ci: read main's sccache on every run, write it only from main

The sccache install step now runs on every event in ci.yml and
ci-test-suite.yml, and each Enable sccache step sets
SCCACHE_GHA_RW_MODE to READ_WRITE only on refs/heads/main and READ_ONLY
everywhere else (sccache docs/GHA.md; the action installs the latest
release, v0.19.0). Pull request and branch-dispatch runs, which now run
the full sharded suite, get main's compiler cache hits without writing
the refs/pull/N entries that once pushed the repo past its 10 GiB cache
cap. rust-cache save-if stays main-only and the Windows RUSTC_WRAPPER
guard from MAIN-GREEN is unchanged.

release-workflows.test.js now checks, per file, that every sccache
enable carries the off-main read-only switch and that no sccache install
is gated on main; both checks were confirmed to fail on a mutated copy.

Windows test legs keep debug 0: no test asserts on backtrace content
(only RUST_BACKTRACE pass-through lists and compile-time
Location::caller, which debug info does not affect).

Checks: node .github/scripts/release-workflows.test.js passed (211
paths, 11 wrapper cases x 12 steps, 15 hermetic checks, workflow
contracts); actionlint 1.7.12 with the repo's CI args (-ignore SC2129
-ignore SC2221 -ignore SC2222) clean on all workflows;
scripts/test_ci_migration_wiring.py 19 tests OK.

First drafted by a duplicate agent run in this worktree; reviewed and
tightened here.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

.github/AUTHOR_MAP

@@ -319,3 +319,5 @@ namezzb = sailorZ <[email protected]>
 alphapku-ux = Alpha Wu <[email protected]>
 BX166 = BX166 <[email protected]>
 c020627 = c020627 <[email protected]>
+[email protected] = Parman Mohammadalizadeh <[email protected]>
+Parman Mohammadalizadeh = Parman Mohammadalizadeh <[email protected]>

.github/scripts/release-workflows.test.js

@@ -32,6 +32,7 @@ function namedStep(source, name) {
 }
 
 const ci = read(".github/workflows/ci.yml");
+const ciSuite = read(".github/workflows/ci-test-suite.yml");
 const nightly = read(".github/workflows/nightly.yml");
 const candidate = read(".github/workflows/release-candidate.yml");
 const artifacts = read(".github/workflows/release-artifacts.yml");
@@ -46,28 +47,212 @@ const archiveInstaller = read("scripts/release/install.sh");
 const cliDispatcher = read("crates/cli/src/lib.rs");
 const runbook = read("docs/RELEASE_RUNBOOK.md");
 
-const ciTestJob = ci.slice(ci.indexOf("\n  test:\n"));
-assert.match(
-  ciTestJob,
-  /matrix\.os == 'ubuntu-latest'.*github\.event_name == 'pull_request'/,
-  "heavy pull requests must select the real Ubuntu test lane",
-);
-assert.match(
-  ciTestJob,
-  /cargo nextest run --workspace --all-features --locked --profile ci/,
-  "the Ubuntu pull-request lane must run workspace nextest",
-);
-assert.match(
-  ciTestJob,
-  /name: Linux test location \(CNB\)/,
-  "the non-PR CNB fallback must be named explicitly",
-);
+function jobBlock(source, job) {
+  const match = source.match(new RegExp(`^  ${job}:\\n([\\s\\S]*?)(?=^  \\S|(?![\\s\\S]))`, "m"));
+  assert.ok(match, `missing job: ${job}`);
+  return match[1];
+}
 
