContains(provider.requests, continueNudgeMarker) {
-		t.Fatalf("expected a continue nudge (%q) to be injected", continueNudgeMarker)
-	}
-}
-
-// A pending plan must nudge non-colon continuation announcements too — not only
-// colon-terminated cues — so the first mid-step text turn cannot finalize while
-// steps remain. A later clean final answer is accepted after the bounded nudges.
-func TestPlanPendingNudgesNonColonContinuationText(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewUpdatePlanTool())
-	midStep := "Let me inspect the configuration."
-	final := "Provider profiles are loaded from config and merged with defaults."
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		planTurn("in_progress", "pending", "pending"),
-		textTurn(midStep),
-		textTurn(final), textTurn(final), textTurn(final), textTurn(final),
-	}}
-
-	result, err := Run(context.Background(), "inspect and configure", provider, Options{
-		Registry: registry,
-		MaxTurns: 10,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.Incomplete {
-		t.Fatalf("expected success after nudge + final answer, got Incomplete (%q)", result.IncompleteReason)
-	}
-	if result.FinalAnswer != final {
-		t.Fatalf("final answer = %q, want %q", result.FinalAnswer, final)
-	}
-	if len(provider.requests) != 1+1+maxContinueNudges {
-		t.Fatalf("expected %d turns (plan + mid-step + %d nudged finals), got %d",
-			1+1+maxContinueNudges, maxContinueNudges, len(provider.requests))
-	}
-	if !someRequestContains(provider.requests, continueNudgeMarker) {
-		t.Fatalf("expected a continue nudge after the mid-step text with pending plan")
-	}
-}
-
-// After bounded nudges, a confident final answer with stale plan bookkeeping must
-// still succeed on the interactive/TUI path (no infinite loop, no false incomplete).
-func TestPlanPendingAcceptsFinalAnswerAfterBoundedNudgesWithoutHeadlessGate(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewUpdatePlanTool())
-	done := "All provider profiles are documented above."
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		planTurn("in_progress", "pending"),
-		textTurn(done), textTurn(done), textTurn(done), textTurn(done),
-	}}
-
-	result, err := Run(context.Background(), "explain provider loading", provider, Options{
-		Registry: registry,
-		MaxTurns: 10,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.Incomplete {
-		t.Fatalf("stale plan alone must not force Incomplete on TUI path; reason=%q", result.IncompleteReason)
-	}
-	if result.FinalAnswer != done {
-		t.Fatalf("final answer = %q, want %q", result.FinalAnswer, done)
-	}
-	if !someRequestContains(provider.requests, continueNudgeMarker) {
-		t.Fatalf("expected at least one continue nudge before accepting completion")
-	}
-}
-
-// With the gate OFF and NO pending plan, a continuation-cue turn is still
-// accepted as the final answer — no behavior change for short single-step tasks.
-func TestCompletionGateOffPreservesLegacyBehavior(t *testing.T) {
-	cue := "Let me check the config:"
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		textTurn(cue),
-	}}
-
-	result, err := Run(context.Background(), "do a thing", provider, Options{
-		Registry: tools.NewRegistry(),
-		MaxTurns: 10,
-		// RequireCompletionSignal deliberately left false.
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.Incomplete {
-		t.Fatalf("legacy path must never set Incomplete; final=%q", result.FinalAnswer)
-	}
-	if result.FinalAnswer != cue {
-		t.Fatalf("final answer = %q, want %q (legacy: text-only turn is the answer)", result.FinalAnswer, cue)
-	}
-	if len(provider.requests) != 1 {
-		t.Fatalf("legacy path must not re-prompt; got %d turns", len(provider.requests))
-	}
-}
-
-// review #6: the continuation-cue detector must catch a mid-line action announcement
-// that stops on a colon (the git-multibranch case), without flagging genuine closers
-// — a recommendation, a plain summary colon, or a sign-off.
-func TestContinuationCueMatching(t *testing.T) {
-	cases := []struct {
-		text string
-		cue  bool
-	}{
-		{"Now I need to configure the SSH server. Let me check the current SSH configuration:", true},
-		{"Let me read the file:", true},
-		{"Now I'll run the tests:", true},
-		{"Next, I suggest reviewing the changes.", false}, // recommendation, no colon
-		{"Here is the summary:", false},                   // summary colon, no action lead-in
-		{"Let me know if you need anything:", false},      // sign-off
-		{"The function is implemented and all tests pass.", false},
-	}
-	for _, c := range cases {
-		if got := endsWithContinuationCue(c.text); got != c.cue {
-			t.Errorf("endsWithContinuationCue(%q) = %v, want %v", c.text, got, c.cue)
-		}
-	}
-}
-
-// Cancellation during a plan-pending stall must return immediately without extra
-// nudges or incomplete synthesis.
-func TestRunCancellationDuringPlanPendingStopsImmediately(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewUpdatePlanTool())
-	ctx, cancel := context.WithCancel(context.Background())
-	_, err := Run(ctx, "do work", cancelMidStreamProvider{cancel: cancel}, Options{
-		Registry: registry,
-	})
-	if !errors.Is(err, context.Canceled) {
-		t.Fatalf("expected context.Canceled, got %v", err)
-	}
-}
-
-// Repeated identical update_plan calls must still hit the max-turns ceiling rather
-// than looping silently forever.
-func TestRunRepeatedIdenticalUpdatePlanHitsMaxTurns(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewUpdatePlanTool())
-	same := toolTurn("plan", "update_plan", `{"plan":[{"content":"step","status":"in_progress"}]}`)
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{same, same, same, textTurn("halted summary")}}
-
-	result, err := Run(context.Background(), "keep planning", provider, Options{
-		Registry: registry,
-		MaxTurns: 3,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	// MaxTurns=3 tool turns, then finalAnswerAfterMaxTurns issues one more request.
-	if len(provider.requests) != 4 {
-		t.Fatalf("expected 3 tool turns + 1 max-turns final-answer call, got %d", len(provider.requests))
-	}
-	if result.FinalAnswer == "" {
-		t.Fatal("expected a max-turns final answer")
-	}
-}
-
-// review #4: a run that loops to the MaxTurns ceiling (always calling a tool, so it
-// never reaches the no-tool-call gate) was reported as success. Under the headless
-// gate, a max-turns cutoff is INCOMPLETE — the agent was stopped mid-run, not done.
-func TestMaxTurnsCutoffIsIncompleteUnderGate(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewUpdatePlanTool())
-	toolEvery := toolTurn("c", "update_plan", `{"plan":[{"content":"step","status":"in_progress"}]}`)
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		toolEvery, toolEvery, toolEvery, toolEvery,
-	}}
-
-	result, err := Run(context.Background(), "keep working", provider, Options{
-		Registry:                registry,
-		MaxTurns:                2,
-		RequireCompletionSignal: true,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if !result.Incomplete {
-		t.Fatalf("a max-turns cutoff under the gate must be Incomplete; final=%q", result.FinalAnswer)
-	}
-	if !strings.Contains(result.IncompleteReason, "max-turns") {
-		t.Fatalf("IncompleteReason = %q, want it to cite max-turns", result.IncompleteReason)
-	}
-}

internal/agent/confirmation_policy.md

@@ -81,11 +81,3 @@ Otherwise, confirm before executing.
 6. Avoid redundant confirmations if you already confirmed and there is no material new risk.
 7. Write/shell operations inside the workspace do not need the same scrutiny as operations outside it.
 8. Interactive programs (editors, pagers, REPLs, `ssh`/`psql`/`mysql` without a command, `top`/`htop`, `git rebase -i`) will be blocked because they hang the agent; always use the non-interactive alternative.
-
-## Sandbox Permission Requests
-
-- Prefer sandboxed additional permissions over full escalation: use `sandbox_permissions: "with_additional_permissions"` with the required file-system or network permission when that can satisfy the command.
-- Use `sandbox_permissions: "require_escalated"` only when the command must run outside the sandbox, such as inspecting or controlling host/global state that the sandbox intentionally hides.
-- If a sandboxed shell command returns only sandbox-local state when the user asked for host/global state, do not answer from that misleading output. Rerun the needed command with `sandbox_permissions: "require_escalated"`, a short user-facing `justification`, and a narrow `prefix_rule` when appropriate.
-- For `require_escalated`, include a short user-facing `justification`. When a reusable approval is appropriate, include a narrow `prefix_rule`; do not suggest broad shell, interpreter, or launcher prefixes.
-- Shell commands are evaluated as independent command segments across `|`, `&&`, `||`, and `;`. A reusable prefix must apply safely to the command segment it is meant to authorize.

internal/agent/context_measurement.go

@@ -1,66 +0,0 @@
-package agent
-
-import (
-	"encoding/json"
-
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-// Context budget.
-//
-// MeasureContext breaks a request's estimated token footprint into the
-// categories that actually compete for the window — the system prompt, the
-// advertised tool definitions, and the conversation messages — so the agent can
-// report context utilization (e.g. "62% used: 4k system, 6k tools, 30k history")
-// and reason about what to compact. Counts are estimates (see estimateTokens),
-// not a real tokenizer: good for proportions and budget bars, not billing.
-
-// ContextBreakdown is a per-category estimate of a request's token footprint.
-type ContextBreakdown struct {
-	SystemTokens  int     // leading system messages (prompt + injected context)
-	ToolTokens    int     // advertised tool definitions
-	MessageTokens int     // non-system conversation messages
-	TotalTokens   int     // SystemTokens + ToolTokens + MessageTokens
-	ContextWindow int     // model context window; 0 when unknown
-	UsedFraction  float64 // TotalTokens / ContextWindow; 0 when window unknown
-}
-
-// MeasureContext estimates the per-category token footprint of a request: the
-// leading system messages, the advertised tool definitions, and the remaining
-// conversation messages.
-func MeasureContext(messages []zeroruntime.Message, tools []zeroruntime.ToolDefinition, contextWindow int) ContextBreakdown {
-	systemEnd := 0
-	for systemEnd < len(messages) && messages[systemEnd].Role == zeroruntime.MessageRoleSystem {
-		systemEnd++
-	}
-
-	breakdown := ContextBreakdown{
-		SystemTokens:  estimateTokens(messages[:systemEnd]),
-		MessageTokens: estimateTokens(messages[systemEnd:]),
-		ToolTokens:    estimateToolTokens(tools),
-		ContextWindow: contextWindow,
-	}
-	breakdown.TotalTokens = breakdown.SystemTokens + breakdown.ToolTokens + breakdown.MessageTokens
-	if contextWindow > 0 {
-		breakdown.UsedFraction = float64(breakdown.TotalTokens) / float64(contextWindow)
-	}
-	return breakdown
-}
-
-// estimateToolTokens approximates the token footprint of advertised tool
-// definitions (name + description + JSON schema), using the same ApproxTextTokens
-// heuristic as estimateTokens so all categories share one scale.
-func estimateToolTokens(tools []zeroruntime.ToolDefinition) int {
-	total := 0
-	for _, tool := range tools {
-		total += ApproxTextTokens(tool.Name)
-		total += ApproxTextTokens(tool.Description)
-		if len(tool.Parameters) > 0 {
-			if encoded, err := json.Marshal(tool.Parameters); err == nil {
-				total += ApproxTextTokens(string(encoded))
-			}
-		}
-		total += 4 // per-tool overhead
-	}
-	return total
-}

internal/agent/context_measurement_test.go

@@ -1,68 +0,0 @@
-package agent
-
-import (
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-func TestMeasureContextSplitsByCategory(t *testing.T) {
-	messages := []zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleSystem, Content: filler(2000)},
-		{Role: zeroruntime.MessageRoleUser, Content: filler(400)},
-		{Role: zeroruntime.MessageRoleAssistant, Content: filler(400)},
-	}
-	tools := []zeroruntime.ToolDefinition{
-		{Name: "read_file", Description: filler(200), Parameters: map[string]any{"type": "object"}},
-	}
-
-	breakdown := MeasureContext(messages, tools, 100_000)
-
-	if breakdown.SystemTokens <= 0 || breakdown.MessageTokens <= 0 || breakdown.ToolTokens <= 0 {
-		t.Fatalf("expected all categories non-zero, got %#v", breakdown)
-	}
-	if got := breakdown.SystemTokens + breakdown.ToolTokens + breakdown.MessageTokens; got != breakdown.TotalTokens {
-		t.Fatalf("TotalTokens %d != sum of categories %d", breakdown.TotalTokens, got)
-	}
-	// System content (2000 chars) dominates the conversation (800 chars total).
-	if breakdown.SystemTokens <= breakdown.MessageTokens {
-		t.Fatalf("expected system to dominate messages, got system=%d messages=%d", breakdown.SystemTokens, breakdown.MessageTokens)
-	}
-	wantFraction := float64(breakdown.TotalTokens) / 100_000
-	if breakdown.UsedFraction != wantFraction {
-		t.Fatalf("UsedFraction = %v, want %v", breakdown.UsedFraction, wantFraction)
-	}
-}
-
-func TestMeasureContextUnknownWindowHasZeroFraction(t *testing.T) {
-	breakdown := MeasureContext([]zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleUser, Content: "hi"},
-	}, nil, 0)
-	if breakdown.UsedFraction != 0 {
-		t.Fatalf("UsedFraction = %v, want 0 when context window unknown", breakdown.UsedFraction)
-	}
-	if breakdown.ToolTokens != 0 {
-		t.Fatalf("ToolTokens = %d, want 0 with no tools", breakdown.ToolTokens)
-	}
-}
-
-func TestMeasureContextNoSystemMessage(t *testing.T) {
-	breakdown := MeasureContext([]zeroruntime.Message{
-		{Role: zeroruntime.MessageRoleUser, Content: filler(400)},
-	}, nil, 1000)
-	if breakdown.SystemTokens != 0 {
-		t.Fatalf("SystemTokens = %d, want 0 when there is no system message", breakdown.SystemTokens)
-	}
-	if breakdown.MessageTokens <= 0 {
-		t.Fatalf("MessageTokens = %d, want > 0", breakdown.MessageTokens)
-	}
-}
-
-// filler returns a string of n characters for sizing assertions.
-func filler(n int) string {
-	b := make([]byte, n)
-	for i := range b {
-		b[i] = 'x'
-	}
-	return string(b)
-}

internal/agent/decode_args_test.go

@@ -1,99 +0,0 @@
-package agent
-
-import "testing"
-
-// decodeToolArguments must (a) decode a normal single-object payload unchanged,
-// (b) tolerate a weak model packing multiple concatenated top-level JSON objects
-// into one arguments string by decoding the FIRST object and ignoring the rest
-// (the minimax-m3 "invalid character '{' after top-level value" failure), and
-// (c) still reject a genuinely malformed/truncated first object.
-func TestDecodeToolArguments(t *testing.T) {
-	t.Run("single object decodes normally", func(t *testing.T) {
-		var a map[string]any
-		if err := decodeToolArguments(`{"path":"x"}`, &a); err != nil {
-			t.Fatalf("err = %v, want nil", err)
-		}
-		if a["path"] != "x" {
-			t.Fatalf("decoded %v, want path=x", a)
-		}
-	})
-
-	t.Run("concatenated objects decode the first, ignore trailing", func(t *testing.T) {
-		var b map[string]any
-		if err := decodeToolArguments(`{"path":"first"}{"path":"second"}`, &b); err != nil {
-			t.Fatalf("multi-object should decode the first, got err = %v", err)
-		}
-		if b["path"] != "first" {
-			t.Fatalf("decoded %v, want the first object (path=first)", b)
-		}
-	})
-
-	t.Run("distinct tools' args concatenated (real minimax case): first object used", func(t *testing.T) {
-		var c map[string]any
-		if err := decodeToolArguments(`{"plan":[1,2]}{"cmd":"go version"}`, &c); err != nil {
-			t.Fatalf("err = %v, want nil", err)
-		}
-		if _, ok := c["plan"]; !ok || c["cmd"] != nil {
-			t.Fatalf("decoded %v, want only the first object's keys", c)
-		}
-	})
-
-	t.Run("truncated first object still errors", func(t *testing.T) {
-		var d map[string]any
-		if err := decodeToolArguments(`{"path":`, &d); err == nil {
-			t.Fatal("truncated JSON must still error")
-		}
-	})
-
-	t.Run("non-JSON trailing garbage still errors (not masked)", func(t *testing.T) {
-		var g map[string]any
-		if err := decodeToolArguments(`{"x":1}xyz`, &g); err == nil {
-			t.Fatal("a valid first object followed by non-JSON garbage must still error")
-		}
-	})
-
-	t.Run("recoverableToolArguments: only whole-JSON trailing is tolerated", func(t *testing.T) {
-		cases := []struct {
-			in    string
-			ok    bool
-			first string
-		}{
-			{`{"a":1}`, true, `{"a":1}`},
-			{`{"a":1}{"b":2}`, true, `{"a":1}`},
-			{`{"a":1}  {"b":2}  `, true, `{"a":1}`},
-			{`{"a":1}xyz`, false, ""},
-			{`{"a":`, false, ""},
-			{`   `, false, ""},
-		}
-		for _, c := range cases {
-			first, ok := recoverableToolArguments(c.in)
-			if ok != c.ok || (ok && first != c.first) {
-				t.Fatalf("recoverableToolArguments(%q) = (%q,%v), want (%q,%v)", c.in, first, ok, c.first, c.ok)
-			}
-		}
-	})
-
-	t.Run("historySafeToolCalls recovers the first object, not {}", func(t *testing.T) {
-		out := historySafeToolCalls([]ToolCall{
-			{ID: "c1", Name: "read_file", Arguments: `{"path":"README.md"}{"path":"AGENTS.md"}`}, // concatenated -> first
-			{ID: "c2", Name: "read_file", Arguments: `{"path":"x"}xyz`},                          // garbage -> {}
-			{ID: "c3", Name: "read_file", Arguments: `{"path":"ok"}`},                            // valid -> unchanged
-		})
-		if out[0].Arguments != `{"path":"README.md"}` {
-			t.Fatalf("concatenated args should record the first object, got %q", out[0].Arguments)
-		}
-		if out[1].Arguments != "{}" {
-			t.Fatalf("garbage-trailing args should record {}, got %q", out[1].Arguments)
-		}
-		if out[2].Arguments != `{"path":"ok"}` {
-			t.Fatalf("valid args should be unchanged, got %q", out[2].Arguments)
-		}
-	})
-
-	t.Run("whitespace-only errors (caller guards empty)", func(t *testing.T) {
-		var e map[string]any
-		if err := decodeToolArguments("   ", &e); err == nil {
-			t.Fatal("whitespace-only must error")
-		}
-	})
-}