+function jobIf(source, job) {
+  const lines = jobBlock(source, job).split("\n");
+  const at = lines.findIndex((line) => /^    if:/.test(line));
+  assert.notEqual(at, -1, `${job} must declare an if`);
+  const first = lines[at].replace(/^    if:\s*/, "");
+  if (first !== ">-") return first.replace(/^'(.*)'$/, "$1");
+  const folded = [];
+  for (let i = at + 1; i < lines.length && /^      \S/.test(lines[i]); i += 1) folded.push(lines[i].trim());
+  return folded.join(" ");
+}
+
+function evaluate(expression, context) {
+  return Boolean(vm.runInNewContext(expression, context));
+}
+
+// The sharded suite: one archive per OS, partitions of it in parallel shards,
+// doctests once, and a check that the shards add up to the archive's list.
 assert.match(
-  namedStep(ciTestJob, "Build canonical executable for acceptance tests"),
+  namedStep(ciSuite, "Archive test binaries"),
+  /cargo nextest archive --workspace --all-features --locked --profile ci/,
+  "the suite must archive the same workspace, features and profile the old single run used",
+);
+assert.match(
+  namedStep(ciSuite, "Build canonical executable for acceptance tests"),
   /cargo build -p codewhale-cli --bin codewhale --all-features --locked/,
   "Engine acceptance must build the canonical executable first",
 );
+assert.match(
+  namedStep(ciSuite, "Run doctests"),
+  /cargo test --workspace --all-features --locked --doc/,
+  "nextest does not run doctests; the suite must run them once per OS",
+);
+const shardRun = namedStep(ciSuite, "Run tests");
+assert.match(shardRun, /--partition "count:\$\{\{ matrix\.shard \}\}\/\$\{\{ inputs\.shards \}\}"/);
+assert.match(shardRun, /-E "not \(\$NATIVE_FILTER\)"/, "every shard must exclude the native tests that shard 1 runs apart");
+assert.match(namedStep(ciSuite, "Run native containment tests"), /-E "\$NATIVE_FILTER"/);
+assert.match(ciSuite, /nextest-shard-count\.py check counts/, "the suite must prove the shards add up to the archive's list");
+for (const name of ["Run tests", "Run native containment tests", "Run doctests"]) {
+  const step = namedStep(ciSuite, name);
+  assert.match(step, /shell: bash/, `${name} must invoke the POSIX helper on Windows too`);
+}
+assert.match(ciSuite, /RUST_MIN_STACK: '16777216'/);
+
+// The native receipt script requires its tests in one JUnit report, so the
+// shard that runs them must select exactly the script's list.
+const receiptScript = read("scripts/compiled-host-native-receipt.py");
+const receiptNames = [
+  receiptScript.match(/^COMPILED = "([^"]+)"$/m)[1],
+  receiptScript.match(/^MEMORY = "([^"]+)"$/m)[1],
+  ...[...receiptScript.match(/^WINDOWS = \[([\s\S]*?)^\]/m)[1].matchAll(/"([^"]+)"/g)].map((m) => m[1]),
+];
+const nativeFilter = ciSuite.match(/NATIVE_FILTER: >-\n([\s\S]*?)\n    steps:/)[1];
+assert.deepEqual(
+  [...nativeFilter.matchAll(/test\(=([^)]+)\)/g)].map((m) => m[1]).sort(),
+  receiptNames.sort(),
+  "NATIVE_FILTER must list exactly the tests scripts/compiled-host-native-receipt.py requires",
+);
+
+// Shard counts: each caller's shards equals the length of its shard_list, the
+// suite's own list agrees with it, and every OS keeps a pipeline.
+const suiteCallers = {};
+for (const job of ["test-ubuntu", "test-macos", "test-windows"]) {
+  const block = jobBlock(ci, job);
+  const os = block.match(/^      os: (\S+)$/m)[1];
+  const shards = Number(block.match(/^      shards: (\d+)$/m)[1]);
+  const list = JSON.parse(block.match(/^      shard_list: '(.+)'$/m)[1]);
+  assert.deepEqual(list, Array.from({ length: shards }, (_, i) => i + 1), `${job} shard_list must be 1..shards`);
+  assert.match(block, /uses: \.\/\.github\/workflows\/ci-test-suite\.yml/);
+  assert.equal(jobIf(ci, job), jobIf(ci, "test-ubuntu"), `${job} must share the suite condition`);
+  suiteCallers[os] = shards;
+}
+assert.deepEqual(Object.keys(suiteCallers).sort(), ["macos-latest", "ubuntu-latest", "windows-latest"]);
+
+// Required contexts. The ruleset requires exactly "Lint" and "Test
+// (windows-latest)". Only Lint and the compile-only test-windows-pr job may
+// produce them; a shard, build or suite check must never share a name.
+const jobNames = (source) => [...source.matchAll(/^  [A-Za-z0-9_-]+:\n    name: (.+)$/gm)].map((m) => m[1]);
+assert.deepEqual(
+  [...jobNames(ci), ...jobNames(ciSuite)].filter((name) => /^Test \(/.test(name) || name === "Lint"),
+  ["Lint", "Test (windows-latest)"],
+  "only Lint and the compile-only Windows leg may use the required context names",
+);
+assert.match(jobBlock(ci, "test-windows-pr"), /^    name: Test \(windows-latest\)$/m);
+assert.equal(jobIf(ci, "test-windows-pr"), "github.event_name == 'pull_request'");
+assert.match(jobBlock(ci, "test-windows-pr"), /cargo check --workspace --all-features --all-targets|cargo check --workspace --all-targets --all-features --locked/);
+
+// Each gate that pull requests lost on 2026-11-07 (99cb5a0f4f) runs on a heavy
+// pull request again, plus on dispatch; a docs-only pull request keeps skipping
+// them; a scoped dispatch skips all of them.
+const gates = ["safety-gate", "msrv", "test-ubuntu", "test-macos", "test-windows", "portable-boundaries", "npm-wrapper-smoke", "mobile-smoke"];