internal/agent/defer_savings_test.go

@@ -1,83 +0,0 @@
-package agent
-
-import (
-	"context"
-	"encoding/json"
-	"fmt"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/tools"
-)
-
-// richMCPTool mimics a real MCP tool's schema (Exa/Firecrawl-style search): an
-// object with several typed, described properties. Real MCP tools are this size
-// or larger, which is why shipping their full schemas every turn is expensive.
-type richMCPTool struct{ name string }
-
-func (t richMCPTool) Name() string { return t.name }
-func (t richMCPTool) Description() string {
-	return "Search the web and return ranked, structured results with titles, URLs, snippets, and optional cleaned full-page content. Supports natural-language and keyword queries with domain and date filtering."
-}
-func (t richMCPTool) Parameters() tools.Schema {
-	return tools.Schema{
-		Type:                 "object",
-		AdditionalProperties: false,
-		Required:             []string{"query"},
-		Properties: map[string]tools.PropertySchema{
-			"query":                {Type: "string", Description: "The search query to run against the web index. Supports natural-language and keyword queries."},
-			"num_results":          {Type: "integer", Description: "Maximum number of results to return (1-100). Defaults to 10."},
-			"include_domains":      {Type: "array", Description: "Only return results from these domains.", Items: &tools.PropertySchema{Type: "string"}},
-			"exclude_domains":      {Type: "array", Description: "Never return results from these domains.", Items: &tools.PropertySchema{Type: "string"}},
-			"start_published_date": {Type: "string", Description: "ISO-8601 date; only results published on or after this date are returned."},
-			"end_published_date":   {Type: "string", Description: "ISO-8601 date; only results published on or before this date are returned."},
-			"category":             {Type: "string", Description: "Restrict results to a content category.", Enum: []string{"news", "research paper", "company", "tweet", "pdf"}},
-			"include_text":         {Type: "boolean", Description: "Include cleaned full-page text for each result (slower, larger payload)."},
-		},
-	}
-}
-func (t richMCPTool) Safety() tools.Safety {
-	return tools.Safety{SideEffect: tools.SideEffectNetwork, Permission: tools.PermissionAllow}
-}
-func (t richMCPTool) Run(_ context.Context, _ map[string]any) tools.Result {
-	return tools.Result{Status: tools.StatusOK}
-}
-func (t richMCPTool) Deferred() bool { return true }
-
-// TestDeferralTokenSavingsMeasurement quantifies the win from the lowered
-// deferThreshold default: with a typical small MCP server (6 tools), threshold 10
-// (the OLD default) kept deferral OFF so all six full schemas shipped every turn,
-// while threshold 3 (the NEW default) collapses them to compact reminder lines.
-func TestDeferralTokenSavingsMeasurement(t *testing.T) {
-	const mcpToolCount = 6
-	registry := tools.NewRegistry()
-	for i := 0; i < mcpToolCount; i++ {
-		registry.Register(richMCPTool{name: fmt.Sprintf("mcp__websearch__search_v%d", i)})
-	}
-	registry.Register(fakeToolSearchTool{})
-
-	measure := func(threshold int) (int, int) {
-		defs, reminder := partitionTools(registry, PermissionModeAuto, Options{DeferThreshold: threshold}, map[string]bool{})
-		encoded, err := json.Marshal(defs)
-		if err != nil {
-			t.Fatalf("marshal tool defs: %v", err)
-		}
-		bytes := len(encoded) + len(reminder)
-		return bytes, bytes / 4 // ~tokens (approx 4 chars/token)
-	}
-
-	eagerBytes, eagerTok := measure(10) // 6 < 10 -> inactive -> all schemas eager
-	deferBytes, deferTok := measure(3)  // 6 >= 3 -> active  -> compact reminder
-	savedPct := 100 * float64(eagerBytes-deferBytes) / float64(eagerBytes)
-
-	t.Logf("MCP tools in set: %d", mcpToolCount)
-	t.Logf("threshold 10 (OLD, eager schemas): %d bytes (~%d tokens) per turn", eagerBytes, eagerTok)
-	t.Logf("threshold  3 (NEW, deferred lines): %d bytes (~%d tokens) per turn", deferBytes, deferTok)
-	t.Logf("SAVED: ~%d tokens per turn (%.0f%% smaller tool payload)", eagerTok-deferTok, savedPct)
-
-	if deferBytes >= eagerBytes {
-		t.Fatalf("deferral should shrink the tool payload: eager=%d defer=%d", eagerBytes, deferBytes)
-	}
-	if savedPct < 50 {
-		t.Fatalf("expected a large reduction, got only %.0f%%", savedPct)
-	}
-}

internal/agent/deferred_loop_test.go

@@ -1,800 +0,0 @@
-package agent
-
-import (
-	"context"
-	"errors"
-	"reflect"
-	"sort"
-	"strings"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/tools"
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-func TestOptionsDeferThresholdFieldExists(t *testing.T) {
-	options := Options{DeferThreshold: 10}
-	if options.DeferThreshold != 10 {
-		t.Fatalf("expected DeferThreshold 10, got %d", options.DeferThreshold)
-	}
-}
-
-func TestToolResultLoadedToolsField(t *testing.T) {
-	result := ToolResult{LoadedTools: []string{"Alpha", "Beta"}}
-	if len(result.LoadedTools) != 2 || result.LoadedTools[0] != "Alpha" || result.LoadedTools[1] != "Beta" {
-		t.Fatalf("expected LoadedTools [Alpha Beta], got %#v", result.LoadedTools)
-	}
-	// Default zero value is nil for an ordinary result.
-	if (ToolResult{}).LoadedTools != nil {
-		t.Fatalf("expected nil LoadedTools by default")
-	}
-}
-
-// loadSignalTool returns Meta["load_tools"] like tool_search does, so we can
-// assert executeToolCall lifts it into ToolResult.LoadedTools.
-type loadSignalTool struct{ value string }
-
-func (t loadSignalTool) Name() string        { return "load_signal" }
-func (t loadSignalTool) Description() string { return "emits a load_tools signal" }
-func (t loadSignalTool) Parameters() tools.Schema {
-	return tools.Schema{Type: "object", AdditionalProperties: false}
-}
-func (t loadSignalTool) Safety() tools.Safety {
-	return tools.Safety{SideEffect: tools.SideEffectRead, Permission: tools.PermissionAllow}
-}
-func (t loadSignalTool) Run(_ context.Context, _ map[string]any) tools.Result {
-	return tools.Result{Status: tools.StatusOK, Output: "ok"}
-}
-func (t loadSignalTool) RunWithOptions(_ context.Context, _ map[string]any, _ tools.RunOptions) tools.Result {
-	return tools.Result{Status: tools.StatusOK, Output: "ok", Meta: map[string]string{"load_tools": t.value}}
-}
-
-func TestExecuteToolCallLiftsLoadTools(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(loadSignalTool{value: " Alpha , Beta ,, "})
-
-	result, abortErr := executeToolCall(
-		context.Background(),
-		registry,
-		ToolCall{ID: "c1", Name: "load_signal", Arguments: ""},
-		PermissionModeAuto,
-		Options{},
-	)
-	if abortErr != nil {
-		t.Fatalf("unexpected abort error: %v", abortErr)
-	}
-	want := []string{"Alpha", "Beta"}
-	if len(result.LoadedTools) != len(want) {
-		t.Fatalf("expected LoadedTools %#v, got %#v", want, result.LoadedTools)
-	}
-	for i := range want {
-		if result.LoadedTools[i] != want[i] {
-			t.Fatalf("expected LoadedTools %#v, got %#v", want, result.LoadedTools)
-		}
-	}
-}
-
-func TestExecuteToolCallNoLoadToolsMetaLeavesNil(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(secretEmittingTool{output: "plain"})
-
-	result, abortErr := executeToolCall(
-		context.Background(),
-		registry,
-		ToolCall{ID: "c1", Name: "leak", Arguments: ""},
-		PermissionModeAuto,
-		Options{},
-	)
-	if abortErr != nil {
-		t.Fatalf("unexpected abort error: %v", abortErr)
-	}
-	if result.LoadedTools != nil {
-		t.Fatalf("expected nil LoadedTools for a tool with no load_tools meta, got %#v", result.LoadedTools)
-	}
-}
-
-// fakeDeferredTool is deferred-eligible (implements Deferred() bool) like an MCP
-// tool wrapper, so partitionTools counts and (when active) hides it.
-type fakeDeferredTool struct {
-	name string
-	desc string
-}
-
-func (t fakeDeferredTool) Name() string        { return t.name }
-func (t fakeDeferredTool) Description() string { return t.desc }
-func (t fakeDeferredTool) Parameters() tools.Schema {
-	return tools.Schema{Type: "object", AdditionalProperties: false}
-}
-func (t fakeDeferredTool) Safety() tools.Safety {
-	return tools.Safety{SideEffect: tools.SideEffectRead, Permission: tools.PermissionAllow}
-}
-func (t fakeDeferredTool) Run(_ context.Context, _ map[string]any) tools.Result {
-	return tools.Result{Status: tools.StatusOK, Output: "ok"}
-}
-func (t fakeDeferredTool) Deferred() bool { return true }
-
-// fakeToolSearchTool stands in for component D's tool_search (a non-deferred
-// builtin) so the inactive path can assert it is dropped.
-type fakeToolSearchTool struct{}
-
-func (fakeToolSearchTool) Name() string        { return "tool_search" }
-func (fakeToolSearchTool) Description() string { return "load deferred tool schemas" }
-func (fakeToolSearchTool) Parameters() tools.Schema {
-	return tools.Schema{Type: "object", AdditionalProperties: false}
-}
-func (fakeToolSearchTool) Safety() tools.Safety {
-	return tools.Safety{SideEffect: tools.SideEffectNone, Permission: tools.PermissionAllow, AdvertiseInAuto: true}
-}
-func (fakeToolSearchTool) Run(_ context.Context, _ map[string]any) tools.Result {
-	return tools.Result{Status: tools.StatusOK, Output: "ok"}
-}
-
-func toolDefinitionByName(definitions []zeroruntime.ToolDefinition, name string) *zeroruntime.ToolDefinition {
-	for i := range definitions {
-		if definitions[i].Name == name {
-			return &definitions[i]
-		}
-	}
-	return nil
-}
-
-func assertNoDeferredDiscoveryMessage(t *testing.T, request zeroruntime.CompletionRequest) {
-	t.Helper()
-	for _, message := range request.Messages {
-		if message.Role == zeroruntime.MessageRoleUser && strings.Contains(message.Content, "Deferred tools:") {
-			t.Fatalf("deferred discovery must not be appended as a user message: %q", message.Content)
-		}
-	}
-}
-
-func TestPartitionToolsInactiveIsByteIdenticalAndDropsToolSearch(t *testing.T) {
-	root := t.TempDir()
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewReadFileTool(root))
-	registry.Register(fakeDeferredTool{name: "mcp__srv__a", desc: "tool a"})
-	registry.Register(fakeToolSearchTool{})
-
-	options := Options{DeferThreshold: 0} // 0 => deferral disabled => inactive path.
-
-	// DeferThreshold 0 => deferral disabled => inactive path.
-	exposed, reminder := partitionTools(registry, PermissionModeAuto, options, map[string]bool{})
-
-	if reminder != "" {
-		t.Fatalf("expected empty reminder on inactive path, got %q", reminder)
-	}
-
-	// Strong byte-identity assertion: build the reference the LEGACY way (the old
-	// toolDefinitions construction — every visible/advertised tool's full schema,
-	// EXCEPT tool_search, alpha-sorted by name) and require an exact DeepEqual. This
-	// pins that the inactive partition produces the pre-deferral output verbatim,
-	// not merely the same set of names.
-	reference := legacyToolDefinitions(registry, PermissionModeAuto, options)
-	if !reflect.DeepEqual(exposed, reference) {
-		t.Fatalf("inactive partition not byte-identical to legacy toolDefinitions:\n got %#v\nwant %#v", exposed, reference)
-	}
-
-	// Belt-and-suspenders: tool_search is dropped, the deferred tool keeps its full
-	// schema, and only the expected names are present.
-	for _, def := range exposed {
-		if def.Name == "tool_search" {
-			t.Fatalf("tool_search must be dropped on inactive path, got %#v", exposed)
-		}
-	}
-	wantNames := map[string]bool{"read_file": true, "mcp__srv__a": true}
-	if len(exposed) != len(wantNames) {
-		t.Fatalf("expected %d exposed tools, got %d: %#v", len(wantNames), len(exposed), exposed)
-	}
-	for _, def := range exposed {
-		if !wantNames[def.Name] {
-			t.Fatalf("unexpected exposed tool %q", def.Name)
-		}
-		if def.Name == "mcp__srv__a" && def.Parameters["type"] != "object" {
-			t.Fatalf("expected full schema for deferred tool on inactive path, got %#v", def.Parameters)
-		}
-	}
-}
-
-// legacyToolDefinitions reconstructs the PRE-deferral tool-list builder: every
-// tool that is visible (passes the operator filters) and advertised for the mode,
-// EXCEPT tool_search, rendered with its full schema and alpha-sorted by name. It
-// is the byte-identity reference for the inactive partition path.
-func legacyToolDefinitions(registry *tools.Registry, permissionMode PermissionMode, options Options) []zeroruntime.ToolDefinition {
-	definitions := make([]zeroruntime.ToolDefinition, 0)
-	for _, tool := range registry.All() {
-		if !ToolVisible(tool, permissionMode, options.EnabledTools, options.DisabledTools) {
-			continue
-		}
-		if tool.Name() == tools.ToolSearchToolName {
-			continue
-		}
-		definitions = append(definitions, zeroruntime.ToolDefinition{
-			Name:        tool.Name(),
-			Description: tool.Description(),
-			Parameters:  schemaToRuntimeMap(tool.Parameters()),
-		})
-	}
-	sort.Slice(definitions, func(left, right int) bool {
-		return definitions[left].Name < definitions[right].Name
-	})
-	return definitions
-}
-
-// Below-threshold-but-eligible (count < threshold) is also inactive.
-func TestPartitionToolsBelowThresholdInactive(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(fakeDeferredTool{name: "mcp__srv__a", desc: "a"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__b", desc: "b"})
-
-	exposed, reminder := partitionTools(registry, PermissionModeAuto, Options{DeferThreshold: 10}, map[string]bool{})
-	if reminder != "" {
-		t.Fatalf("expected empty reminder below threshold, got %q", reminder)
-	}
-	if len(exposed) != 2 {
-		t.Fatalf("expected both deferred tools exposed below threshold, got %#v", exposed)
-	}
-}
-
-// FIX 6(a): a deferred tool removed by DisabledTools must NOT count toward the
-// eligible total. With N deferred registered and threshold == N, disabling one
-// drops the surviving eligible count to N-1 < threshold, so the partition takes
-// the INACTIVE path (empty discovery text, all VISIBLE tools exposed with full
-// schemas, the disabled one filtered out entirely).
-func TestPartitionToolsDisabledDeferredDropsBelowThresholdInactive(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(fakeDeferredTool{name: "mcp__srv__alpha", desc: "alpha"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__beta", desc: "beta"})
-
-	exposed, reminder := partitionTools(registry, PermissionModeAuto, Options{
-		DeferThreshold: 2,
-		DisabledTools:  []string{"mcp__srv__beta"},
-	}, map[string]bool{})
-
-	// Eligible drops to 1 (< threshold 2) => inactive: empty reminder.
-	if reminder != "" {
-		t.Fatalf("expected inactive path (empty reminder) when a disable drops eligible below threshold, got %q", reminder)
-	}
-	// Only the surviving visible tool is exposed, with its FULL schema; the disabled
-	// tool is filtered out entirely.
-	if len(exposed) != 1 || exposed[0].Name != "mcp__srv__alpha" {
-		t.Fatalf("expected only mcp__srv__alpha exposed on inactive path, got %#v", exposed)
-	}
-	if exposed[0].Parameters["type"] != "object" {
-		t.Fatalf("surviving deferred tool must keep its full schema on inactive path, got %#v", exposed[0].Parameters)
-	}
-}
-
-// FIX 6(b): with deferral ACTIVE, a DisabledTools-hidden deferred tool must never
-// influence discovery NOR be exposed — it is filtered out before the partition
-// even considers it.
-func TestPartitionToolsActiveExcludesDisabledDeferredFromDiscoveryAndExposed(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(fakeDeferredTool{name: "mcp__srv__alpha", desc: "alpha"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__beta", desc: "beta"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__gamma", desc: "gamma"})
-	registry.Register(fakeToolSearchTool{}) // usable loader => deferral can activate
-
-	// 3 deferred, disable beta => 2 surviving eligible, threshold 2 => active.
-	exposed, discovery := partitionTools(registry, PermissionModeAuto, Options{
-		DeferThreshold: 2,
-		DisabledTools:  []string{"mcp__srv__beta"},
-	}, map[string]bool{})
-
-	if discovery == "" {
-		t.Fatalf("expected active path with discovery text for the unloaded tools")
-	}
-	if strings.Contains(discovery, "mcp__srv__beta") {
-		t.Fatalf("disabled deferred tool must not appear in discovery text, got %q", discovery)
-	}
-	for _, def := range exposed {
-		if def.Name == "mcp__srv__beta" {
-			t.Fatalf("disabled deferred tool must not be exposed, got %#v", exposed)
-		}
-	}
-	search := toolDefinitionByName(exposed, tools.ToolSearchToolName)
-	if search == nil {
-		t.Fatalf("tool_search must be exposed on active path, got %#v", exposed)
-	}
-	if search.Description != discovery {
-		t.Fatalf("tool_search description must carry discovery text\n got: %q\nwant: %q", search.Description, discovery)
-	}
-	if !strings.Contains(discovery, "mcp") {
-		t.Fatalf("discovery text must list the surviving deferred source, got %q", discovery)
-	}
-}
-
-func TestPartitionToolsActiveHidesUnloadedExposesLoaded(t *testing.T) {
-	root := t.TempDir()
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewReadFileTool(root)) // non-deferred builtin
-	registry.Register(fakeToolSearchTool{})        // non-deferred, must stay exposed
-	registry.Register(fakeDeferredTool{name: "mcp__srv__alpha", desc: "alpha tool"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__beta", desc: "beta tool"})
-
-	loaded := map[string]bool{"mcp__srv__alpha": true}
-
-	// 2 eligible deferred tools, threshold 2 => active.
-	exposed, discovery := partitionTools(registry, PermissionModeAuto, Options{DeferThreshold: 2}, loaded)
-
-	exposedNames := map[string]bool{}
-	for _, def := range exposed {
-		exposedNames[def.Name] = true
-	}
-	if !exposedNames["read_file"] {
-		t.Fatalf("expected builtin read_file exposed, got %#v", exposed)
-	}
-	if !exposedNames["tool_search"] {
-		t.Fatalf("expected tool_search exposed on active path, got %#v", exposed)
-	}
-	if !exposedNames["mcp__srv__alpha"] {
-		t.Fatalf("expected loaded deferred tool exposed, got %#v", exposed)
-	}
-	if exposedNames["mcp__srv__beta"] {
-		t.Fatalf("unloaded deferred tool must be hidden from exposed, got %#v", exposed)
-	}
-	if discovery == "" {
-		t.Fatalf("expected non-empty discovery text for the hidden tool")
-	}
-	search := toolDefinitionByName(exposed, tools.ToolSearchToolName)
-	if search == nil {
-		t.Fatalf("expected tool_search exposed on active path, got %#v", exposed)
-	}
-	if search.Description != discovery || !strings.Contains(search.Description, "mcp") {
-		t.Fatalf("tool_search description must carry discovery source, got %q", search.Description)
-	}
-	if strings.Contains(search.Description, "mcp__srv__alpha") {
-		t.Fatalf("tool_search description must not list already-loaded tool names, got %q", search.Description)
-	}
-	if !strings.Contains(search.Description, "mcp__srv__beta") {
-		t.Fatalf("tool_search description must list exact hidden tool names, got %q", search.Description)
-	}
-}
-
-func TestPartitionToolsActiveNothingHiddenEmptyReminder(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(fakeDeferredTool{name: "mcp__srv__alpha", desc: "alpha"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__beta", desc: "beta"})
-	registry.Register(fakeToolSearchTool{}) // usable loader => deferral can activate
-
-	loaded := map[string]bool{"mcp__srv__alpha": true, "mcp__srv__beta": true}
-	exposed, reminder := partitionTools(registry, PermissionModeAuto, Options{DeferThreshold: 2}, loaded)
-
-	exposedNames := map[string]bool{}
-	for _, def := range exposed {
-		exposedNames[def.Name] = true
-	}
-	if !exposedNames["mcp__srv__alpha"] || !exposedNames["mcp__srv__beta"] {
-		t.Fatalf("expected both loaded deferred tools exposed, got %#v", exposed)
-	}
-	if !exposedNames["tool_search"] {
-		t.Fatalf("expected tool_search exposed on active path, got %#v", exposed)
-	}
-	// No hidden tools means no dynamic discovery text is needed.
-	if reminder != "" {
-		t.Fatalf("expected empty discovery text when nothing is hidden, got %q", reminder)
-	}
-}
-
-func TestRunLoadsDeferredToolThenAdvertisesNextTurn(t *testing.T) {
-	registry := tools.NewRegistry()
-	// load_signal asks the loop to load mcp__srv__alpha next turn.
-	registry.Register(loadSignalTool{value: "mcp__srv__alpha"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__alpha", desc: "alpha tool"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__beta", desc: "beta tool"})
-	// A real, usable tool_search must be registered for deferral to activate
-	// (otherwise the loop falls back to eager so it never strands the loader).
-	registry.Register(tools.NewToolSearchTool(registry))
-
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		{ // turn 1: call load_signal
-			{Type: zeroruntime.StreamEventToolCallStart, ToolCallID: "c1", ToolName: "load_signal"},
-			{Type: zeroruntime.StreamEventToolCallEnd, ToolCallID: "c1"},
-			{Type: zeroruntime.StreamEventDone},
-		},
-		{ // turn 2: final answer
-			{Type: zeroruntime.StreamEventText, Content: "done"},
-			{Type: zeroruntime.StreamEventDone},
-		},
-	}}
-
-	result, err := Run(context.Background(), "go", provider, Options{
-		Registry:       registry,
-		DeferThreshold: 2, // 2 deferred tools => active
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if len(provider.requests) != 2 {
-		t.Fatalf("expected two turns, got %d", len(provider.requests))
-	}
-
-	// Turn 1: alpha is hidden (not advertised), and discovery lives on the
-	// tool_search definition rather than a trailing user message.
-	turn1Tools := map[string]bool{}
-	for _, def := range provider.requests[0].Tools {
-		turn1Tools[def.Name] = true
-	}
-	if turn1Tools["mcp__srv__alpha"] {
-		t.Fatalf("turn 1 must not advertise the not-yet-loaded deferred tool")
-	}
-	assertNoDeferredDiscoveryMessage(t, provider.requests[0])
-	search1 := toolDefinitionByName(provider.requests[0].Tools, tools.ToolSearchToolName)
-	if search1 == nil {
-		t.Fatalf("turn 1 must expose tool_search, got %#v", provider.requests[0].Tools)
-	}
-	if !strings.Contains(search1.Description, "Tool discovery") || !strings.Contains(search1.Description, "mcp") {
-		t.Fatalf("tool_search description must carry deferred discovery, got %q", search1.Description)
-	}
-
-	// Turn 2: alpha is now loaded and advertised with a full schema.
-	turn2Tools := map[string]bool{}
-	for _, def := range provider.requests[1].Tools {
-		turn2Tools[def.Name] = true
-	}
-	if !turn2Tools["mcp__srv__alpha"] {
-		t.Fatalf("turn 2 must advertise the loaded deferred tool, got %#v", provider.requests[1].Tools)
-	}
-	if turn2Tools["mcp__srv__beta"] {
-		t.Fatalf("beta was never loaded; it must stay hidden in turn 2")
-	}
-
-	// The discovery text must NOT persist into the returned message history.
-	for _, m := range result.Messages {
-		if m.Role == zeroruntime.MessageRoleUser && strings.Contains(m.Content, "Tool discovery") {
-			t.Fatalf("deferred discovery must not be persisted in result.Messages")
-		}
-	}
-}
-
-// TestRunReactiveRetryKeepsLoadedDeferredToolAndDiscovery drives a mid-run
-// context-limit error that triggers reactive compaction+retry while deferral is
-// ACTIVE and a deferred tool is already loaded. The retried turn must NOT be
-// degraded to the empty-loaded/no-discovery state: it must still advertise the
-// loaded tool's FULL schema and keep tool_search discovery for hidden tools.
-func TestRunReactiveRetryKeepsLoadedDeferredToolAndDiscovery(t *testing.T) {
-	registry := tools.NewRegistry()
-	// load_signal asks the loop to load mcp__srv__alpha next turn.
-	registry.Register(loadSignalTool{value: "mcp__srv__alpha"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__alpha", desc: "alpha tool"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__beta", desc: "beta tool"})
-	// A real, usable tool_search must be registered for deferral to activate
-	// (otherwise the loop falls back to eager so it never strands the loader).
-	registry.Register(tools.NewToolSearchTool(registry))
-
-	// Request indices (mockProvider plays one turn per request, in order):
-	//   0: turn 1 — calls load_signal (loads mcp__srv__alpha for later turns)
-	//   1: turn 2 — emits a context-limit error MID-stream -> reactive recover
-	//   2: the summarize call inside Compact (must return non-empty text)
-	//   3: the RETRY request rebuilt after compaction — asserted below
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		{ // turn 1: call load_signal
-			{Type: zeroruntime.StreamEventToolCallStart, ToolCallID: "c1", ToolName: "load_signal"},
-			{Type: zeroruntime.StreamEventToolCallEnd, ToolCallID: "c1"},
-			{Type: zeroruntime.StreamEventDone},
-		},
-		{ // turn 2: mid-stream context-limit error -> reactive compaction
-			{Type: zeroruntime.StreamEventError, Error: "prompt is too long: 250000 tokens > 200000 maximum"},
-		},
-		{ // summarize call inside Compact
-			{Type: zeroruntime.StreamEventText, Content: "SUMMARY"},
-			{Type: zeroruntime.StreamEventDone},
-		},
-		{ // retry of turn 2 after compaction: final answer
-			{Type: zeroruntime.StreamEventText, Content: "done"},
-			{Type: zeroruntime.StreamEventDone},
-		},
-	}}
-
-	// A large user prompt so the elided middle is big enough that reactive
-	// compaction actually shrinks the history (recover only retries when it can
-	// shrink). ContextWindow is set high enough that PROACTIVE compaction (which
-	// fires at 0.8 * window at the top of a turn) does NOT trigger first — so the
-	// only summarize call is the reactive one, keeping the request sequence
-	// predictable: turn1, errored turn2, summarize, retry.
-	bigPrompt := strings.Repeat("work on this task. ", 2000)
-	result, err := Run(context.Background(), bigPrompt, provider, Options{
-		Registry:               registry,
-		DeferThreshold:         2, // 2 deferred tools => active
-		ContextWindow:          1_000_000,
-		CompactionPreserveLast: 2,
-	})
-	if err != nil {
-		t.Fatalf("run returned error: %v", err)
-	}
-	if result.FinalAnswer != "done" {
-		t.Fatalf("expected final answer from the retried turn, got %q", result.FinalAnswer)
-	}
-	// 4 provider calls: turn1, errored turn2, summarize, retry.
-	if len(provider.requests) != 4 {
-		t.Fatalf("expected 4 provider requests (turn1, errored turn2, summarize, retry), got %d", len(provider.requests))
-	}
-
-	retry := provider.requests[3]
-
-	// The retry must advertise the loaded deferred tool with its FULL schema —
-	// NOT re-hide it as the empty-loaded partition would.
-	var alpha *zeroruntime.ToolDefinition
-	for i := range retry.Tools {
-		if retry.Tools[i].Name == "mcp__srv__alpha" {
-			alpha = &retry.Tools[i]
-		}
-		if retry.Tools[i].Name == "mcp__srv__beta" {
-			t.Fatalf("retry must keep the never-loaded deferred tool hidden, got %#v", retry.Tools)
-		}
-	}
-	if alpha == nil {
-		t.Fatalf("retry must still advertise the loaded deferred tool mcp__srv__alpha, got %#v", retry.Tools)
-	}
-	if alpha.Parameters["type"] != "object" {
-		t.Fatalf("retry must advertise the loaded tool's FULL schema, got %#v", alpha.Parameters)
-	}
-
-	assertNoDeferredDiscoveryMessage(t, retry)
-	search := toolDefinitionByName(retry.Tools, tools.ToolSearchToolName)
-	if search == nil {
-		t.Fatalf("retry must expose tool_search for still-hidden tools, got %#v", retry.Tools)
-	}
-	if !strings.Contains(search.Description, "Tool discovery") || !strings.Contains(search.Description, "mcp") {
-		t.Fatalf("retry tool_search description must carry deferred discovery, got %q", search.Description)
-	}
-
-	// The discovery text must NOT persist into the returned message history, even on
-	// the reactive-retry path.
-	for _, m := range result.Messages {
-		if m.Role == zeroruntime.MessageRoleUser && strings.Contains(m.Content, "Tool discovery") {
-			t.Fatalf("deferred discovery must not be persisted in result.Messages")
-		}
-	}
-}
-
-// connectErrorProvider returns a connect-time error (StreamCompletion itself
-// returns a non-nil error) on the request at index errAtRequest, exercising the
-// FIRST reactive-recovery block in the loop (loop.go:99). Every other request
-// streams the corresponding turn's events. It mirrors mockProvider's recording so
-// the rebuilt retry request can be asserted.
-type connectErrorProvider struct {
-	turns        [][]zeroruntime.StreamEvent
-	errAtRequest int
-	errText      string
-	requests     []zeroruntime.CompletionRequest
-}
-
-func (provider *connectErrorProvider) StreamCompletion(_ context.Context, request zeroruntime.CompletionRequest) (<-chan zeroruntime.StreamEvent, error) {
-	provider.requests = append(provider.requests, request)
-	index := len(provider.requests) - 1
-	if index == provider.errAtRequest {
-		return nil, errors.New(provider.errText)
-	}
-	events := []zeroruntime.StreamEvent{{Type: zeroruntime.StreamEventDone}}
-	if index < len(provider.turns) {
-		events = provider.turns[index]
-	}
-	ch := make(chan zeroruntime.StreamEvent, len(events))
-	for _, event := range events {
-		ch <- event
-	}
-	close(ch)
-	return ch, nil
-}
-
-// TestRunConnectTimeReactiveRetryKeepsLoadedDeferredToolAndDiscovery mirrors
-// TestRunReactiveRetryKeepsLoadedDeferredToolAndDiscovery but drives the
-// CONNECT-TIME error path: StreamCompletion itself returns a non-nil
-// context-limit error (rather than a mid-stream StreamEventError). With deferral
-// ACTIVE and a deferred tool already loaded, the rebuilt retry request must still
-// advertise the loaded tool's FULL schema AND carry tool_search discovery —
-// i.e. the first reactive block reuses exposed, not the empty-loaded
-// partition.
-func TestRunConnectTimeReactiveRetryKeepsLoadedDeferredToolAndDiscovery(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(loadSignalTool{value: "mcp__srv__alpha"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__alpha", desc: "alpha tool"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__beta", desc: "beta tool"})
-	// A real, usable tool_search must be registered for deferral to activate
-	// (otherwise the loop falls back to eager so it never strands the loader).
-	registry.Register(tools.NewToolSearchTool(registry))
-
-	// Request indices:
-	//   0: turn 1 — calls load_signal (loads mcp__srv__alpha for later turns)
-	//   1: turn 2 — StreamCompletion returns a connect-time context-limit error
-	//   2: the summarize call inside Compact (must return non-empty text)
-	//   3: the RETRY request rebuilt after compaction — asserted below
-	provider := &connectErrorProvider{
-		errAtRequest: 1,
-		errText:      "prompt is too long: 250000 tokens > 200000 maximum",
-		turns: [][]zeroruntime.StreamEvent{
-			{ // turn 1: call load_signal
-				{Type: zeroruntime.StreamEventToolCallStart, ToolCallID: "c1", ToolName: "load_signal"},
-				{Type: zeroruntime.StreamEventToolCallEnd, ToolCallID: "c1"},
-				{Type: zeroruntime.StreamEventDone},
-			},
-			nil, // index 1: replaced by the connect-time error
-			{ // index 2: summarize call inside Compact
-				{Type: zeroruntime.StreamEventText, Content: "SUMMARY"},
-				{Type: zeroruntime.StreamEventDone},
-			},
-			{ // index 3: retry of turn 2 after compaction: final answer
-				{Type: zeroruntime.StreamEventText, Content: "done"},
-				{Type: zeroruntime.StreamEventDone},
-			},
-		},
-	}
-
-	bigPrompt := strings.Repeat("work on this task. ", 2000)
-	result, err := Run(context.Background(), bigPrompt, provider, Options{
-		Registry:               registry,
-		DeferThreshold:         2, // 2 deferred tools => active
-		ContextWindow:          1_000_000,
-		CompactionPreserveLast: 2,
-	})
-	if err != nil {
-		t.Fatalf("run returned error: %v", err)
-	}
-	if result.FinalAnswer != "done" {
-		t.Fatalf("expected final answer from the retried turn, got %q", result.FinalAnswer)
-	}
-	// 4 provider calls: turn1, errored-connect turn2, summarize, retry.
-	if len(provider.requests) != 4 {
-		t.Fatalf("expected 4 provider requests (turn1, errored turn2, summarize, retry), got %d", len(provider.requests))
-	}
-
-	retry := provider.requests[3]
-
-	// The retry must advertise the loaded deferred tool with its FULL schema and
-	// keep the never-loaded one hidden.
-	var alpha *zeroruntime.ToolDefinition
-	for i := range retry.Tools {
-		if retry.Tools[i].Name == "mcp__srv__alpha" {
-			alpha = &retry.Tools[i]
-		}
-		if retry.Tools[i].Name == "mcp__srv__beta" {
-			t.Fatalf("retry must keep the never-loaded deferred tool hidden, got %#v", retry.Tools)
-		}
-	}
-	if alpha == nil {
-		t.Fatalf("retry must still advertise the loaded deferred tool mcp__srv__alpha, got %#v", retry.Tools)
-	}
-	if alpha.Parameters["type"] != "object" {
-		t.Fatalf("retry must advertise the loaded tool's FULL schema, got %#v", alpha.Parameters)
-	}
-
-	assertNoDeferredDiscoveryMessage(t, retry)
-	search := toolDefinitionByName(retry.Tools, tools.ToolSearchToolName)
-	if search == nil {
-		t.Fatalf("retry must expose tool_search for still-hidden tools, got %#v", retry.Tools)
-	}
-	if !strings.Contains(search.Description, "Tool discovery") || !strings.Contains(search.Description, "mcp") {
-		t.Fatalf("retry tool_search description must carry deferred discovery, got %q", search.Description)
-	}
-
-	// The discovery text must NOT persist into the returned message history.
-	for _, m := range result.Messages {
-		if m.Role == zeroruntime.MessageRoleUser && strings.Contains(m.Content, "Tool discovery") {
-			t.Fatalf("deferred discovery must not be persisted in result.Messages")
-		}
-	}
-}
-
-// TestAllowlistedDeferredToolsKeepToolSearchReachable is the FIX 1+2 dead-end
-// regression: N deferred tools are registered alongside tool_search, the operator
-// allowlists ONLY the N deferred names (NOT tool_search), and DeferThreshold == N.
-// Deferral must ACTIVATE, the partition must STILL expose tool_search (the gateway
-// to the allowlisted tools), and a tool_search call must be DISPATCH-ALLOWED.
-func TestAllowlistedDeferredToolsKeepToolSearchReachable(t *testing.T) {
-	registry := tools.NewRegistry()
-	deferredNames := []string{"mcp__srv__alpha", "mcp__srv__beta"}
-	for _, name := range deferredNames {
-		registry.Register(fakeDeferredTool{name: name, desc: name + " tool"})
-	}
-	// The REAL tool_search (so partitionTools can look it up by name and dispatch
-	// can run it through registry.RunWithOptions).
-	registry.Register(tools.NewToolSearchTool(registry))
-
-	options := Options{
-		DeferThreshold: len(deferredNames), // threshold == N => active
-		// Allowlist the deferred tools but NOT tool_search.
-		EnabledTools: append([]string{}, deferredNames...),
-	}
-
-	exposed, discovery := partitionTools(registry, PermissionModeAuto, options, map[string]bool{})
-
-	// Deferral active => non-empty tool_search discovery for hidden deferred tools.
-	if discovery == "" {
-		t.Fatalf("expected deferral active (non-empty discovery) at threshold N, got empty")
-	}
-	if !strings.Contains(discovery, "tool_search") {
-		t.Fatalf("discovery must instruct the model to call tool_search, got %q", discovery)
-	}
-
-	// FIX 2(a): tool_search is exposed even though the allowlist omits it.
-	exposedNames := map[string]bool{}
-	for _, def := range exposed {
-		exposedNames[def.Name] = true
-	}
-	if !exposedNames["tool_search"] {
-		t.Fatalf("tool_search must be exposed on the active path despite the allowlist omitting it, got %#v", exposed)
-	}
-	search := toolDefinitionByName(exposed, tools.ToolSearchToolName)
-	if search == nil || search.Description != discovery {
-		t.Fatalf("tool_search description must carry discovery text, got %#v discovery=%q", search, discovery)
-	}
-	// The allowlisted-but-unloaded deferred tools are hidden (not exposed).
-	for _, name := range deferredNames {
-		if exposedNames[name] {
-			t.Fatalf("unloaded deferred tool %q must be hidden when active, got %#v", name, exposed)
-		}
-	}
-
-	// FIX 2(b): a tool_search call must be DISPATCH-ALLOWED (not rejected by the
-	// allowlist that omits it). It runs through the registry and reports a load.
-	result, abortErr := executeToolCall(
-		context.Background(),
-		registry,
-		ToolCall{ID: "c1", Name: "tool_search", Arguments: `{"query":"select:mcp__srv__alpha"}`},
-		PermissionModeAuto,
-		options,
-	)
-	if abortErr != nil {
-		t.Fatalf("unexpected abort error: %v", abortErr)
-	}
-	if result.Status != tools.StatusOK {
-		t.Fatalf("tool_search call was rejected: status=%s output=%q", result.Status, result.Output)
-	}
-	if result.DenialReason == DenialFiltered {
-		t.Fatalf("tool_search must not be denied by the allowlist, got DenialFiltered: %q", result.Output)
-	}
-	if len(result.LoadedTools) != 1 || result.LoadedTools[0] != "mcp__srv__alpha" {
-		t.Fatalf("tool_search call must load mcp__srv__alpha, got LoadedTools=%#v", result.LoadedTools)
-	}
-}
-
-// TestDisabledToolSearchStaysHiddenAndRejectedWhenActive verifies an explicit
-// DisabledTools entry for tool_search is STILL honored on the active path: the
-// loader is not exposed and a call to it is rejected (FIX 2 exempts the allowlist
-// only, never the denylist).
-func TestDisabledToolSearchFallsBackToEager(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(fakeDeferredTool{name: "mcp__srv__alpha", desc: "alpha"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__beta", desc: "beta"})
-	registry.Register(tools.NewToolSearchTool(registry))
-
-	options := Options{
-		DeferThreshold: 2,
-		DisabledTools:  []string{"tool_search"},
-	}
-
-	// tool_search is explicitly disabled, so it can never run. Deferral must NOT
-	// activate — otherwise the loop would hide the deferred tools behind a loader
-	// the dispatch gate rejects (an inescapable dead-end). Expect the eager /
-	// inactive fallback: every deferred tool exposed with its full schema, no
-	// tool_search definition, and an empty reminder.
-	exposed, reminder := partitionTools(registry, PermissionModeAuto, options, map[string]bool{})
-	if reminder != "" {
-		t.Fatalf("inactive fallback must emit no reminder, got %q", reminder)
-	}
-	exposedNames := make(map[string]bool, len(exposed))
-	for _, def := range exposed {
-		exposedNames[def.Name] = true
-	}
-	if exposedNames["tool_search"] {
-		t.Fatalf("tool_search must not be advertised when disabled, got %#v", exposed)
-	}
-	if !exposedNames["mcp__srv__alpha"] || !exposedNames["mcp__srv__beta"] {
-		t.Fatalf("deferred tools must be exposed eagerly when deferral is inactive, got %#v", exposed)
-	}
-
-	// The deferred tools themselves stay directly callable (not stranded behind a
-	// disabled loader).
-	result, abortErr := executeToolCall(
-		context.Background(),
-		registry,
-		ToolCall{ID: "c1", Name: "mcp__srv__alpha", Arguments: `{}`},
-		PermissionModeAuto,
-		options,
-	)
-	if abortErr != nil {
-		t.Fatalf("unexpected abort error: %v", abortErr)
-	}
-	if result.Status != tools.StatusOK {
-		t.Fatalf("deferred tool must be callable under eager fallback, got status=%s output=%q", result.Status, result.Output)
-	}
-}