+for (const job of gates) {
+  const guard = jobIf(ci, job);
+  const base = { inputs: { workspace_test_mode: "nextest" } };
+  const run = (event, heavy, extra = {}) =>
+    evaluate(guard, {
+      github: { event_name: event },
+      needs: { changes: { outputs: { heavy: String(heavy), mobile: "true", scoped: "false", ...extra } } },
+      ...base,
+    });
+  assert.equal(run("pull_request", true), true, `${job} must run on a heavy pull request`);
+  if (job !== "npm-wrapper-smoke") {
+    assert.equal(run("pull_request", false), false, `${job} must skip a docs-only pull request`);
+  }
+  assert.equal(run("workflow_dispatch", true), true, `${job} must run on a full dispatch`);
+  assert.equal(run("workflow_dispatch", true, { scoped: "true" }), false, `${job} must skip a scoped dispatch`);
+}
+assert.equal(
+  evaluate(jobIf(ci, "mobile-smoke"), {
+    github: { event_name: "pull_request" },
+    needs: { changes: { outputs: { heavy: "true", mobile: "false", scoped: "false" } } },
+  }),
+  false,
+  "mobile smoke keeps its path filter on pull requests",
+);
+for (const job of ["test-ubuntu", "test-macos", "test-windows"]) {
+  assert.equal(
+    evaluate(jobIf(ci, job), {
+      github: { event_name: "workflow_dispatch" },
+      inputs: { workspace_test_mode: "shared-process-twice" },
+      needs: { changes: { outputs: { heavy: "true", scoped: "false" } } },
+    }),
+    false,
+    `${job} must yield to the shared-process qualification`,
+  );
+}
+const sharedGuard = jobIf(ci, "test");
+for (const [event, mode, scoped, expected] of [
+  ["workflow_dispatch", "shared-process-twice", "false", true],
+  ["workflow_dispatch", "nextest", "false", false],
+  ["workflow_dispatch", "shared-process-twice", "true", false],
+  ["pull_request", "nextest", "false", false],
+]) {
+  assert.equal(
+    evaluate(sharedGuard, {
+      github: { event_name: event },
+      inputs: { workspace_test_mode: mode },
+      needs: { changes: { outputs: { heavy: "true", scoped } } },
+    }),
+    expected,
+    `shared-process job: ${event} ${mode} scoped=${scoped}`,
+  );
+}
+assert.equal(
+  evaluate(jobIf(ci, "scoped-test"), { needs: { changes: { outputs: { scoped: "true" } } } }),
+  true,
+);
+assert.equal(
+  evaluate(jobIf(ci, "scoped-test"), { needs: { changes: { outputs: { scoped: "false" } } } }),
+  false,
+);
+assert.match(ci, /^      expected_sha:\n        description: .*\n        required: true\n        type: string$/m, "expected_sha stays required");
+for (const input of ["crate", "filter", "os"]) {
+  assert.match(ci, new RegExp(`^      ${input}:\\n        description: .*\\n        required: false\\n`, "m"), `scoped dispatch input ${input}`);
+}
+
+// Concurrency: a new push to the same pull request or non-main branch cancels
+// the older run; runs on main and scoped dispatches never cancel anything.
+const concurrency = ci.match(/^concurrency:\n  group: (.+)\n  cancel-in-progress: (.+)$/m);
+assert.ok(concurrency, "ci.yml must declare its concurrency");
+const strip = (value) => value.replace(/^\$\{\{ /, "").replace(/ \}\}$/, "");
+const format = (template, ...args) => template.replace(/\{(\d)\}/g, (_, i) => args[Number(i)]);
+const groupOf = (context) =>
+  vm.runInNewContext(strip(concurrency[1]), { ...context, format });
+const cancelOf = (context) => vm.runInNewContext(strip(concurrency[2]), context);
+const dispatch = (ref, sha, inputs = {}) => ({
+  github: { event_name: "workflow_dispatch", ref, sha, workflow: "CI", run_id: `run-${sha}` },
+  inputs: { crate: "", filter: "", ...inputs },
+});
+const pr = (number, sha) => ({
+  github: { event_name: "pull_request", ref: `refs/pull/${number}/merge`, sha, workflow: "CI", event: { pull_request: { number } }, run_id: `run-${sha}` },
+  inputs: {},
+});
+assert.equal(groupOf(pr(7, "a")), groupOf(pr(7, "b")), "a new push to a pull request shares its group");
+assert.notEqual(groupOf(pr(7, "a")), groupOf(pr(8, "a")));
+assert.equal(cancelOf(pr(7, "a")), true);
+assert.equal(groupOf(dispatch("refs/heads/ci/x", "a")), groupOf(dispatch("refs/heads/ci/x", "b")), "a new dispatch on the same ci branch shares its group");
+assert.notEqual(groupOf(dispatch("refs/heads/ci/x", "a")), groupOf(dispatch("refs/heads/ci/y", "a")));
+assert.equal(cancelOf(dispatch("refs/heads/ci/x", "a")), true);
+assert.notEqual(groupOf(dispatch("refs/heads/main", "a")), groupOf(dispatch("refs/heads/main", "b")), "runs on main never share a group");
+assert.equal(cancelOf(dispatch("refs/heads/main", "a")), false, "runs on main never cancel");
+assert.notEqual(
+  groupOf(dispatch("refs/heads/ci/x", "a", { crate: "codewhale-tui" })),
+  groupOf(dispatch("refs/heads/ci/x", "a")),
+  "a scoped dispatch never shares the full run's group",
+);
+assert.equal(cancelOf(dispatch("refs/heads/ci/x", "a", { filter: "test(x)" })), false, "a scoped dispatch cancels nothing");
+
+// Security for pull requests from forks: the workflows use pull_request (no
+// pull_request_target), read-only permissions, no secrets, and write caches
+// only from main.
+for (const [name, source] of [["ci.yml", ci], ["ci-test-suite.yml", ciSuite]]) {