internal/agent/denial_test.go

@@ -1,41 +0,0 @@
-package agent
-
-import (
-	"context"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/sandbox"
-	"github.com/Gitlawb/zero/internal/tools"
-)
-
-func TestExecuteToolCallCategorizesFilteredDenial(t *testing.T) {
-	root := t.TempDir()
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewReadFileTool(root))
-
-	result, _ := executeToolCall(context.Background(), registry, ToolCall{ID: "c1", Name: "read_file", Arguments: `{"path":"x"}`}, PermissionModeAuto, Options{
-		DisabledTools: []string{"read_file"},
-	})
-	if result.DenialReason != DenialFiltered {
-		t.Fatalf("DenialReason = %q, want %q", result.DenialReason, DenialFiltered)
-	}
-	if result.Status != tools.StatusError {
-		t.Fatalf("status = %q, want error", result.Status)
-	}
-}
-
-func TestDeniedPermissionResultCategories(t *testing.T) {
-	// Plain approval-declined.
-	plain := deniedPermissionResult(ToolCall{ID: "c1", Name: "bash"}, "needs approval", PermissionEvent{ToolName: "bash"})
-	if plain.DenialReason != DenialPermissionDenied {
-		t.Fatalf("plain denial = %q, want %q", plain.DenialReason, DenialPermissionDenied)
-	}
-	// Sandbox-block-driven denial.
-	sandboxDenied := deniedPermissionResult(ToolCall{ID: "c2", Name: "bash"}, "outside workspace", PermissionEvent{
-		ToolName: "bash",
-		Block:    &sandbox.Block{Code: "outside_workspace"},
-	})
-	if sandboxDenied.DenialReason != DenialSandboxBlock {
-		t.Fatalf("sandbox denial = %q, want %q", sandboxDenied.DenialReason, DenialSandboxBlock)
-	}
-}

internal/agent/effective_permission_test.go

@@ -1,44 +0,0 @@
-package agent
-
-import (
-	"context"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/tools"
-)
-
-// argsPermTool implements tools.ArgsPermissioner; plainTool does not.
-type argsPermTool struct{ perm tools.Permission }
-
-func (argsPermTool) Name() string                                        { return "args" }
-func (argsPermTool) Description() string                                 { return "" }
-func (argsPermTool) Parameters() tools.Schema                            { return tools.Schema{} }
-func (argsPermTool) Safety() tools.Safety                                { return tools.Safety{Permission: tools.PermissionPrompt} }
-func (argsPermTool) Run(context.Context, map[string]any) tools.Result    { return tools.Result{} }
-func (t argsPermTool) PermissionForArgs(map[string]any) tools.Permission { return t.perm }
-
-type plainTool struct{ perm tools.Permission }
-
-func (plainTool) Name() string                                     { return "plain" }
-func (plainTool) Description() string                              { return "" }
-func (plainTool) Parameters() tools.Schema                         { return tools.Schema{} }
-func (t plainTool) Safety() tools.Safety                           { return tools.Safety{Permission: t.perm} }
-func (plainTool) Run(context.Context, map[string]any) tools.Result { return tools.Result{} }
-
-func TestEffectivePermission(t *testing.T) {
-	// A tool implementing ArgsPermissioner refines its permission per call.
-	if got := effectivePermission(argsPermTool{perm: tools.PermissionAllow}, nil); got != tools.PermissionAllow {
-		t.Fatalf("args-aware tool: got %q, want allow", got)
-	}
-	if got := effectivePermission(argsPermTool{perm: tools.PermissionPrompt}, nil); got != tools.PermissionPrompt {
-		t.Fatalf("args-aware tool: got %q, want prompt", got)
-	}
-	// A tool WITHOUT the interface falls back to its static Safety().Permission —
-	// the args-aware path must never change a normal tool's behavior.
-	if got := effectivePermission(plainTool{perm: tools.PermissionAllow}, nil); got != tools.PermissionAllow {
-		t.Fatalf("plain tool: got %q, want its static allow", got)
-	}
-	if got := effectivePermission(plainTool{perm: tools.PermissionPrompt}, nil); got != tools.PermissionPrompt {
-		t.Fatalf("plain tool: got %q, want its static prompt", got)
-	}
-}

internal/agent/escalate_loop_test.go

@@ -1,318 +0,0 @@
-package agent
-
-import (
-	"context"
-	"errors"
-	"strings"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/tools"
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-// TestOptionsModelSwitcherFieldExists asserts the escalation hook is an
-// assignable field on Options with the agreed signature, and that nil is its
-// zero value (escalation disabled by default).
-func TestOptionsModelSwitcherFieldExists(t *testing.T) {
-	var options Options
-	if options.ModelSwitcher != nil {
-		t.Fatalf("expected ModelSwitcher to default to nil, got non-nil")
-	}
-	options.ModelSwitcher = func(_ context.Context, modelID string) (Provider, error) {
-		return &mockProvider{}, nil
-	}
-	provider, err := options.ModelSwitcher(context.Background(), "claude-sonnet-4.5")
-	if err != nil {
-		t.Fatal(err)
-	}
-	if _, ok := provider.(*mockProvider); !ok {
-		t.Fatalf("expected ModelSwitcher to return a Provider, got %T", provider)
-	}
-}
-
-// TestToolResultRequestedModelFieldExists asserts the loop-level escalation
-// signal field is present on ToolResult and empty for a normal result.
-func TestToolResultRequestedModelFieldExists(t *testing.T) {
-	var result ToolResult
-	if result.RequestedModel != "" {
-		t.Fatalf("expected RequestedModel to default to empty, got %q", result.RequestedModel)
-	}
-	result.RequestedModel = "claude-opus-4.1"
-	if result.RequestedModel != "claude-opus-4.1" {
-		t.Fatalf("expected RequestedModel to round-trip, got %q", result.RequestedModel)
-	}
-	// Keep the zeroruntime import load-bearing so the file compiles standalone.
-	_ = zeroruntime.MessageRoleAssistant
-}
-
-// escalatingTool is a registered fake tool that returns the escalation signal
-// in result Meta (mirroring the real escalate_model tool's contract), used to
-// drive the loop-level switch in tests without depending on the tools package.
-type escalatingTool struct {
-	target string
-}
-
-func (t escalatingTool) Name() string        { return "escalate" }
-func (t escalatingTool) Description() string { return "requests a model switch for testing" }
-func (t escalatingTool) Parameters() tools.Schema {
-	return tools.Schema{Type: "object", AdditionalProperties: false}
-}
-func (t escalatingTool) Safety() tools.Safety {
-	return tools.Safety{SideEffect: tools.SideEffectRead, Permission: tools.PermissionAllow}
-}
-func (t escalatingTool) Run(_ context.Context, _ map[string]any) tools.Result {
-	meta := map[string]string{}
-	if t.target != "" {
-		meta["escalate_to_model"] = t.target
-	}
-	return tools.Result{Status: tools.StatusOK, Output: "escalating", Meta: meta}
-}
-
-// TestExecuteToolCallLiftsEscalationMeta verifies executeToolCall promotes the
-// tool's Meta["escalate_to_model"] into ToolResult.RequestedModel.
-func TestExecuteToolCallLiftsEscalationMeta(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(escalatingTool{target: "claude-opus-4.1"})
-
-	result, abortErr := executeToolCall(
-		context.Background(),
-		registry,
-		ToolCall{ID: "c1", Name: "escalate"},
-		PermissionModeAuto,
-		Options{},
-	)
-	if abortErr != nil {
-		t.Fatal(abortErr)
-	}
-	if result.RequestedModel != "claude-opus-4.1" {
-		t.Fatalf("expected RequestedModel lifted from meta, got %q", result.RequestedModel)
-	}
-}
-
-// TestExecuteToolCallNoEscalationMetaLeavesRequestedModelEmpty verifies a normal
-// tool result (no escalate_to_model meta) leaves RequestedModel empty.
-func TestExecuteToolCallNoEscalationMetaLeavesRequestedModelEmpty(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(escalatingTool{target: ""})
-
-	result, abortErr := executeToolCall(
-		context.Background(),
-		registry,
-		ToolCall{ID: "c1", Name: "escalate"},
-		PermissionModeAuto,
-		Options{},
-	)
-	if abortErr != nil {
-		t.Fatal(abortErr)
-	}
-	if result.RequestedModel != "" {
-		t.Fatalf("expected empty RequestedModel without escalation meta, got %q", result.RequestedModel)
-	}
-}
-
-// escalateThenAnswerTurns builds a two-turn provider script: turn 1 calls the
-// escalate tool, turn 2 returns a final answer. Reused by the switch tests.
-func escalateThenAnswerTurns(answer string) [][]zeroruntime.StreamEvent {
-	return [][]zeroruntime.StreamEvent{
-		{
-			{Type: zeroruntime.StreamEventToolCallStart, ToolCallID: "c1", ToolName: "escalate"},
-			{Type: zeroruntime.StreamEventToolCallEnd, ToolCallID: "c1"},
-			{Type: zeroruntime.StreamEventDone},
-		},
-		{
-			{Type: zeroruntime.StreamEventText, Content: answer},
-			{Type: zeroruntime.StreamEventDone},
-		},
-	}
-}
-
-// TestRunSwitchesProviderOnEscalationRequest verifies that when a tool requests
-// a model and a ModelSwitcher is wired, the loop swaps the active provider (the
-// NEXT turn streams from the new provider) and updates options.Model.
-func TestRunSwitchesProviderOnEscalationRequest(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(escalatingTool{target: "claude-opus-4.1"})
-
-	firstProvider := &mockProvider{turns: escalateThenAnswerTurns("done")}
-	secondProvider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		{
-			{Type: zeroruntime.StreamEventText, Content: "answered on the upgraded model"},
-			{Type: zeroruntime.StreamEventDone},
-		},
-	}}
-
-	var switchedTo string
-	switchCount := 0
-	result, err := Run(context.Background(), "go", firstProvider, Options{
-		Registry: registry,
-		Model:    "claude-sonnet-4.5",
-		MaxTurns: 4,
-		ModelSwitcher: func(_ context.Context, modelID string) (Provider, error) {
-			switchCount++
-			switchedTo = modelID
-			return secondProvider, nil
-		},
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if switchCount != 1 {
-		t.Fatalf("expected exactly one switch, got %d", switchCount)
-	}
-	if switchedTo != "claude-opus-4.1" {
-		t.Fatalf("expected switch to claude-opus-4.1, got %q", switchedTo)
-	}
-	// The first provider only handled the escalation turn (1 request); the second
-	// provider handled the answer turn (1 request) — proving the swap took effect.
-	if len(firstProvider.requests) != 1 {
-		t.Fatalf("expected first provider to handle exactly the escalation turn, got %d requests", len(firstProvider.requests))
-	}
-	if len(secondProvider.requests) != 1 {
-		t.Fatalf("expected second provider to handle the post-switch turn, got %d requests", len(secondProvider.requests))
-	}
-	if result.FinalAnswer != "answered on the upgraded model" {
-		t.Fatalf("expected final answer from the swapped provider, got %q", result.FinalAnswer)
-	}
-}
-
-// TestRunIgnoresEscalationWhenSwitcherNil verifies that without a ModelSwitcher
-// the escalation request is ignored: the original provider serves every turn.
-func TestRunIgnoresEscalationWhenSwitcherNil(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(escalatingTool{target: "claude-opus-4.1"})
-
-	provider := &mockProvider{turns: escalateThenAnswerTurns("done on original")}
-
-	result, err := Run(context.Background(), "go", provider, Options{
-		Registry: registry,
-		Model:    "claude-sonnet-4.5",
-		MaxTurns: 4,
-		// ModelSwitcher intentionally nil.
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	// Both turns (escalate + answer) must run on the single original provider.
-	if len(provider.requests) != 2 {
-		t.Fatalf("expected the original provider to serve both turns, got %d requests", len(provider.requests))
-	}
-	if result.FinalAnswer != "done on original" {
-		t.Fatalf("expected final answer from the original provider, got %q", result.FinalAnswer)
-	}
-}
-
-// TestRunEscalationSwitcherErrorIsNonFatal verifies a ModelSwitcher error does
-// not abort the run: the loop continues on the current provider and a note is
-// recorded in the transcript.
-func TestRunEscalationSwitcherErrorIsNonFatal(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(escalatingTool{target: "claude-opus-4.1"})
-
-	provider := &mockProvider{turns: escalateThenAnswerTurns("recovered")}
-
-	result, err := Run(context.Background(), "go", provider, Options{
-		Registry: registry,
-		Model:    "claude-sonnet-4.5",
-		MaxTurns: 4,
-		ModelSwitcher: func(_ context.Context, _ string) (Provider, error) {
-			return nil, errors.New("provider build blew up")
-		},
-	})
-	if err != nil {
-		t.Fatalf("expected escalation error to be non-fatal, got %v", err)
-	}
-	// The run continues on the same provider through the answer turn.
-	if len(provider.requests) != 2 {
-		t.Fatalf("expected the run to continue on the original provider, got %d requests", len(provider.requests))
-	}
-	if result.FinalAnswer != "recovered" {
-		t.Fatalf("expected final answer after non-fatal switch error, got %q", result.FinalAnswer)
-	}
-	// A brief note about the failed switch must reach the model on the next turn.
-	var sawNote bool
-	for _, m := range provider.requests[1].Messages {
-		if m.Role == zeroruntime.MessageRoleUser && strings.Contains(strings.ToLower(m.Content), "could not switch") {
-			sawNote = true
-		}
-	}
-	if !sawNote {
-		t.Fatalf("expected a non-fatal switch-failure note on the next turn, messages: %+v", provider.requests[1].Messages)
-	}
-}
-
-// TestRunEscalationSwitcherNilProviderKeepsOriginal verifies that when a
-// ModelSwitcher returns (nil, nil) — no error, but also no replacement provider —
-// the loop does NOT null out the active provider. The run must continue on the
-// ORIGINAL provider and return a result rather than panic on a nil provider.
-func TestRunEscalationSwitcherNilProviderKeepsOriginal(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(escalatingTool{target: "claude-opus-4.1"})
-
-	provider := &mockProvider{turns: escalateThenAnswerTurns("answered on original")}
-
-	switchCount := 0
-	result, err := Run(context.Background(), "go", provider, Options{
-		Registry: registry,
-		Model:    "claude-sonnet-4.5",
-		MaxTurns: 4,
-		ModelSwitcher: func(_ context.Context, _ string) (Provider, error) {
-			switchCount++
-			// No error, but no provider either: the loop must NOT swap to nil.
-			return nil, nil
-		},
-	})
-	if err != nil {
-		t.Fatalf("expected (nil,nil) switch to be non-fatal, got %v", err)
-	}
-	if switchCount != 1 {
-		t.Fatalf("expected exactly one switch attempt, got %d", switchCount)
-	}
-	// Both turns (escalate + answer) must run on the single original provider; if
-	// the loop nil'd out the provider on a (nil,nil) return, the second turn would
-	// panic or never run.
-	if len(provider.requests) != 2 {
-		t.Fatalf("expected the original provider to serve both turns, got %d requests", len(provider.requests))
-	}
-	if result.FinalAnswer != "answered on original" {
-		t.Fatalf("expected final answer from the original provider, got %q", result.FinalAnswer)
-	}
-}
-
-// TestRunSwitchesAtMostOncePerTurn verifies that when a single turn carries two
-// escalation requests, only the first triggers a switch.
-func TestRunSwitchesAtMostOncePerTurn(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(escalatingTool{target: "claude-opus-4.1"})
-
-	firstProvider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		{
-			{Type: zeroruntime.StreamEventToolCallStart, ToolCallID: "c1", ToolName: "escalate"},
-			{Type: zeroruntime.StreamEventToolCallEnd, ToolCallID: "c1"},
-			{Type: zeroruntime.StreamEventToolCallStart, ToolCallID: "c2", ToolName: "escalate"},
-			{Type: zeroruntime.StreamEventToolCallEnd, ToolCallID: "c2"},
-			{Type: zeroruntime.StreamEventDone},
-		},
-	}}
-	secondProvider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		{
-			{Type: zeroruntime.StreamEventText, Content: "done"},
-			{Type: zeroruntime.StreamEventDone},
-		},
-	}}
-
-	switchCount := 0
-	if _, err := Run(context.Background(), "go", firstProvider, Options{
-		Registry: registry,
-		Model:    "claude-sonnet-4.5",
-		MaxTurns: 4,
-		ModelSwitcher: func(_ context.Context, _ string) (Provider, error) {
-			switchCount++
-			return secondProvider, nil
-		},
-	}); err != nil {
-		t.Fatal(err)
-	}
-	if switchCount != 1 {
-		t.Fatalf("expected at most one switch per turn, got %d", switchCount)
-	}
-}

internal/agent/file_diagnostics.go

@@ -1,62 +0,0 @@
-package agent
-
-import (
-	"context"
-	"os"
-	"path/filepath"
-	"strings"
-	"time"
-
-	"github.com/Gitlawb/zero/internal/lsp"
-)
-
-// fileDiagnosticsTimeout bounds one inline post-edit diagnostics check so a
-// slow or wedged language server can never hang a tool call; on timeout the
-// edit simply reports without a diagnostics block.
-const fileDiagnosticsTimeout = 10 * time.Second
-
-// NewFileDiagnostics adapts an *lsp.Manager to the per-edit inline diagnostics
-// callback (tools.RunOptions.Diagnostics): it reads the just-written file,
-// checks it against the file's language server, and formats error-severity
-// diagnostics for the model. Warnings and hints are excluded — nagging about
-// style on every edit is noise, while a type error the edit just introduced is
-// exactly what the model should see before its next step. Diagnostics are
-// rendered with workspace-relative paths: the absolute path would put the
-// local username/home directory into the model prompt and session transcript
-// on every edit. Returns nil when manager is nil, disabling inline diagnostics
-// entirely.
-func NewFileDiagnostics(manager *lsp.Manager, workspaceRoot string) func(context.Context, string) string {
-	if manager == nil {
-		return nil
-	}
-	return func(ctx context.Context, absPath string) string {
-		text, err := os.ReadFile(absPath)
-		if err != nil {
-			return ""
-		}
-		checkCtx, cancel := context.WithTimeout(ctx, fileDiagnosticsTimeout)
-		defer cancel()
-		diagnostics, err := manager.Check(checkCtx, absPath, string(text))
-		if err != nil {
-			return ""
-		}
-		errors := lsp.FilterBySeverity(diagnostics, lsp.SeverityError)
-		if len(errors) == 0 {
-			return ""
-		}
-		return lsp.FormatDiagnostics(diagnosticsDisplayPath(workspaceRoot, absPath), errors)
-	}
-}
-
-// diagnosticsDisplayPath renders absPath relative to the workspace root for
-// model-facing output, falling back to the file's base name when the path is
-// outside the workspace (a bare name still identifies the file without
-// exposing the directory layout).
-func diagnosticsDisplayPath(workspaceRoot, absPath string) string {
-	if workspaceRoot != "" {
-		if rel, err := filepath.Rel(workspaceRoot, absPath); err == nil && !strings.HasPrefix(rel, "..") {
-			return rel
-		}
-	}
-	return filepath.Base(absPath)
-}

internal/agent/file_diagnostics_test.go

@@ -1,24 +0,0 @@
-package agent
-
-import (
-	"path/filepath"
-	"testing"
-)
-
-// Diagnostics are model-facing: absolute paths would leak the local username
-// and directory layout into the prompt and session transcript on every edit.
-func TestDiagnosticsDisplayPath(t *testing.T) {
-	root := filepath.Join("/Users", "someone", "project")
-	cases := []struct {
-		root, abs, want string
-	}{
-		{root, filepath.Join(root, "internal", "a.go"), filepath.Join("internal", "a.go")},
-		{root, filepath.Join("/etc", "other.go"), "other.go"}, // outside root -> base name only
-		{"", filepath.Join("/home", "user", "x.go"), "x.go"},  // no root -> base name only
-	}
-	for _, c := range cases {
-		if got := diagnosticsDisplayPath(c.root, c.abs); got != c.want {
-			t.Errorf("diagnosticsDisplayPath(%q, %q) = %q, want %q", c.root, c.abs, got, c.want)
-		}
-	}
-}

internal/agent/guardrails.go

@@ -1,7 +1,6 @@
 package agent
 
 import (
-	"encoding/json"
 	"strconv"
 	"strings"
 
@@ -22,18 +21,6 @@ const (
 	// calls have executed since the last update_plan call.
 	staleToolCallThreshold = 10
 
-	// stalePlanTurnThreshold injects the same one-shot reminder once this many
-	// turns have passed since the last update_plan while plan items are still
-	// pending — the turn-based complement to staleToolCallThreshold, catching a
-	// plan that drifts stale across many turns that each make few tool calls.
-	stalePlanTurnThreshold = 8
-
-	// toolOnlyProgressReminderAt injects a one-shot progress nudge after this
-	// many consecutive turns contain tool calls but no visible assistant text.
-	// It does not stop the run; it tells the model to synthesize what it already
-	// knows before spending more tool turns.
-	toolOnlyProgressReminderAt = 6
-
 	// planReminderTurn is the turn (1-based) by the end of which a multi-step
 	// task should have called update_plan; if it hasn't, a one-time reminder is
 	// injected. Set to 3 (not 2) so short, legitimate two-step tasks finish
@@ -49,275 +36,10 @@ const (
 	toolFailureHintAt = 2
 	// toolFailureStopAt halts the run after a tool fails this many times in a row
 	// with the same error, so NO model (weak or strong) burns turns looping on a
-	// bad call. Set to 6 (not 4): a corrective hint fires at toolFailureHintAt (2),
-	// and a model iterating on a genuinely tricky edit can legitimately fail a few
-	// times after the hint while converging — stopping at 4 cut those runs short.
-	// The streak still resets the moment the tool succeeds or hits a different
-	// error, so this only affects true same-error loops.
-	toolFailureStopAt = 6
-
-	// maxContinueNudges bounds how many times the headless completion gate
-	// (Options.RequireCompletionSignal) re-prompts a model that stopped without a
-	// tool call while work clearly remained. Once spent, the run finalizes as
-	// INCOMPLETE rather than nudging forever (and it is still bounded by maxTurns
-	// and the run deadline).
-	maxContinueNudges = 3
+	// bad call.
+	toolFailureStopAt = 4
 )
 
-// continueNudgeMarker is a stable substring for tests.
-const continueNudgeMarker = "the task is not finished"
-
-// continueNudge tells a model that stopped without a tool call — while work
-// clearly remained — to keep going, or to mark the plan complete and summarize if
-// it is genuinely done. The second path gives it a clean route to a legitimate
-// completion (a finished plan + no continuation cue then exits as success).
-func continueNudge(reason string) string {
-	return "You stopped without calling a tool, but " + continueNudgeMarker + " (" + reason + "). " +
-		"Do not stop here: take the next concrete action with a tool now. " +
-		"If you are genuinely finished, first mark the plan complete with update_plan, then give your final summary."
-}
-
-// endsWithContinuationCue reports whether an assistant message ends mid-thought —
-// the model announcing its OWN next action and then stopping on a colon
-// ("…Let me check the SSH configuration:") rather than concluding. It requires
-// BOTH a trailing colon AND an action lead-in on the last line, so genuine closers
-// are NOT flagged: a forward-looking recommendation ("Next, I suggest reviewing
-// the changes."), a sign-off ("Let me know if you need anything"), or a summary
-// that merely ends in a colon ("Here is the summary:"). A bare trailing colon alone
-// is too common in legitimate final answers to use as a signal.
-func endsWithContinuationCue(text string) bool {
-	trimmed := strings.TrimSpace(text)
-	if trimmed == "" {
-		return false
-	}
-	lines := strings.Split(trimmed, "\n")
-	last := ""
-	for i := len(lines) - 1; i >= 0; i-- {
-		if s := strings.TrimSpace(lines[i]); s != "" {
-			last = strings.ToLower(s)
-			break
-		}
-	}
-	if !strings.HasSuffix(last, ":") {
-		return false
-	}
-	// Inspect the final clause (after the last sentence break) so a mid-line action
-	// announcement is caught ("…configure the server. Let me check the config:")
-	// while a plain summary colon ("Here is the summary:") or a recommendation is not.
-	clause := last
-	if idx := strings.LastIndex(last, ". "); idx >= 0 {
-		clause = strings.TrimSpace(last[idx+2:])
-	}
-	if strings.HasPrefix(clause, "let me know") { // sign-off, not a mid-step action
-		return false
-	}
-	for _, cue := range []string{
-		"let me ", "let's ", "now i'll ", "now i will ", "now let me ", "let me now ",
-		"i'll now ", "i will now ", "next i'll ", "next, i'll ", "first, i'll ", "first let me ",
-	} {
-		if strings.HasPrefix(clause, cue) {
-			return true
-		}
-	}
-	return false
-}
-
-// planStatusRemaining reports whether a raw update_plan status string represents
-// unfinished work. Anything not clearly completed/failed (incl. empty/unknown,
-// which the update_plan tool coerces to "pending") counts as remaining, matching
-// that tool's own status normalization.
-func planStatusRemaining(status string) bool {
-	switch strings.ToLower(strings.TrimSpace(status)) {
-	case "completed", "complete", "done", "finished", "resolved", "✓", "x", "[x]",
-		"failed", "fail", "error", "errored", "blocked", "cancelled", "canceled", "abandoned", "skipped":
-		return false
-	default:
-		return true
-	}
-}
-
-// acceptanceNudgeMarker is a stable substring for tests; it is embedded verbatim
-// in acceptanceVerificationNudge.
-const acceptanceNudgeMarker = "verify your work against the task's stated success criterion"
-
-// selfReportPhrases are high-signal admissions of guessing/fabrication or stated
-// uncertainty about the produced result. They are matched by plain substring with
-// NO context guard, so every entry must be FIRST-PERSON or unambiguously about the
-// model's own output — NOT a phrase that also describes implemented behavior.
-// (Earlier versions included "fall back to" / "placeholder value" / "as a
-// fallback" / bare "best guess" / "without proper", which match legitimate final
-// answers like "the parser will fall back to UTF-8" or "I replaced the placeholder
-// value", wrongly downgrading completed runs — so those are removed. Admissions of
-// INABILITY are handled separately by inabilityStems, which carry a guard.)
-var selfReportPhrases = []string{
-	// first-person admissions of guessing / fabrication / assumption
-	"i guessed", "my best guess", "i'm guessing", "this is a guess", "just a guess",
-	"i made it up", "i fabricated", "i assumed a", "i had to assume",
-	// first-person lack of capability (inability stems also cover cannot/could-not)
-	"i do not have the ability", "i don't have the ability", "i lack the ability",
-	"no way for me to", "not possible for me to",
-	// stated uncertainty about the correctness of the produced result
-	"may not be correct", "might not be correct", "may be incorrect", "might be incorrect",
-	"this may not work", "this might not work", "not fully functional", "not fully working",
-}
-
-// inabilityStems are first-person "I cannot / can't / could not / am unable to /
-// do not have" stems. Matching the STEM (not a fixed verb) generalizes over
-// whatever action the model claims it could not perform — "analyze", "determine",
-// "do", "complete", "verify", "see", etc. — so the detector is not defeated by
-// re-phrasing (the chess case slipped a fixed "cannot analyze" list by writing
-// "…which I cannot do without proper image analysis capabilities").
-var inabilityStems = []string{
-	"i cannot ", "i can't ", "i can not ", "i could not ", "i couldn't ",
-	"i am unable to", "i'm unable to", "i was unable to", "i wasn't able to",
-	"i was not able to", "i do not have", "i don't have", "unable to ",
-	"without being able to",
-}
-
-// successNegationTails are negated phrasings that indicate SUCCESS, not an
-// admission ("I could not find any remaining issues", "I cannot reproduce the
-// bug"). When an inability stem is immediately followed by one of these, it is not
-// treated as an admission, so a clean result is not misreported as INCOMPLETE.
-var successNegationTails = []string{
-	"find any", "found any", "find a ", "see any", "detect any", "identify any",
-	"reproduce", "spot any", "locate any",
-}
-
-// narrativeMarkers flag a sentence as RETELLING a past exchange rather than
-// reporting the outcome of the current objective, so an inability admission in
-// that sentence is about THEN, not NOW. Grounded in a real false positive: a
-// conversational recap ("You asked if I could work autonomously … I was honest
-// that my sandbox had no repo … so I couldn't actually do it at the time") was
-// downgraded to INCOMPLETE on the quoted "i couldn't " stem even though the
-// current turn's objective (summarize where we left off) was fully met. Every
-// marker is second-person or explicitly past-referential, so a first-person
-// admission about the current task ("I couldn't complete the refactor") is
-// never masked.
-var narrativeMarkers = []string{
-	"you asked", "you said", "you wanted", "you mentioned",
-	"we discussed", "we talked", "we covered",
-	"when we last", "last time", "last session", "previous session",
-	"previous conversation", "earlier session", "earlier conversation",
-	"at the time", "back then",
-}
-
-// stripQuoted removes spans enclosed in double quotes (straight or curly) or
-// backticks, so an admission the model merely QUOTES — its own earlier message,
-// a log line, an error string — cannot fire the detector. Only BALANCED spans
-// are removed: an opening delimiter that never closes is kept as literal text,
-// so a stray quote cannot swallow the rest of the message (and with it a
-// genuine admission the detector must see). Single quotes are left alone: they
-// are overwhelmingly apostrophes ("couldn't"), and treating them as quote
-// delimiters would swallow the text between two contractions.
-func stripQuoted(s string) string {
-	var b strings.Builder
-	var span strings.Builder // pending text since the open delimiter, kept if it never closes
-	open := rune(0)
-	for _, r := range s {
-		switch {
-		case open != 0:
-			if (open == '"' && r == '"') || (open == '“' && r == '”') || (open == '`' && r == '`') {
-				open = 0
-				span.Reset()
-				continue
-			}
-			span.WriteRune(r)
-		case r == '"' || r == '“' || r == '`':
-			open = r
-			span.WriteRune(r)
-		default:
-			b.WriteRune(r)
-		}
-	}
-	b.WriteString(span.String()) // dangling delimiter: restore the span verbatim
-	return b.String()
-}
-
-// admissionSentences splits lowered text into sentence-ish fragments so the
-// detector judges each claim in its own context. Newlines split too (markdown
-// bullets are separate claims); the exact boundaries only need to keep an
-// admission next to its own narrative/negation context, not be grammatical.
-func admissionSentences(lower string) []string {
-	return strings.FieldsFunc(lower, func(r rune) bool {
-		return r == '.' || r == '!' || r == '?' || r == '\n'
-	})
-}
-
-// selfReportedIncompletion returns a short reason when the model's final text
-// admits it guessed or could not meet the objective, else "". Case-insensitive.
-// Matching is per sentence, after dropping quoted spans, and a sentence that
-// retells a past exchange (narrativeMarkers) is skipped entirely — an admission
-// must be the model's own report about the CURRENT objective, not general
-// language that merely resembles one.
-func selfReportedIncompletion(text string) string {
-	for _, sentence := range admissionSentences(strings.ToLower(stripQuoted(text))) {
-		if containsAny(sentence, narrativeMarkers) {
-			continue
-		}
-		for _, phrase := range selfReportPhrases {
-			if strings.Contains(sentence, phrase) {
-				return selfReportReason(phrase)
-			}
-		}
-		for _, stem := range inabilityStems {
-			// Scan EVERY occurrence of the stem, not just the first: an earlier
-			// success-negation use ("I could not find any examples, so I could not
-			// implement it") must not mask a later genuine admission with the same stem.
-			for start := 0; ; {
-				rel := strings.Index(sentence[start:], stem)
-				if rel < 0 {
-					break
-				}
-				abs := start + rel
-				tail := strings.TrimSpace(sentence[abs+len(stem):])
-				if !hasAnyPrefix(tail, successNegationTails) {
-					return selfReportReason(strings.TrimSpace(stem) + " …")
-				}
-				start = abs + len(stem)
-			}
-		}
-	}
-	return ""
-}
-
-func containsAny(s string, subs []string) bool {
-	for _, sub := range subs {
-		if strings.Contains(s, sub) {
-			return true
-		}
-	}
-	return false
-}
-
-func selfReportReason(marker string) string {
-	return `the final message admits the objective was not met ("` + marker + `")`
-}
-
-func hasAnyPrefix(s string, prefixes []string) bool {
-	for _, p := range prefixes {
-		if strings.HasPrefix(s, p) {
-			return true
-		}
-	}
-	return false
-}
-
-// acceptanceVerificationNudge forces a TASK-GROUNDED acceptance check before a run
-// may finalize as success. It explicitly rejects the three false-success patterns
-// the bug-hunt surfaced: well-formed output treated as correct; pre-existing tests
-// passing treated as the objective being met; and a result that merely matches a
-// baseline the task asked to beat or improve. General — no task-specific content.
-func acceptanceVerificationNudge() string {
-	return "Before this task can be marked complete, " + acceptanceNudgeMarker + " — " +
-		"NOT the shape or format of your output, NOT that pre-existing tests pass, and NOT that your " +
-		"result merely matches a baseline you were asked to beat or improve. " +
-		"Re-read the original task, then run a concrete check that exercises the actual requirement: " +
-		"execute the program or tests that demonstrate the required behavior, or directly probe the specific " +
-		"thing the task asked you to produce, recover, fix, or optimize. " +
-		"If that check passes, reply PASS and cite the evidence. " +
-		"If it does not pass — or you cannot run such a check — say so plainly and keep working; do not claim success."
-}
-
 // toolFailureHintMarker is a stable substring for tests.
 const toolFailureHintMarker = "kept failing with the same error"
 