+  assert.doesNotMatch(source, /pull_request_target/, `${name} must not use pull_request_target`);
+  assert.doesNotMatch(source, /\$\{\{\s*secrets\./, `${name} must not read secrets`);
+  assert.match(source, /^permissions:\n  contents: read\n/m, `${name} must keep read-only default permissions`);
+  assert.doesNotMatch(source.replace(/^permissions:\n  contents: read\n/m, ""), /^\s+(?:contents|actions|id-token|pull-requests|packages):\s*write/m, `${name} must not grant write permissions`);
+  for (const match of source.matchAll(/Swatinem\/rust-cache@[^\n]*\n(?:\s+(?:with|cache-bin):[^\n]*\n)*\s+save-if: ([^\n]+)/g)) {
+    assert.match(match[1], /refs\/heads\/main|false/, `${name}: rust-cache must save only on main`);
+  }
+  assert.doesNotMatch(source, /actions\/cache(?:\/save)?@/, `${name} must not write caches outside rust-cache and sccache`);
+}
 const npmSmokeJob = ci.match(/^  npm-wrapper-smoke:\n([\s\S]*?)(?=^  \S)/m)?.[1];
 assert.ok(npmSmokeJob, "CI must retain the npm-wrapper job");
 assert.match(
@@ -79,13 +264,13 @@ assert.match(
   /run: node scripts\/release\/npm-wrapper-smoke\.js/,
 );
 const npmSmokeSteps = npmSmokeJob.split(/(?=^      - )/m).slice(1);
-// Exercise the workflow's actual Boolean guards. The smoke is manual-only
-// (workflow_dispatch); pull requests, pushes and schedules skip the job.
+// Exercise the workflow's actual Boolean guards. The smoke runs on pull
+// requests and dispatches (§180); pushes and schedules skip the job.
 const npmSmokeCases = [
   // name, event, heavy, OS, trusted, cache success, execute, Linux deps, CNB
-  ["own PR", "pull_request", true, "ubuntu-latest", true, true, false, false, false],
-  ["fork PR", "pull_request", true, "ubuntu-latest", false, true, false, false, false],
-  ["PR cache failure", "pull_request", true, "ubuntu-latest", false, false, false, false, false],
+  ["own PR", "pull_request", true, "ubuntu-latest", true, true, true, true, false],
+  ["fork PR", "pull_request", true, "ubuntu-latest", false, true, true, true, false],
+  ["PR cache failure", "pull_request", true, "ubuntu-latest", false, false, true, true, false],
   ["light PR", "pull_request", false, "ubuntu-latest", true, true, false, false, false],
   ["manual Ubuntu", "workflow_dispatch", true, "ubuntu-latest", true, true, true, true, false],
   ["main Ubuntu", "push", true, "ubuntu-latest", true, true, false, false, false],
@@ -95,15 +280,22 @@ const npmSmokeCases = [
   ["light main", "push", false, "ubuntu-latest", true, true, false, false, false],
   ["schedule", "schedule", true, "ubuntu-latest", true, true, false, false, false],
 ];
-// The sccache GitHub Actions backend is main-only, mirroring rust-cache's
-// save-if: pull requests never install or enable it (cache bloat, PLAN D).
+// The sccache GitHub Actions backend is read-only off main: every run reads
+// main's cache, and only main writes it (SCCACHE_GHA_RW_MODE), mirroring
+// rust-cache's save-if (cache bloat, PLAN D).
+for (const [name, source] of [["ci.yml", ci], ["ci-test-suite.yml", ciSuite]]) {
+  const enabled = source.match(/SCCACHE_GHA_ENABLED=true/g) ?? [];
+  const modes = source.match(/if \[ "\$\{GITHUB_REF\}" = refs\/heads\/main \]; then rw=READ_WRITE; else rw=READ_ONLY; fi\n\s+echo "SCCACHE_GHA_RW_MODE=\$\{rw\}"/g) ?? [];
+  assert.ok(enabled.length > 0, `${name} must enable sccache`);
+  assert.equal(modes.length, enabled.length, `${name}: every sccache enable must be read-only off main`);
+  assert.doesNotMatch(source, /sccache-action[^\n]*\n(?:[^\n]*\n){0,4}\s+if: [^\n]*github\.ref == 'refs\/heads\/main'/, `${name}: sccache must install on every run`);
+}
 const sccacheInstallStep = "mozilla-actions/[email protected]";
 for (const [label, event, heavy, os, trusted, cache, execute, linuxDeps, cnb] of npmSmokeCases) {
   const ref = event === "pull_request" ? "refs/pull/1/merge" : "refs/heads/main";
-  const onMain = ref === "refs/heads/main";
-  const installed = execute && onMain;
+  const installed = execute;
   const context = {
-    needs: { changes: { outputs: { heavy: String(heavy), trusted: String(trusted) } } },
+    needs: { changes: { outputs: { heavy: String(heavy), trusted: String(trusted), scoped: "false" } } },
     github: { event_name: event, ref },
     matrix: { os },
     steps: { sccache: { outcome: installed ? (cache ? "success" : "failure") : "skipped" } },
@@ -688,7 +880,8 @@ assert.doesNotMatch(
 // itself runs none: its parity job calls release-parity.yml.
 let hermeticInvocations = 0;
 for (const [label, workflow, expected] of [
-  ["CI", ci, 6],
+  ["CI", ci, 5],
+  ["CI test suite", ciSuite, 3],
   ["release", release, 0],
   ["release parity", parityWorkflow, 4],
   ["CNB", cnb, 3],
@@ -707,11 +900,6 @@ for (const [label, workflow, expected] of [
   }
   hermeticInvocations += commands.length;
 }
-for (const name of ["Run tests", "Run doctests"]) {
-  const step = namedStep(ciTestJob, name);
-  assert.match(step, /shell: bash/, `${name} must invoke the POSIX helper on Windows too`);
-  assert.match(step, /RUST_MIN_STACK: '16777216'/);
-}
 console.log(`Hermetic Rust workflow invocations OK: ${hermeticInvocations} checks passed.`);
 