@@ -362,56 +84,18 @@ func toolFailureStopAnswer(toolName string, count int) string {
 		"Please check the request or adjust the tool arguments."
 }
 
-// The no-output stop answer is assembled from these fixed parts (only the turn
-// count varies). IsNoProgressStop matches all three so a legitimate message that
-// merely quotes the marker substring is not misclassified as a failed empty run.
-const (
-	noOutputStopPrefix = "Agent stopped after "
-	noOutputStopMarker = "with no output (no visible text and no tool calls)"
-	noOutputStopSuffix = "to avoid consuming tokens without making progress."
-)
-
 // noOutputStopAnswer is the final answer returned when the no-output guard
 // stops the run. The turn count is interpolated at the call site.
 func noOutputStopAnswer(turns int) string {
-	return noOutputStopPrefix + strconv.Itoa(turns) + " turns " + noOutputStopMarker + " " + noOutputStopSuffix
-}
-
-// IsNoProgressStop reports whether content IS the no-output guardrail stop answer
-// (a run that produced no visible text and no tool calls). It matches the EXACT
-// structure noOutputStopAnswer emits — prefix + "<int> turns " + marker + " " +
-// suffix, where only the integer turn count varies — rather than just looking for
-// the three parts in order. A loose check (prefix && contains-marker && suffix)
-// would misclassify a genuine assistant/tool message that merely quotes the
-// marker amid other prose, which would wrongly hide a real session from /resume
-// and skip its title generation.
-func IsNoProgressStop(content string) bool {
-	trimmed := strings.TrimSpace(content)
-	if !strings.HasPrefix(trimmed, noOutputStopPrefix) {
-		return false
-	}
-	rest := trimmed[len(noOutputStopPrefix):]
-	const turnsSep = " turns "
-	sep := strings.Index(rest, turnsSep)
-	if sep < 0 {
-		return false
-	}
-	// The text between the prefix and " turns " must be exactly the bare integer
-	// count; anything else means this isn't the guard's own answer.
-	if _, err := strconv.Atoi(rest[:sep]); err != nil {
-		return false
-	}
-	// The marker must be immediately followed (one space) by the suffix and then
-	// end — no arbitrary text wedged in between.
-	return rest[sep+len(turnsSep):] == noOutputStopMarker+" "+noOutputStopSuffix
+	return "Agent stopped after " + strconv.Itoa(turns) +
+		" turns with no output (no visible text and no tool calls) to avoid consuming tokens without making progress."
 }
 
 // Reminder markers are stable substrings used both to build the reminder text
 // and to assert in tests that the right reminder was injected exactly once.
 const (
-	planNotCalledReminderMarker    = "you have not called update_plan"
-	planStaleReminderMarker        = "haven't updated the plan via update_plan"
-	toolOnlyProgressReminderMarker = "consecutive tool-only turns"
+	planNotCalledReminderMarker = "you have not called update_plan"
+	planStaleReminderMarker     = "haven't updated the plan via update_plan"
 )
 
 // planNotCalledReminder nudges the model to track a multi-step task with
@@ -430,34 +114,18 @@ func planStaleReminder(callsSinceUpdate int) string {
 		" in a while. Update the plan to reflect completed and remaining steps, then continue."
 }
 
-func toolOnlyProgressReminder(turns int) string {
-	return "Reminder: you've made " + strconv.Itoa(turns) + " " + toolOnlyProgressReminderMarker +
-		" without visible progress. Before calling more tools, summarize what you already know, state the next concrete step, and finish if you have enough information."
-}
-
 // guardState tracks the per-run signals the guardrails need. It is observable
 // purely from tool-call names and per-turn output, matching what the loop holds.
 type guardState struct {
 	emptyTurns               int
 	totalToolCalls           int
 	toolCallsSincePlanUpdate int
-	// turnsSincePlanUpdate counts turns (not individual tool calls) since the last
-	// update_plan, so a plan that goes stale across many low-tool-call turns is
-	// still caught — the tool-call counter alone can take many turns to trip when
-	// the model makes only one call per turn.
-	turnsSincePlanUpdate  int
-	planEverCalled        bool
-	notCalledReminderSent bool
+	planEverCalled           bool
+	notCalledReminderSent    bool
 	// staleReminderSent records whether the stale reminder has already fired for
 	// the current stale interval. It is cleared when a plan update opens a new
 	// interval, making the reminder one-shot per interval rather than per turn.
-	staleReminderSent    bool
-	toolOnlyTurns        int
-	toolOnlyReminderSent bool
-	// planItemsPending is the number of remaining (pending/in_progress) items in
-	// the most recent update_plan call, so the headless completion gate can tell
-	// whether work is unfinished when the model stops without a tool call.
-	planItemsPending int
+	staleReminderSent bool
 	// toolFailures tracks consecutive same-error failures per tool, keyed by tool
 	// name, so the loop can hint then halt instead of looping forever.
 	toolFailures map[string]*toolFailureRecord
@@ -506,45 +174,20 @@ func (state *guardState) observeToolResult(name string, failed bool, output stri
 func (state *guardState) observeTurn(collected zeroruntime.CollectedStream) (stop bool) {
 	hasToolCalls := len(collected.ToolCalls) > 0
 	hasVisibleText := strings.TrimSpace(collected.Text) != ""
-	hasReasoning := collected.HasReasoning || len(collected.ReasoningBlocks) > 0
 
-	productive := hasToolCalls || hasVisibleText
-	switch {
-	case productive:
+	if hasToolCalls || hasVisibleText {
 		state.emptyTurns = 0
-	case hasReasoning && state.planItemsPending > 0:
-		// Reasoning alone does not prove progress while plan steps remain — weaker
-		// models (e.g. HY3 via OpenAI-compatible gateways) can stream long
-		// reasoning-only turns without tools or visible text, burning the turn
-		// budget while the plan panel stays on the first in_progress step.
-		state.emptyTurns++
-	case hasReasoning:
-		state.emptyTurns = 0
-	default:
-		state.emptyTurns++
-	}
-	if hasToolCalls && !hasVisibleText {
-		state.toolOnlyTurns++
 	} else {
-		state.toolOnlyTurns = 0
-		state.toolOnlyReminderSent = false
+		state.emptyTurns++
 	}
 
-	// One turn has passed; the plan-update below resets this to 0 when the model
-	// refreshes the plan this turn.
-	state.turnsSincePlanUpdate++
-
 	for _, call := range collected.ToolCalls {
 		state.totalToolCalls++
 		if call.Name == planToolName {
 			state.planEverCalled = true
 			state.toolCallsSincePlanUpdate = 0
-			state.turnsSincePlanUpdate = 0
 			// A fresh plan update opens a new stale interval.
 			state.staleReminderSent = false
-			// Record how many items remain so the completion gate knows whether
-			// work is unfinished if the model later stops without a tool call.
-			state.observePlanUpdate(call.Arguments)
 		} else {
 			state.toolCallsSincePlanUpdate++
 		}
@@ -553,52 +196,15 @@ func (state *guardState) observeTurn(collected zeroruntime.CollectedStream) (sto
 	return state.emptyTurns >= maxEmptyTurns
 }
 
-// pendingPlanItems reports whether the most recent update_plan call still has
-// unfinished (pending/in_progress) items. False when no plan was ever recorded.
-func (state *guardState) pendingPlanItems() bool {
-	return state.planItemsPending > 0
-}
-
-// observePlanUpdate parses an update_plan call's raw arguments and records how
-// many items are still remaining. Malformed arguments leave the prior count
-// unchanged (best-effort — the plan panel itself tolerates the same).
-func (state *guardState) observePlanUpdate(arguments string) {
-	var parsed struct {
-		Plan []struct {
-			Status string `json:"status"`
-		} `json:"plan"`
-	}
-	if json.Unmarshal([]byte(arguments), &parsed) != nil {
-		return
-	}
-	pending := 0
-	for _, item := range parsed.Plan {
-		if planStatusRemaining(item.Status) {
-			pending++
-		}
-	}
-	state.planItemsPending = pending
-}
-
-func (state *guardState) progressReminder() string {
-	if state.toolOnlyReminderSent || state.toolOnlyTurns < toolOnlyProgressReminderAt {
-		return ""
-	}
-	state.toolOnlyReminderSent = true
-	return toolOnlyProgressReminder(state.toolOnlyTurns)
-}
-
 // planReminder returns a one-shot reminder message to inject before the next
 // turn, or an empty string when no reminder applies. `turn` is 1-based (the
 // number of turns completed so far).
 func (state *guardState) planReminder(turn int) string {
 	// STALE reminder takes priority: a long run without a plan update is the
-	// stronger signal. Fires on either the tool-call streak OR a turn streak with
-	// pending items (so a plan drifting stale across many low-call turns is caught,
-	// while a fully-completed plan is left alone). One-shot per stale interval.
-	if state.planEverCalled && !state.staleReminderSent &&
-		(state.toolCallsSincePlanUpdate >= staleToolCallThreshold ||
-			(state.turnsSincePlanUpdate >= stalePlanTurnThreshold && state.planItemsPending > 0)) {
+	// stronger signal. One-shot per stale interval.
+	if state.planEverCalled &&
+		!state.staleReminderSent &&
+		state.toolCallsSincePlanUpdate >= staleToolCallThreshold {
 		state.staleReminderSent = true
 		return planStaleReminder(state.toolCallsSincePlanUpdate)
 	}

internal/agent/guardrails_test.go

@@ -22,24 +22,6 @@ func textTurn(content string) []zeroruntime.StreamEvent {
 	}
 }
 
-// reasoningTurn produces live reasoning without visible assistant text.
-func reasoningTurn(content string) []zeroruntime.StreamEvent {
-	return []zeroruntime.StreamEvent{
-		{Type: zeroruntime.StreamEventReasoning, Content: content},
-		{Type: zeroruntime.StreamEventDone},
-	}
-}
-
-// multiReasoningTurn streams several reasoning chunks inside ONE provider turn.
-func multiReasoningTurn(chunks ...string) []zeroruntime.StreamEvent {
-	events := make([]zeroruntime.StreamEvent, 0, len(chunks)+1)
-	for _, chunk := range chunks {
-		events = append(events, zeroruntime.StreamEvent{Type: zeroruntime.StreamEventReasoning, Content: chunk})
-	}
-	events = append(events, zeroruntime.StreamEvent{Type: zeroruntime.StreamEventDone})
-	return events
-}
-
 // toolTurn produces a turn that calls a named tool with the given args JSON.
 func toolTurn(callID string, toolName string, args string) []zeroruntime.StreamEvent {
 	return []zeroruntime.StreamEvent{
@@ -119,120 +101,6 @@ func TestRunResetsEmptyTurnCounterOnVisibleOutput(t *testing.T) {
 	}
 }
 
-func TestRunStopsReasoningOnlyTurnsWhilePlanPending(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewUpdatePlanTool())
-
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		toolTurn("plan", "update_plan", `{"plan":[{"content":"step 1","status":"in_progress"},{"content":"step 2","status":"pending"}]}`),
-		reasoningTurn("thinking 1"),
-		reasoningTurn("thinking 2"),
-		reasoningTurn("thinking 3"),
-		textTurn("should never reach here"),
-	}}
-
-	result, err := Run(context.Background(), "multi-step task", provider, Options{
-		Registry: registry,
-		MaxTurns: 12,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if len(provider.requests) != maxEmptyTurns+1 {
-		t.Fatalf("expected %d turns (1 plan + %d reasoning-only stalls), got %d", maxEmptyTurns+1, maxEmptyTurns, len(provider.requests))
-	}
-	if !result.Incomplete {
-		t.Fatalf("expected Incomplete=true after reasoning-only stall with pending plan, got false")
-	}
-	if !strings.Contains(result.FinalAnswer, noOutputStopMarker) {
-		t.Fatalf("expected no-output stop answer, got %q", result.FinalAnswer)
-	}
-}
-
-func TestRunMultiChunkReasoningCountsAsOneNonProductiveTurnWhilePlanPending(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewUpdatePlanTool())
-
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		toolTurn("plan", "update_plan", `{"plan":[{"content":"step 1","status":"in_progress"},{"content":"step 2","status":"pending"}]}`),
-		multiReasoningTurn("chunk-1 ", "chunk-2 ", "chunk-3 ", "chunk-4 ", "chunk-5 "),
-		multiReasoningTurn("more-1 ", "more-2 ", "more-3 "),
-		multiReasoningTurn("final-1 ", "final-2 "),
-		textTurn("should never reach here"),
-	}}
-
-	result, err := Run(context.Background(), "multi-step task", provider, Options{
-		Registry: registry,
-		MaxTurns: 12,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if len(provider.requests) != maxEmptyTurns+1 {
-		t.Fatalf("expected %d turns (1 plan + %d multi-chunk reasoning stalls), got %d",
-			maxEmptyTurns+1, maxEmptyTurns, len(provider.requests))
-	}
-	if !result.Incomplete {
-		t.Fatalf("expected Incomplete=true, got false")
-	}
-}
-
-func TestRunReasoningThenToolCallSucceedsWithPendingPlan(t *testing.T) {
-	root := t.TempDir()
-	writeAgentTestFile(t, root+"/notes.txt", "alpha")
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewReadFileTool(root))
-	registry.Register(tools.NewUpdatePlanTool())
-
-	provider := &mockProvider{turns: [][]zeroruntime.StreamEvent{
-		toolTurn("plan", "update_plan", `{"plan":[{"content":"step 1","status":"in_progress"},{"content":"step 2","status":"pending"}]}`),
-		multiReasoningTurn("thinking ", "more thinking "),
-		toolTurn("read", "read_file", `{"path":"notes.txt"}`),
-		toolTurn("plan2", "update_plan", `{"plan":[{"content":"step 1","status":"completed"},{"content":"step 2","status":"completed"}]}`),
-		textTurn("Both steps are done."),
-	}}
-
-	result, err := Run(context.Background(), "multi-step task", provider, Options{
-		Registry: registry,
-		MaxTurns: 12,
-		Cwd:      root,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.Incomplete {
-		t.Fatalf("run should succeed after productive tool call, got Incomplete (%q)", result.IncompleteReason)
-	}
-	if result.FinalAnswer != "Both steps are done." {
-		t.Fatalf("final answer = %q", result.FinalAnswer)
-	}
-}
-
-func TestRunResetsEmptyTurnCounterOnReasoning(t *testing.T) {
-	provider := &mockProvider{
-		turns: [][]zeroruntime.StreamEvent{
-			reasoningTurn("thinking 1"),
-			reasoningTurn("thinking 2"),
-			reasoningTurn("thinking 3"),
-			textTurn("done"),
-		},
-	}
-
-	result, err := Run(context.Background(), "go", provider, Options{
-		Registry: tools.NewRegistry(),
-		MaxTurns: 12,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.FinalAnswer != "done" {
-		t.Fatalf("expected reasoning-only turns to keep the run live until final answer, got %q", result.FinalAnswer)
-	}
-	if len(provider.requests) != 4 {
-		t.Fatalf("expected 4 turns, got %d", len(provider.requests))
-	}
-}
-
 func TestRunResetsEmptyTurnCounterOnToolCall(t *testing.T) {
 	root := t.TempDir()
 	writeAgentTestFile(t, root+"/notes.txt", "alpha")
@@ -268,26 +136,6 @@ func TestRunResetsEmptyTurnCounterOnToolCall(t *testing.T) {
 	}
 }
 
-func TestGuardStateResetsToolOnlyStreakOnEmptyNonToolTurn(t *testing.T) {
-	var state guardState
-	toolOnly := zeroruntime.CollectedStream{
-		ToolCalls: []zeroruntime.ToolCall{{ID: "call", Name: "read_file", Arguments: `{}`}},
-	}
-
-	for range toolOnlyProgressReminderAt - 1 {
-		state.observeTurn(toolOnly)
-	}
-	state.observeTurn(zeroruntime.CollectedStream{})
-	state.observeTurn(toolOnly)
-
-	if reminder := state.progressReminder(); reminder != "" {
-		t.Fatalf("expected empty non-tool turn to reset tool-only progress reminder, got %q", reminder)
-	}
-	if state.toolOnlyTurns != 1 {
-		t.Fatalf("expected tool-only streak to restart at 1, got %d", state.toolOnlyTurns)
-	}
-}
-
 func TestRunDoesNotCountDroppedToolCallTurnsAsEmpty(t *testing.T) {
 	provider := &mockProvider{
 		turns: [][]zeroruntime.StreamEvent{
@@ -364,7 +212,7 @@ func TestRunDoesNotInjectPlanReminderForTrivialTask(t *testing.T) {
 	provider := &mockProvider{
 		turns: [][]zeroruntime.StreamEvent{
 			toolTurn("call-1", "read_file", `{"path":"notes.txt"}`), // single tool call
-			textTurn("done"), // immediately answers
+			textTurn("done"),                                        // immediately answers
 		},
 	}
 
@@ -392,7 +240,7 @@ func TestRunDoesNotInjectNotCalledReminderWhenPlanUsed(t *testing.T) {
 
 	provider := &mockProvider{
 		turns: [][]zeroruntime.StreamEvent{
-			toolTurn("call-1", "update_plan", `{"plan":[{"content":"step one","status":"completed"}]}`),
+			toolTurn("call-1", "update_plan", `{"plan":[{"content":"step one"}]}`),
 			toolTurn("call-2", "read_file", `{"path":"notes.txt"}`),
 			textTurn("done"),
 		},
@@ -428,12 +276,6 @@ func TestRunInjectsStalePlanReminderAfterManyToolCalls(t *testing.T) {
 	for i := 0; i < staleToolCallThreshold+2; i++ {
 		turns = append(turns, toolTurn("call", "read_file", `{"path":"notes.txt"}`))
 	}
-	// The plan-aware completion gate re-prompts up to maxContinueNudges when the
-	// model stops with text while plan items remain; supply those turns before
-	// the final answer.
-	for i := 0; i < maxContinueNudges; i++ {
-		turns = append(turns, textTurn("done."))
-	}
 	turns = append(turns, textTurn("done"))
 
 	provider := &mockProvider{turns: turns}
@@ -468,9 +310,6 @@ func TestRunStalePlanReminderIsOneShotPerInterval(t *testing.T) {
 	for i := 0; i < staleToolCallThreshold*2; i++ {
 		turns = append(turns, toolTurn("call", "read_file", `{"path":"notes.txt"}`))
 	}
-	for i := 0; i < maxContinueNudges; i++ {
-		turns = append(turns, textTurn("done."))
-	}
 	turns = append(turns, textTurn("done"))
 
 	provider := &mockProvider{turns: turns}
@@ -491,50 +330,11 @@ func TestRunStalePlanReminderIsOneShotPerInterval(t *testing.T) {
 	}
 }
 
-func TestRunInjectsToolOnlyProgressReminder(t *testing.T) {
-	root := t.TempDir()
-	writeAgentTestFile(t, root+"/notes.txt", "alpha")
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewReadFileTool(root))
-
-	turns := make([][]zeroruntime.StreamEvent, 0, toolOnlyProgressReminderAt+1)
-	for i := 0; i < toolOnlyProgressReminderAt; i++ {
-		turns = append(turns, toolTurn("call", "read_file", `{"path":"notes.txt"}`))
-	}
-	turns = append(turns, textTurn("done"))
-
-	provider := &mockProvider{turns: turns}
-	result, err := Run(context.Background(), "go", provider, Options{
-		Registry: registry,
-		MaxTurns: len(turns) + 2,
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if result.FinalAnswer != "done" {
-		t.Fatalf("expected final answer, got %q", result.FinalAnswer)
-	}
-	if count := countUserMessagesContaining(result.Messages, toolOnlyProgressReminderMarker); count != 1 {
-		t.Fatalf("expected one tool-only progress reminder, got %d", count)
-	}
-	found := false
-	for _, message := range provider.requests[toolOnlyProgressReminderAt].Messages {
-		if message.Role == zeroruntime.MessageRoleUser && strings.Contains(message.Content, toolOnlyProgressReminderMarker) {
-			found = true
-		}
-	}
-	if !found {
-		t.Fatalf("expected reminder on request after tool-only streak, messages: %+v", provider.requests[toolOnlyProgressReminderAt].Messages)
-	}
-}
-
 type alwaysFailingTool struct{}
 
-func (alwaysFailingTool) Name() string        { return "flaky" }
-func (alwaysFailingTool) Description() string { return "always fails for testing" }
-func (alwaysFailingTool) Parameters() tools.Schema {
-	return tools.Schema{Type: "object", AdditionalProperties: false}
-}
+func (alwaysFailingTool) Name() string             { return "flaky" }
+func (alwaysFailingTool) Description() string       { return "always fails for testing" }
+func (alwaysFailingTool) Parameters() tools.Schema  { return tools.Schema{Type: "object", AdditionalProperties: false} }
 func (alwaysFailingTool) Safety() tools.Safety {
 	return tools.Safety{SideEffect: tools.SideEffectRead, Permission: tools.PermissionAllow}
 }

internal/agent/hook_wiring_test.go

@@ -1,95 +0,0 @@
-package agent
-
-import (
-	"context"
-	"strings"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/hooks"
-	"github.com/Gitlawb/zero/internal/tools"
-)
-
-func TestAppendHookFeedbackFormatsOutput(t *testing.T) {
-	// Blank feedback leaves the original output untouched and reports no redaction.
-	if got, redacted := appendHookFeedback("tool output", "   "); got != "tool output" || redacted {
-		t.Fatalf("blank feedback should not change output or redact, got %q redacted=%v", got, redacted)
-	}
-	// Feedback is appended under a header alongside the existing output.
-	got, redacted := appendHookFeedback("tool output", "gofmt reformatted main.go")
-	if !strings.Contains(got, "tool output") || !strings.Contains(got, "Hook output:") || !strings.Contains(got, "gofmt reformatted main.go") {
-		t.Fatalf("expected combined tool + hook output, got %q", got)
-	}
-	if redacted {
-		t.Fatal("clean feedback should not be reported as redacted")
-	}
-	// With no original output the hook feedback stands alone under the header.
-	if got, _ := appendHookFeedback("", "validation ran"); !strings.HasPrefix(got, "Hook output:") || !strings.Contains(got, "validation ran") {
-		t.Fatalf("expected standalone hook output, got %q", got)
-	}
-}
-
-func TestAppendHookFeedbackScrubsSecretsAndFlagsRedaction(t *testing.T) {
-	// A hook that echoes a secret must be scrubbed before it reaches the model,
-	// and the redaction must be reported so the caller can flag ToolResult.Redacted.
-	got, redacted := appendHookFeedback("tool output", "leaked token ghp_abcdefghijklmnopqrstuvwxyz0123456789")
-	if !redacted {
-		t.Fatal("expected redaction to be reported for secret-bearing feedback")
-	}
-	if strings.Contains(got, "ghp_abcdefghijklmnopqrstuvwxyz0123456789") {
-		t.Fatalf("secret leaked into hook feedback: %q", got)
-	}
-}
-
-func TestBlockedByHookResultCarriesReasonAndDenial(t *testing.T) {
-	out := blockedByHookResult(
-		ToolCall{ID: "c1", Name: "write_file"},
-		hooks.DispatchOutcome{Blocked: true, BlockedBy: "policy", Reason: "writes under /etc are denied"},
-	)
-	if out.Status != tools.StatusError {
-		t.Fatalf("status = %v, want error", out.Status)
-	}
-	if out.DenialReason != DenialHookBlocked {
-		t.Fatalf("denial = %q, want %q", out.DenialReason, DenialHookBlocked)
-	}
-	if out.ToolCallID != "c1" || out.Name != "write_file" {
-		t.Fatalf("call identity not propagated: %#v", out)
-	}
-	for _, want := range []string{"write_file", "policy", "writes under /etc are denied"} {
-		if !strings.Contains(out.Output, want) {
-			t.Fatalf("output %q missing %q", out.Output, want)
-		}
-	}
-	if out.Redacted {
-		t.Fatal("a clean hook reason must not flag Redacted")
-	}
-}
-
-func TestBlockedByHookResultScrubsSecretReason(t *testing.T) {
-	out := blockedByHookResult(
-		ToolCall{ID: "c3", Name: "bash"},
-		hooks.DispatchOutcome{Blocked: true, BlockedBy: "secret-scan", Reason: "denied: found ghp_abcdefghijklmnopqrstuvwxyz0123456789 in args"},
-	)
-	if !out.Redacted {
-		t.Fatal("a scrubbed hook reason must set ToolResult.Redacted")
-	}
-	if strings.Contains(out.Output, "ghp_abcdefghijklmnopqrstuvwxyz0123456789") {
-		t.Fatalf("secret leaked into blocked-hook output: %q", out.Output)
-	}
-}
-
-func TestBlockedByHookResultFallsBackWhenReasonEmpty(t *testing.T) {
-	out := blockedByHookResult(ToolCall{ID: "c2", Name: "bash"}, hooks.DispatchOutcome{Blocked: true, BlockedBy: "x"})
-	if !strings.Contains(out.Output, "blocked by a beforeTool hook") {
-		t.Fatalf("expected a default reason, got %q", out.Output)
-	}
-}
-
-func TestDispatchHelpersAreNoopWithoutDispatcher(t *testing.T) {
-	options := Options{} // Hooks is nil
-	if _, blocked := dispatchBeforeTool(context.Background(), options, ToolCall{Name: "bash"}, nil); blocked {
-		t.Fatal("a nil dispatcher must never block a tool")
-	}
-	if feedback := dispatchAfterTool(context.Background(), options, ToolCall{Name: "bash"}, nil, tools.Result{}); feedback != "" {
-		t.Fatalf("a nil dispatcher must yield no feedback, got %q", feedback)
-	}
-}

internal/agent/image_rejection_test.go

@@ -1,62 +0,0 @@
-package agent
-
-import (
-	"errors"
-	"testing"
-)
-
-func TestIsImageRejectionError(t *testing.T) {
-	tests := []struct {
-		name string
-		err  error
-		want bool
-	}{
-		{
-			name: "400 with image mention",
-			err:  errors.New("HTTP 400: image content is not supported by this model"),
-			want: true,
-		},
-		{
-			name: "400 with multimodal mention",
-			err:  errors.New("400 Bad Request: model does not support multimodal input"),
-			want: true,
-		},
-		{
-			name: "400 with vision mention",
-			err:  errors.New("error 400: this model does not have vision capabilities"),
-			want: true,
-		},
-		{
-			name: "400 with unsupported content type",
-			err:  errors.New("400: unsupported content type"),
-			want: true,
-		},
-		{
-			name: "generic 400 no image keywords",
-			err:  errors.New("400 Bad Request: invalid model id"),
-			want: false,
-		},
-		{
-			name: "500 server error",
-			err:  errors.New("500 Internal Server Error"),
-			want: false,
-		},
-		{
-			name: "nil error",
-			err:  nil,
-			want: false,
-		},
-		{
-			name: "context limit error",
-			err:  errors.New("context length exceeded"),
-			want: false,
-		},
-	}
-	for _, tt := range tests {
-		t.Run(tt.name, func(t *testing.T) {
-			if got := isImageRejectionError(tt.err); got != tt.want {
-				t.Errorf("isImageRejectionError(%v) = %v, want %v", tt.err, got, tt.want)
-			}
-		})
-	}
-}

internal/agent/loop_images_test.go

@@ -1,87 +0,0 @@
-package agent
-
-import (
-	"bytes"
-	"context"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-// imageEchoProvider records the messages of the first request it receives, then
-// returns an empty final answer so the loop terminates after one turn.
-type imageEchoProvider struct {
-	seen []zeroruntime.Message
-}
-
-func (p *imageEchoProvider) StreamCompletion(ctx context.Context, request zeroruntime.CompletionRequest) (<-chan zeroruntime.StreamEvent, error) {
-	if p.seen == nil {
-		p.seen = append([]zeroruntime.Message{}, request.Messages...)
-	}
-	events := make(chan zeroruntime.StreamEvent)
-	close(events)
-	return events, nil
-}
-
-func TestRunSeedsImagesIntoUserTurn(t *testing.T) {
-	provider := &imageEchoProvider{}
-	images := []zeroruntime.ImageBlock{{MediaType: "image/png", Data: []byte{0x89, 0x50}}}
-
-	if _, err := Run(context.Background(), "look at this", provider, Options{
-		MaxTurns: 1,
-		Images:   images,
-	}); err != nil {
-		t.Fatalf("Run returned error: %v", err)
-	}
-
-	if len(provider.seen) < 2 {
-		t.Fatalf("provider saw %d messages, want >= 2", len(provider.seen))
-	}
-	user := provider.seen[len(provider.seen)-1]
-	if user.Role != zeroruntime.MessageRoleUser {
-		t.Fatalf("last seeded message role = %q, want user", user.Role)
-	}
-	if len(user.Images) != 1 || user.Images[0].MediaType != "image/png" {
-		t.Fatalf("user.Images = %#v, want one image/png block", user.Images)
-	}
-}
-
-// TestCopyMessagesDeepCopiesImageBytes locks the anti-aliasing guarantee for
-// copyMessages: copies must carry INDEPENDENT image bytes, so mutating the
-// source message's Data never bleeds into a history/request/result copy.
-func TestCopyMessagesDeepCopiesImageBytes(t *testing.T) {
-	source := []Message{
-		{
-			Role:    zeroruntime.MessageRoleUser,
-			Content: "look",
-			Images: []zeroruntime.ImageBlock{
-				{MediaType: "image/png", Data: []byte{0x89, 0x50, 0x4e, 0x47}},
-			},
-		},
-	}
-
-	copied := copyMessages(source)
-	if len(copied) != 1 || len(copied[0].Images) != 1 {
-		t.Fatalf("unexpected copy shape: %#v", copied)
-	}
-
-	// Mutating the source bytes must not change the copy.
-	source[0].Images[0].Data[0] = 0x00
-	if !bytes.Equal(copied[0].Images[0].Data, []byte{0x89, 0x50, 0x4e, 0x47}) {
-		t.Fatalf("copy image bytes aliased the source: %v", copied[0].Images[0].Data)
-	}
-	if &source[0].Images[0].Data[0] == &copied[0].Images[0].Data[0] {
-		t.Fatal("copy Data shares backing array with source")
-	}
-}
-
-func TestRunWithoutImagesSeedsNilImages(t *testing.T) {
-	provider := &imageEchoProvider{}
-	if _, err := Run(context.Background(), "hello", provider, Options{MaxTurns: 1}); err != nil {
-		t.Fatalf("Run returned error: %v", err)
-	}
-	user := provider.seen[len(provider.seen)-1]
-	if user.Images != nil {
-		t.Fatalf("user.Images = %#v, want nil for text-only run", user.Images)
-	}
-}

internal/agent/member_auto_test.go

@@ -1,51 +0,0 @@
-package agent
-
-import (
-	"context"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/tools"
-)
-
-type safetyTool struct {
-	name   string
-	safety tools.Safety
-}
-
-func (t safetyTool) Name() string        { return t.name }
-func (t safetyTool) Description() string { return "test tool" }
-func (t safetyTool) Parameters() tools.Schema {
-	return tools.Schema{Type: "object", AdditionalProperties: false}
-}
-func (t safetyTool) Safety() tools.Safety { return t.safety }
-func (t safetyTool) Run(context.Context, map[string]any) tools.Result {
-	return tools.Result{Status: tools.StatusOK}
-}
-
-// member-auto advertises the in-workspace mutators a headless member needs to
-// build (write/edit + shell) that plain Auto hides, but NOT network or denied
-// tools — the sandbox still gates the advertised ones at call time.
-func TestToolAdvertisedMemberAuto(t *testing.T) {
-	write := safetyTool{name: "write_file", safety: tools.Safety{SideEffect: tools.SideEffectWrite, Permission: tools.PermissionPrompt}}
-	shell := safetyTool{name: "bash", safety: tools.Safety{SideEffect: tools.SideEffectShell, Permission: tools.PermissionPrompt}}
-	read := safetyTool{name: "read_file", safety: tools.Safety{SideEffect: tools.SideEffectRead, Permission: tools.PermissionAllow}}
-	network := safetyTool{name: "net_tool", safety: tools.Safety{SideEffect: tools.SideEffectNetwork, Permission: tools.PermissionPrompt}}
-	denied := safetyTool{name: "blocked", safety: tools.Safety{SideEffect: tools.SideEffectRead, Permission: tools.PermissionDeny}}
-
-	// Plain Auto hides prompt-requiring mutators (the read-only member problem).
-	if ToolAdvertised(write, PermissionModeAuto) || ToolAdvertised(shell, PermissionModeAuto) {
-		t.Fatal("Auto must NOT advertise write/shell prompt tools")
-	}
-
-	for _, tool := range []tools.Tool{write, shell, read} {
-		if !ToolAdvertised(tool, PermissionModeMemberAuto) {
-			t.Fatalf("member-auto must advertise %q", tool.Name())
-		}
-	}
-	if ToolAdvertised(network, PermissionModeMemberAuto) {
-		t.Fatal("member-auto must NOT advertise a network prompt tool")
-	}
-	if ToolAdvertised(denied, PermissionModeMemberAuto) {
-		t.Fatal("member-auto must NOT advertise a denied tool")
-	}
-}

internal/agent/no_progress_stop_test.go

@@ -1,36 +0,0 @@
-package agent
-
-import "testing"
-
-func TestIsNoProgressStopRequiresFullStructure(t *testing.T) {
-	// The actual guardrail answer (any turn count) is recognized.
-	for _, turns := range []int{1, 3, 12} {
-		if !IsNoProgressStop(noOutputStopAnswer(turns)) {
-			t.Fatalf("the real no-output stop answer (turns=%d) should be recognized", turns)
-		}
-	}
-
-	// A legitimate message that merely QUOTES the marker must NOT be classified as
-	// a no-progress stop — that would wrongly hide a real session from /resume.
-	rejected := []string{
-		`The previous run ended "with no output (no visible text and no tool calls)" so here is what I did instead.`,
-		"with no output (no visible text and no tool calls)", // bare marker, no prefix/suffix
-		"Agent stopped after 3 turns.",                       // prefix only
-		"to avoid consuming tokens without making progress.", // suffix only
-		"Here is your e-commerce site.",                      // unrelated
-		"",
-		// Hostile input: prefix + marker + suffix ALL present, but the marker is
-		// quoted mid-sentence rather than being the guard's own structured answer.
-		// A loose prefix/contains/suffix check misclassifies this.
-		`Agent stopped after 3 turns. The marker is "with no output (no visible text and no tool calls)" so here it is: to avoid consuming tokens without making progress.`,
-		// Arbitrary text wedged between the marker and the suffix.
-		"Agent stopped after 3 turns with no output (no visible text and no tool calls) and then some more text to avoid consuming tokens without making progress.",
-		// Non-integer "turn count" between the prefix and the marker.
-		"Agent stopped after several turns with no output (no visible text and no tool calls) to avoid consuming tokens without making progress.",
-	}
-	for _, content := range rejected {
-		if IsNoProgressStop(content) {
-			t.Fatalf("content must not be classified as a no-progress stop: %q", content)
-		}
-	}
-}