 const nextest = read(".config/nextest.toml");
@@ -853,6 +1041,7 @@ for (const [name, source] of [
   ["release-parity.yml", parityWorkflow],
   ["release-republish.yml", republish],
   ["ci.yml", ci],
+  ["ci-test-suite.yml", ciSuite],
   ["nightly.yml", nightly],
 ]) {
   assert.deepEqual(

.github/workflows/ci-test-suite.yml

@@ -0,0 +1,388 @@
+name: CI test suite
+
+# Called by ci.yml, once per operating system, so each OS is its own pipeline:
+# one build of the whole workspace's test binaries, parallel shards that run
+# partitions of that one nextest archive, and a check that the shards together
+# ran exactly the tests the archive lists. Running per OS means Linux reports
+# as soon as Linux is done instead of waiting for the slowest build.
+#
+# Trust boundary: this runs with the caller's event. For a pull request from a
+# fork that is the read-only token and no secrets; nothing here writes a cache
+# (rust-cache and sccache save only on main) or uses a write permission. The
+# artifacts passed between jobs live inside one run.
+
+on:
+  workflow_call:
+    inputs:
+      os:
+        description: Hosted runner label
+        required: true
+        type: string
+      shards:
+        description: Number of shards, equal to the length of shard_list
+        required: true
+        type: number
+      shard_list:
+        description: JSON array of shard numbers, for example [1,2,3,4]
+        required: true
+        type: string
+      build_jobs:
+        description: CARGO_BUILD_JOBS for the test build
+        required: true
+        type: string
+      debug:
+        description: Debug-table level for the test build (CARGO_PROFILE_*_DEBUG)
+        required: true
+        type: string
+
+permissions:
+  contents: read
+
+env:
+  CARGO_TERM_COLOR: always
+  CARGO_INCREMENTAL: 0
+  RUSTFLAGS: -Dwarnings
+  RUST_MIN_STACK: 8388608
+
+jobs:
+  build:
+    name: Build
+    # A cold hosted Mac compiled for 65 minutes with one Cargo job (run
+    # 43216596942) and the old one-job cap was 210; this job only compiles, so
+    # it keeps real margin without the test run in it.
+    timeout-minutes: 150
+    runs-on: ${{ inputs.os }}
+    env:
+      # Test builds compile all features of the whole workspace. One Cargo job
+      # kept hosted runners inside memory (exit 143 on 2026-10-19 with four);
+      # trimmed debug tables shrink compiler and linker memory so Linux and
+      # Windows can run two. Assertions and every test are unchanged.
+      CARGO_BUILD_JOBS: ${{ inputs.build_jobs }}
+      CARGO_PROFILE_DEV_DEBUG: ${{ inputs.debug }}
+      CARGO_PROFILE_TEST_DEBUG: ${{ inputs.debug }}
+    steps:
+      - uses: actions/checkout@v7
+      - name: Test Windows installer PATH helper
+        if: inputs.os == 'windows-latest'
+        shell: pwsh
+        run: ./scripts/installer/update-user-path.tests.ps1
+      - name: Install NSIS for Windows installer regression
+        if: inputs.os == 'windows-latest'
+        shell: pwsh
+        run: ./scripts/installer/install-nsis.ps1
+      - name: Test Windows installer PATH regression
+        if: inputs.os == 'windows-latest'
+        shell: pwsh
+        run: ./scripts/installer/installer-path-regression.tests.ps1 -AllowUserPathMutation
+      - uses: dtolnay/rust-toolchain@stable
+      - name: Install macOS LLVM tools
+        if: inputs.os == 'macos-latest'
+        run: rustup component add llvm-tools --toolchain stable
+      - uses: mozilla-actions/[email protected]
+        id: sccache
+        continue-on-error: true
+        # Every run reads the GitHub Actions cache main wrote; only main writes (see Enable sccache).
+      - name: Enable sccache
+        if: steps.sccache.outcome == 'success'
+        shell: bash
+        run: |
+          echo "SCCACHE_GHA_ENABLED=true" >> "${GITHUB_ENV}"
+          if [ "${GITHUB_REF}" = refs/heads/main ]; then rw=READ_WRITE; else rw=READ_ONLY; fi
+          echo "SCCACHE_GHA_RW_MODE=${rw}" >> "${GITHUB_ENV}"
+          echo "RUSTC_WRAPPER=sccache" >> "${GITHUB_ENV}"
+          echo "SCCACHE_IGNORE_SERVER_IO_ERROR=1" >> "${GITHUB_ENV}"
+      - name: Install Linux system dependencies
+        if: inputs.os == 'ubuntu-latest'
+        run: |
+          bash scripts/ci-apt-install.sh libdbus-1-dev pkg-config bubblewrap apparmor-profiles
+          sh scripts/prepare-linux-test-sandbox.sh
+      - name: Ubuntu build headroom
+        # Reuse the same fixed hosted-runner provisioning as Safety and Lint.
+        # Exit 143 is a shutdown observation, not proof of OOM or timeout.
+        if: inputs.os == 'ubuntu-latest'
+        shell: bash
+        run: bash scripts/prepare-ubuntu-build-headroom.sh
+      - uses: Swatinem/rust-cache@v2
+        with:
+          cache-bin: false
+          save-if: ${{ github.ref == 'refs/heads/main' }}
+      - uses: taiki-e/install-action@nextest
+      - uses: actions/[email protected]
+        with:
+          node-version: 22
+      - name: Build canonical executable for acceptance tests
+        shell: bash
+        run: cargo build -p codewhale-cli --bin codewhale --all-features --locked
+        env:
+          RUSTC_WRAPPER: ${{ inputs.os != 'windows-latest' && env.RUSTC_WRAPPER || '' }}
+      - name: Windows Terminal clipboard-to-draft acceptance
+        if: inputs.os == 'windows-latest'
+        timeout-minutes: 5
+        shell: pwsh
+        run: node scripts/release/windows-terminal-clipboard.mjs target/debug/codewhale.exe "${{ runner.temp }}/windows-clipboard"
+      - name: Retain Windows Terminal clipboard-to-draft evidence
+        if: ${{ !cancelled() && inputs.os == 'windows-latest' }}
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: windows-clipboard-${{ github.sha }}-${{ github.run_attempt }}
+          path: ${{ runner.temp }}/windows-clipboard/
+          if-no-files-found: error
+          retention-days: 7
+      - name: Archive test binaries
+        # One compile per OS. Shards run this archive with
+        # `cargo nextest run --archive-file`, so the suite is the same test
+        # binaries, built once. Doctests are not in the archive; they run
+        # below, once, while the library artifacts are still warm.
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p "$RUNNER_TEMP/test-archive" "$RUNNER_TEMP/test-expected"
+          cargo nextest archive --workspace --all-features --locked --profile ci \
+            --archive-file "$RUNNER_TEMP/test-archive/nextest-archive.tar.zst"
+          exe=codewhale
+          if [ "$RUNNER_OS" = Windows ]; then exe=codewhale.exe; fi
+          cp "target/debug/$exe" "$RUNNER_TEMP/test-archive/$exe"
+        env:
+          RUSTC_WRAPPER: ${{ inputs.os != 'windows-latest' && env.RUSTC_WRAPPER || '' }}
+      - name: Count the tests the archive selects
+        shell: bash
+        run: |
+          set -euo pipefail
+          cargo nextest list --archive-file "$RUNNER_TEMP/test-archive/nextest-archive.tar.zst" \
+            --workspace-remap . --profile ci --message-format json > "$RUNNER_TEMP/test-listing.json"
+          tests="$(python3 -I scripts/ci/nextest-shard-count.py list "$RUNNER_TEMP/test-listing.json")"
+          printf '{"os": "%s", "tests": %s, "shards": %s}\n' "${{ inputs.os }}" "$tests" "${{ inputs.shards }}" \
+            > "$RUNNER_TEMP/test-expected/expected-${{ inputs.os }}.json"
+          cat "$RUNNER_TEMP/test-expected/expected-${{ inputs.os }}.json"
+      - name: Run doctests
+        # nextest does not run doctests; this keeps them once per OS.
+        shell: bash
+        run: sh scripts/with-hermetic-test-home.sh cargo test --workspace --all-features --locked --doc
+        env:
+          RUSTC_WRAPPER: ${{ inputs.os != 'windows-latest' && env.RUSTC_WRAPPER || '' }}
+          RUST_MIN_STACK: '16777216'
+      - name: Lockfile drift guard
+        run: git diff --exit-code -- Cargo.lock
+      - name: sccache stats
+        if: steps.sccache.outcome == 'success'
+        continue-on-error: true
+        shell: bash
+        run: sccache --show-stats
+      - name: Retain test archive for the shards
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: test-archive-${{ inputs.os }}
+          path: ${{ runner.temp }}/test-archive/
+          if-no-files-found: error
+          retention-days: 1
+          compression-level: 0
+      - name: Retain expected test count
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: test-expected-${{ inputs.os }}
+          path: ${{ runner.temp }}/test-expected/
+          if-no-files-found: error
+          retention-days: 7
+
+  shard:
+    name: Shard ${{ matrix.shard }}/${{ inputs.shards }}
+    needs: build
+    timeout-minutes: 60
+    runs-on: ${{ inputs.os }}
+    strategy:
+      # A failure in one shard must not erase the others' results; the suite
+      # gate needs every conclusion.
+      fail-fast: false
+      matrix:
+        shard: ${{ fromJSON(inputs.shard_list) }}
+    env:
+      CODEWHALE_EXT_HOST_TESTS: '1'
+      CODEWHALE_EXT_HOST_BUN_TESTS: '1'
+      # Give test threads the stack the product gives itself; the reasoning is
+      # on the shared-process step above and in scripts/dev-cargo.sh.
+      RUST_MIN_STACK: '16777216'
+      # The exact-image Native containment tests (and, on Windows, the LPAC
+      # owner and profile tests) that scripts/compiled-host-native-receipt.py
+      # requires in ONE JUnit report. Shard 1 runs them in their own nextest
+      # invocation and every shard excludes them from its partition, so each
+      # still runs exactly once. release-workflows.test.js keeps this list equal
+      # to the script's.
+      NATIVE_FILTER: >-
+        package(codewhale-engine) & kind(lib) & (
+        test(=extension_host::tests::compiled_native_host_cannot_read_secrets_or_write_outside_its_data_dir) |
+        test(=extension_host::tests::compiled_native_host_memory_cap_is_enforced) |
+        test(=extension_host::windows::tests::windows_native_lpac_node_owner_boundary) |
+        test(=extension_host::windows::tests::windows_native_lpac_bun_owner_boundary) |
+        test(=extension_host::windows::tests::windows_native_lpac_compiled_owner_boundary) |
+        test(=extension_host::windows::tests::windows_native_profiles_are_distinct_and_acl_grant_refuses_junctions) |
+        test(=extension_host::windows::tests::windows_argv_and_environment_keep_exact_values_and_reject_nul) |
+        test(=extension_host::windows::tests::windows_profile_retirement_removes_only_its_grants_and_inherited_data_on_restarts) |
+        test(=extension_host::windows::tests::windows_profile_directory_budget_and_recorded_identity_refuse_before_overwrite))
+    steps:
+      - uses: actions/checkout@v7
+      - name: Point terminal failure diagnostics at the runner temp directory
+        shell: bash
+        run: echo "QA_PTY_DIAGNOSTICS_DIR=${RUNNER_TEMP}/pty-failures" >> "${GITHUB_ENV}"
+      - uses: dtolnay/rust-toolchain@stable
+      - name: Install Linux system dependencies
+        if: inputs.os == 'ubuntu-latest'
+        run: |
+          bash scripts/ci-apt-install.sh libdbus-1-dev pkg-config bubblewrap apparmor-profiles
+          sh scripts/prepare-linux-test-sandbox.sh
+      - uses: taiki-e/install-action@nextest
+      - uses: actions/[email protected]
+        # The extension-host integration tests spawn the real bundle under
+        # Node >= 22.19; CODEWHALE_EXT_HOST_TESTS makes a missing Node a
+        # failure instead of a silent skip.
+        with:
+          node-version: 22
+      - uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2
+        with:
+          bun-version: 1.4.0
+      - name: Build matching compiled-host input for required Native containment receipts
+        shell: bash
+        run: bash scripts/ci/compiled-host-test-input.sh
+      - uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8
+        with:
+          name: test-archive-${{ inputs.os }}
+          path: ${{ runner.temp }}/test-archive
+      - name: Place the canonical executable for acceptance tests
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p target/debug
+          cp "$RUNNER_TEMP"/test-archive/codewhale* target/debug/
+          if [ "$RUNNER_OS" != Windows ]; then chmod +x target/debug/codewhale; fi
+      - name: Run native containment tests
+        if: matrix.shard == 1
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p "$RUNNER_TEMP/test-count" "$RUNNER_TEMP/nextest-extract"
+          sh scripts/with-hermetic-test-home.sh cargo nextest run \
+            --archive-file "$RUNNER_TEMP/test-archive/nextest-archive.tar.zst" \