internal/agent/parallel_tools.go

@@ -1,97 +0,0 @@
-package agent
-
-import (
-	"context"
-	"sync"
-
-	"github.com/Gitlawb/zero/internal/tools"
-)
-
-// Parallel read-ahead for tool batches. When a turn requests several
-// independent lookups (read_file + grep + glob is the common shape), executing
-// them one after another serializes pure I/O waits. A consecutive run of
-// auto-allowed read-only calls is executed concurrently instead; results are
-// then consumed in the original call order, so guard counters, message
-// ordering, abort semantics, and the surface's call/result event pairing are
-// byte-identical to sequential execution. Runs never span a mutating call: a
-// read that follows a write must observe the write, so eligibility is decided
-// per consecutive run, not per batch.
-
-// maxParallelReadTools bounds concurrent read-only tool executions in a turn.
-const maxParallelReadTools = 8
-
-// precomputedToolResult is one parallel read-ahead execution, keyed back to
-// its batch index by the caller.
-type precomputedToolResult struct {
-	result   ToolResult
-	abortErr error
-}
-
-// parallelSafeToolCall reports whether call may run concurrently with its
-// neighbors: the tool must exist, be side-effect-free (SideEffectRead), and be
-// auto-allowed for these args, so no interactive prompt or workspace mutation
-// is on the hot path. Loop-intercepted tools (ask_user, request_permissions)
-// and tool_search (mutates the deferred-tool set) stay sequential.
-func parallelSafeToolCall(registry *tools.Registry, call ToolCall, options Options) bool {
-	switch call.Name {
-	case "ask_user", tools.RequestPermissionsToolName, tools.ToolSearchToolName:
-		return false
-	}
-	tool, found := registry.Get(call.Name)
-	if !found || tool.Safety().SideEffect != tools.SideEffectRead {
-		return false
-	}
-	args := map[string]any{}
-	if call.Arguments != "" {
-		if err := decodeToolArguments(call.Arguments, &args); err != nil {
-			return false
-		}
-	}
-	return effectivePermission(tool, args) == tools.PermissionAllow
-}
-
-// executeParallelReadBatch runs calls[start:end] concurrently (bounded by
-// maxParallelReadTools) and returns results indexed relative to start. All
-// execution-side callbacks that can fire inside executeToolCall are serialized
-// behind one mutex: a permission prompt (a sandbox preflight can demand one
-// even for an auto-allowed read) must never appear twice at once on an
-// interactive front-end, and OnPermission event handlers append to shared
-// session-recording state without their own locking — two batched reads under
-// a granted extra root would otherwise race (the pre-batch serial loop never
-// had two callbacks in flight at once).
-func executeParallelReadBatch(ctx context.Context, registry *tools.Registry, calls []ToolCall, start, end int, permissionMode PermissionMode, options Options) []precomputedToolResult {
-	batchOptions := options
-	var callbackMutex sync.Mutex
-	if options.OnPermissionRequest != nil {
-		inner := options.OnPermissionRequest
-		batchOptions.OnPermissionRequest = func(ctx context.Context, request PermissionRequest) (PermissionDecision, error) {
-			callbackMutex.Lock()
-			defer callbackMutex.Unlock()
-			return inner(ctx, request)
-		}
-	}
-	if options.OnPermission != nil {
-		inner := options.OnPermission
-		batchOptions.OnPermission = func(event PermissionEvent) {
-			callbackMutex.Lock()
-			defer callbackMutex.Unlock()
-			inner(event)
-		}
-	}
-
-	results := make([]precomputedToolResult, end-start)
-	semaphore := make(chan struct{}, maxParallelReadTools)
-	var waitGroup sync.WaitGroup
-	for index := start; index < end; index++ {
-		waitGroup.Add(1)
-		go func(index int) {
-			defer waitGroup.Done()
-			semaphore <- struct{}{}
-			defer func() { <-semaphore }()
-			result, abortErr := executeToolCall(ctx, registry, calls[index], permissionMode, batchOptions)
-			results[index-start] = precomputedToolResult{result: result, abortErr: abortErr}
-		}(index)
-	}
-	waitGroup.Wait()
-	return results
-}

internal/agent/parallel_tools_test.go

@@ -1,204 +0,0 @@
-package agent
-
-import (
-	"context"
-	"fmt"
-	"sync"
-	"testing"
-	"time"
-
-	"github.com/Gitlawb/zero/internal/tools"
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-// probeTool records execution overlap and ordering so tests can assert what
-// actually ran concurrently.
-type probeTool struct {
-	name       string
-	sideEffect tools.SideEffect
-	delay      time.Duration
-	// shared, when set, receives this probe's start/end entries too, so a test
-	// can observe ordering ACROSS probes (e.g. reads vs a write barrier).
-	shared *probeLog
-
-	mu        sync.Mutex
-	active    int
-	maxActive int
-	log       []string
-}
-
-// probeLog is a mutex-guarded event log shared between probes.
-type probeLog struct {
-	mu      sync.Mutex
-	entries []string
-}
-
-func (log *probeLog) append(entry string) {
-	log.mu.Lock()
-	defer log.mu.Unlock()
-	log.entries = append(log.entries, entry)
-}
-
-func (log *probeLog) snapshot() []string {
-	log.mu.Lock()
-	defer log.mu.Unlock()
-	return append([]string(nil), log.entries...)
-}
-
-func (tool *probeTool) Name() string        { return tool.name }
-func (tool *probeTool) Description() string { return "test probe tool" }
-func (tool *probeTool) Parameters() tools.Schema {
-	return tools.Schema{
-		Type:                 "object",
-		Properties:           map[string]tools.PropertySchema{"id": {Type: "string"}},
-		AdditionalProperties: false,
-	}
-}
-func (tool *probeTool) Safety() tools.Safety {
-	return tools.Safety{SideEffect: tool.sideEffect, Permission: tools.PermissionAllow, Reason: "test"}
-}
-func (tool *probeTool) Run(_ context.Context, args map[string]any) tools.Result {
-	id, _ := args["id"].(string)
-	tool.mu.Lock()
-	tool.active++
-	if tool.active > tool.maxActive {
-		tool.maxActive = tool.active
-	}
-	tool.log = append(tool.log, "start:"+id)
-	tool.mu.Unlock()
-	if tool.shared != nil {
-		tool.shared.append("start:" + id)
-	}
-	time.Sleep(tool.delay)
-	if tool.shared != nil {
-		tool.shared.append("end:" + id)
-	}
-	tool.mu.Lock()
-	tool.active--
-	tool.log = append(tool.log, "end:"+id)
-	tool.mu.Unlock()
-	return tools.Result{Status: tools.StatusOK, Output: "probe " + id}
-}
-
-func probeCallEvents(callID, toolName, id string) []zeroruntime.StreamEvent {
-	return []zeroruntime.StreamEvent{
-		{Type: zeroruntime.StreamEventToolCallStart, ToolCallID: callID, ToolName: toolName},
-		{Type: zeroruntime.StreamEventToolCallDelta, ToolCallID: callID, ArgumentsFragment: fmt.Sprintf(`{"id":%q}`, id)},
-		{Type: zeroruntime.StreamEventToolCallEnd, ToolCallID: callID},
-	}
-}
-
-func TestParallelSafeToolCall(t *testing.T) {
-	registry := tools.NewRegistry()
-	registry.Register(&probeTool{name: "probe_read", sideEffect: tools.SideEffectRead})
-	registry.Register(&probeTool{name: "probe_write", sideEffect: tools.SideEffectWrite})
-
-	call := func(name, args string) ToolCall { return ToolCall{ID: "c", Name: name, Arguments: args} }
-	if !parallelSafeToolCall(registry, call("probe_read", `{"id":"a"}`), Options{}) {
-		t.Fatal("auto-allowed read tool must be parallel-safe")
-	}
-	if parallelSafeToolCall(registry, call("probe_write", `{"id":"a"}`), Options{}) {
-		t.Fatal("mutating tool must not be parallel-safe")
-	}
-	if parallelSafeToolCall(registry, call("unknown_tool", `{}`), Options{}) {
-		t.Fatal("unknown tool must not be parallel-safe")
-	}
-	if parallelSafeToolCall(registry, call("probe_read", `{"id":`), Options{}) {
-		t.Fatal("undecodable arguments must not be parallel-safe")
-	}
-	if parallelSafeToolCall(registry, call("ask_user", `{}`), Options{}) {
-		t.Fatal("loop-intercepted tools must stay sequential")
-	}
-}
-
-func TestRunExecutesConsecutiveReadsConcurrently(t *testing.T) {
-	probe := &probeTool{name: "probe_read", sideEffect: tools.SideEffectRead, delay: 60 * time.Millisecond}
-	registry := tools.NewRegistry()
-	registry.Register(probe)
-
-	turnOne := append(probeCallEvents("call-1", "probe_read", "a"), probeCallEvents("call-2", "probe_read", "b")...)
-	turnOne = append(turnOne, probeCallEvents("call-3", "probe_read", "c")...)
-	turnOne = append(turnOne, zeroruntime.StreamEvent{Type: zeroruntime.StreamEventDone})
-	provider := &mockProvider{
-		turns: [][]zeroruntime.StreamEvent{
-			turnOne,
-			{{Type: zeroruntime.StreamEventText, Content: "done"}, {Type: zeroruntime.StreamEventDone}},
-		},
-	}
-
-	var results []ToolResult
-	_, err := Run(context.Background(), "probe", provider, Options{
-		Registry:     registry,
-		OnToolResult: func(result ToolResult) { results = append(results, result) },
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if probe.maxActive < 2 {
-		t.Fatalf("consecutive read-only calls must overlap, max concurrency was %d", probe.maxActive)
-	}
-	// Results must still be recorded in original call order.
-	if len(results) != 3 || results[0].ToolCallID != "call-1" || results[1].ToolCallID != "call-2" || results[2].ToolCallID != "call-3" {
-		t.Fatalf("tool results out of order: %#v", results)
-	}
-	messages := provider.requests[1].Messages
-	var toolOrder []string
-	for _, message := range messages {
-		if message.Role == zeroruntime.MessageRoleTool {
-			toolOrder = append(toolOrder, message.ToolCallID)
-		}
-	}
-	if len(toolOrder) != 3 || toolOrder[0] != "call-1" || toolOrder[1] != "call-2" || toolOrder[2] != "call-3" {
-		t.Fatalf("recorded tool messages out of order: %v", toolOrder)
-	}
-}
-
-func TestRunParallelReadsNeverSpanMutatingCall(t *testing.T) {
-	shared := &probeLog{}
-	read := &probeTool{name: "probe_read", sideEffect: tools.SideEffectRead, delay: 30 * time.Millisecond, shared: shared}
-	write := &probeTool{name: "probe_write", sideEffect: tools.SideEffectWrite, shared: shared}
-	registry := tools.NewRegistry()
-	registry.Register(read)
-	registry.Register(write)
-
-	turnOne := append(probeCallEvents("call-1", "probe_read", "r1"), probeCallEvents("call-2", "probe_read", "r2")...)
-	turnOne = append(turnOne, probeCallEvents("call-3", "probe_write", "w")...)
-	turnOne = append(turnOne, probeCallEvents("call-4", "probe_read", "r3")...)
-	turnOne = append(turnOne, probeCallEvents("call-5", "probe_read", "r4")...)
-	turnOne = append(turnOne, zeroruntime.StreamEvent{Type: zeroruntime.StreamEventDone})
-	provider := &mockProvider{
-		turns: [][]zeroruntime.StreamEvent{
-			turnOne,
-			{{Type: zeroruntime.StreamEventText, Content: "done"}, {Type: zeroruntime.StreamEventDone}},
-		},
-	}
-
-	_, err := Run(context.Background(), "probe", provider, Options{Registry: registry})
-	if err != nil {
-		t.Fatal(err)
-	}
-
-	// Cross-probe ordering on the SHARED log: the write must start only after
-	// both first-batch reads finished, and both second-batch reads must start
-	// only after the write finished — batches never cross a mutating call.
-	log := shared.snapshot()
-	index := func(entry string) int {
-		for i, e := range log {
-			if e == entry {
-				return i
-			}
-		}
-		t.Fatalf("entry %q missing from shared log: %v", entry, log)
-		return -1
-	}
-	writeStart, writeEnd := index("start:w"), index("end:w")
-	if firstBatchMaxEnd := max(index("end:r1"), index("end:r2")); writeStart < firstBatchMaxEnd {
-		t.Fatalf("write started before the first read batch finished: %v", log)
-	}
-	if secondBatchMinStart := min(index("start:r3"), index("start:r4")); secondBatchMinStart < writeEnd {
-		t.Fatalf("second read batch started before the write finished: %v", log)
-	}
-	if read.maxActive < 2 {
-		t.Fatalf("reads within a batch must overlap, max concurrency was %d", read.maxActive)
-	}
-}

internal/agent/partition_cache_stable_test.go

@@ -1,44 +0,0 @@
-package agent
-
-import (
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/tools"
-)
-
-func TestPartitionToolsActiveAppendsLoadedToolAfterEagerBlock(t *testing.T) {
-	// Cache-stability property: loading a deferred tool must APPEND it after the
-	// always-eager tools, not sort it into the middle of them — otherwise the tool
-	// definitions (the provider's cacheable prefix) shift on every load. The old
-	// behaviour alpha-sorted the whole array, so "mcp__srv__alpha" jumped ahead of
-	// "read_file".
-	root := t.TempDir()
-	registry := tools.NewRegistry()
-	registry.Register(tools.NewReadFileTool(root)) // non-deferred, "read_file"
-	registry.Register(fakeToolSearchTool{})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__alpha", desc: "alpha"})
-	registry.Register(fakeDeferredTool{name: "mcp__srv__beta", desc: "beta"})
-
-	exposed, _ := partitionTools(registry, PermissionModeAuto, Options{DeferThreshold: 2}, map[string]bool{"mcp__srv__alpha": true})
-
-	pos := func(name string) int {
-		for i := range exposed {
-			if exposed[i].Name == name {
-				return i
-			}
-		}
-		return -1
-	}
-	readPos, searchPos, alphaPos := pos("read_file"), pos("tool_search"), pos("mcp__srv__alpha")
-	if readPos < 0 || searchPos < 0 || alphaPos < 0 {
-		t.Fatalf("expected read_file, tool_search and loaded alpha all exposed, got %#v", exposed)
-	}
-	// The eager tools must precede the loaded deferred tool (appended, not inserted).
-	if readPos >= alphaPos || searchPos >= alphaPos {
-		t.Fatalf("loaded deferred tool must be appended after the eager block; got read=%d search=%d alpha=%d", readPos, searchPos, alphaPos)
-	}
-	// beta was never loaded — it stays hidden.
-	if pos("mcp__srv__beta") != -1 {
-		t.Fatalf("unloaded deferred tool must stay hidden, got %#v", exposed)
-	}
-}

internal/agent/partition_cache_test.go

@@ -1,63 +0,0 @@
-package agent
-
-import (
-	"context"
-	"reflect"
-	"testing"
-
-	"github.com/Gitlawb/zero/internal/tools"
-	"github.com/Gitlawb/zero/internal/zeroruntime"
-)
-
-// countingSchemaTool records how many times its Parameters() schema is read, so a
-// test can prove the definition cache avoids re-rendering across turns.
-type countingSchemaTool struct {
-	name  string
-	calls *int
-}
-
-func (t countingSchemaTool) Name() string        { return t.name }
-func (t countingSchemaTool) Description() string { return "counts schema reads" }
-func (t countingSchemaTool) Parameters() tools.Schema {
-	*t.calls++
-	return tools.Schema{Type: "object", AdditionalProperties: false, Properties: map[string]tools.PropertySchema{
-		"x": {Type: "string"},
-	}}
-}
-func (t countingSchemaTool) Safety() tools.Safety {
-	return tools.Safety{SideEffect: tools.SideEffectRead, Permission: tools.PermissionAllow}
-}
-func (t countingSchemaTool) Run(_ context.Context, _ map[string]any) tools.Result {
-	return tools.Result{Status: tools.StatusOK}
-}
-
-// The cache renders a tool's schema once and reuses it across calls, and its
-// output is identical to the uncached path.
-func TestPartitionToolsCacheRendersOnceAndMatchesUncached(t *testing.T) {
-	calls := 0
-	registry := tools.NewRegistry()
-	registry.Register(countingSchemaTool{name: "alpha", calls: &calls})
-	registry.Register(countingSchemaTool{name: "beta", calls: &calls})
-
-	uncached, _ := partitionTools(registry, PermissionModeAuto, Options{}, map[string]bool{})
-
-	base := calls
-	cache := map[string]zeroruntime.ToolDefinition{}
-	first, _ := partitionToolsCached(registry, PermissionModeAuto, Options{}, map[string]bool{}, cache)
-	rendersAfterFirst := calls - base
-	second, _ := partitionToolsCached(registry, PermissionModeAuto, Options{}, map[string]bool{}, cache)
-	rendersAfterSecond := calls - base
-
-	if rendersAfterFirst != 2 {
-		t.Fatalf("first cached call should render 2 tools once each, rendered %d times", rendersAfterFirst)
-	}
-	if rendersAfterSecond != 2 {
-		t.Fatalf("second cached call must reuse the cache (no new schema reads), total renders = %d", rendersAfterSecond)
-	}
-	if !reflect.DeepEqual(first, second) {
-		t.Fatal("cached calls must return identical definitions")
-	}
-	if !reflect.DeepEqual(first, uncached) {
-		t.Fatal("cached output must match the uncached partitionTools output")
-	}
-}