+            --extract-to "$RUNNER_TEMP/nextest-extract" --extract-overwrite \
+            --workspace-remap . --profile ci -E "$NATIVE_FILTER"
+          cp target/nextest/ci/junit.xml "$RUNNER_TEMP/native-junit.xml"
+          python3 -I scripts/ci/nextest-shard-count.py junit "$RUNNER_TEMP/native-junit.xml" > "$RUNNER_TEMP/native-count.txt"
+      - name: Run tests
+        # Same test binaries as `cargo test`, run by cargo-nextest: one
+        # process per test, all runner cores busy, slow tests named instead
+        # of stalling the binary. `.config/nextest.toml` serializes the PTY
+        # binary and bounds the integration binary that spawns the real
+        # executable; retries are off, so a flake is a red run, not a hidden
+        # one. This shard runs partition k of N of everything except the
+        # native containment tests; test-suite proves the shards add up to the
+        # archive's full list.
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p "$RUNNER_TEMP/test-count" "$RUNNER_TEMP/nextest-extract"
+          monitor=
+          if [ "$RUNNER_OS" = Linux ]; then
+            # Headroom trace: if the runner is terminated again, the last lines
+            # say whether memory or disk ran out.
+            ( while sleep 30; do
+                echo "[headroom] $(free -m | awk '/^Mem:/{print "mem used " $3 "/" $2 " MiB"} /^Swap:/{print "swap used " $3 "/" $2 " MiB"}' | paste -sd ' ') disk free $(df -h / | awk 'NR==2{print $4}')"
+              done ) &
+            monitor=$!
+          fi
+          status=0
+          sh scripts/with-hermetic-test-home.sh cargo nextest run \
+            --archive-file "$RUNNER_TEMP/test-archive/nextest-archive.tar.zst" \
+            --extract-to "$RUNNER_TEMP/nextest-extract" --extract-overwrite \
+            --workspace-remap . --profile ci \
+            --partition "count:${{ matrix.shard }}/${{ inputs.shards }}" \
+            -E "not ($NATIVE_FILTER)" || status=$?
+          if [ -n "$monitor" ]; then kill "$monitor" 2>/dev/null || true; fi
+          if [ "$status" -ne 0 ]; then exit "$status"; fi
+          tests="$(python3 -I scripts/ci/nextest-shard-count.py junit target/nextest/ci/junit.xml)"
+          if [ -f "$RUNNER_TEMP/native-count.txt" ]; then tests=$((tests + $(cat "$RUNNER_TEMP/native-count.txt"))); fi
+          printf '{"os": "%s", "shard": %s, "tests": %s}\n' "${{ inputs.os }}" "${{ matrix.shard }}" "$tests" \
+            > "$RUNNER_TEMP/test-count/shard-${{ inputs.os }}-${{ matrix.shard }}.json"
+          cat "$RUNNER_TEMP/test-count/shard-${{ inputs.os }}-${{ matrix.shard }}.json"
+      - name: Retain sealed terminal failure diagnostics
+        if: failure()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: terminal-failures-${{ runner.os }}-${{ matrix.shard }}-${{ github.sha }}-${{ github.run_attempt }}
+          path: ${{ runner.temp }}/pty-failures/
+          include-hidden-files: true
+          if-no-files-found: ignore
+          retention-days: 7
+      - name: Export exact compiled-image Native containment receipt
+        # JUnit is from the same archive's native invocation: no second
+        # compile. The official entire CI run must be green before release can
+        # consume this artifact. Missing/changed inputs or silent test skips
+        # are red.
+        if: matrix.shard == 1
+        shell: bash
+        run: |
+          set -euo pipefail
+          test "$(git rev-parse HEAD)" = "$GITHUB_SHA"
+          git diff --exit-code
+          git diff --cached --exit-code
+          python3 scripts/compiled-host-native-receipt.py \
+            --junit "$RUNNER_TEMP/native-junit.xml" \
+            --compiled-image "$CODEWHALE_COMPILED_HOST_TEST_BINARY" \
+            --bundle crates/engine/extension-host/dist/codewhale-extension-host.mjs \
+            --bun "$CODEWHALE_COMPILED_HOST_TEST_BUN" --source-sha "$GITHUB_SHA" \
+            --image-platform "$CODEWHALE_COMPILED_HOST_TEST_PLATFORM" --image-arch "$CODEWHALE_COMPILED_HOST_TEST_ARCH" \
+            --host-sha256 "$CODEWHALE_COMPILED_HOST_TEST_SHA256" \
+            --runtime-sha256 "$CODEWHALE_COMPILED_HOST_TEST_RUNTIME_SHA256" \
+            --bundle-sha256 "$CODEWHALE_COMPILED_HOST_TEST_BUNDLE_SHA256" \
+            --output "$RUNNER_TEMP/native-compiled-host"
+      - name: Retain exact compiled image and Native receipt
+        if: matrix.shard == 1
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: native-compiled-host-${{ runner.os }}-${{ runner.arch }}
+          path: ${{ runner.temp }}/native-compiled-host/
+          if-no-files-found: error
+          retention-days: 30
+      - name: Lockfile drift guard
+        run: git diff --exit-code -- Cargo.lock
+      - name: Retain shard test count
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: test-count-${{ inputs.os }}-${{ matrix.shard }}
+          path: ${{ runner.temp }}/test-count/
+          if-no-files-found: error
+          retention-days: 7
+
+  check:
+    # The job that answers "did this OS's whole suite pass": the build and every
+    # shard succeeded, and the shards together ran exactly the tests the archive
+    # lists. Not a required check (§180).
+    name: Suite check
+    needs: [build, shard]
+    if: '!cancelled()'
+    timeout-minutes: 10
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+        with:
+          sparse-checkout: scripts/ci
+      - name: Every build and shard succeeded
+        shell: bash
+        env:
+          BUILD_RESULT: ${{ needs.build.result }}
+          SHARD_RESULT: ${{ needs.shard.result }}
+        run: |
+          echo "build: ${BUILD_RESULT}"
+          echo "shard: ${SHARD_RESULT}"
+          if [ "${BUILD_RESULT}" != success ] || [ "${SHARD_RESULT}" != success ]; then
+            echo "::error::The sharded suite did not complete successfully."
+            exit 1
+          fi
+      - uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8
+        with:
+          name: test-expected-${{ inputs.os }}
+          path: counts
+      - uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8
+        with:
+          pattern: test-count-${{ inputs.os }}-*
+          path: counts
+          merge-multiple: true
+      - name: Shards ran every listed test exactly once
+        run: python3 -I scripts/ci/nextest-shard-count.py check counts

.github/workflows/ci.yml

@@ -16,13 +16,32 @@ on:
         options:
           - nextest
           - shared-process-twice
+      crate:
+        description: Scoped run, in a few minutes. One workspace package (for example codewhale-tui); with crate or filter set only the scoped test runs, not the full suite
+        required: false
+        default: ''
+        type: string
+      filter:
+        description: Scoped run, in a few minutes. A nextest filterset (for example 'test(auto_review)'), alone or with crate; with crate or filter set only the scoped test runs
+        required: false
+        default: ''
+        type: string
+      os:
+        description: Operating systems for a scoped run, comma separated (ubuntu-latest, macos-latest, windows-latest, or all); ignored by a full run
+        required: false
+        default: ubuntu-latest
+        type: string
 
 permissions:
   contents: read
 
+# A new push to a pull request, or a new dispatch on the same non-main branch
+# (an agent's ci/<task> branch), cancels the older run still in progress.
+# Runs on main are keyed by SHA and never cancel each other, and a scoped
+# dispatch gets its own group so it cannot cancel a full run on the same branch.
 concurrency:
-  group: ${{ github.event_name == 'pull_request' && format('ci-pr-{0}', github.event.pull_request.number) || format('ci-{0}-{1}', github.workflow, github.sha) }}
-  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+  group: ${{ github.event_name == 'pull_request' && format('ci-pr-{0}', github.event.pull_request.number) || (github.event_name == 'workflow_dispatch' && (inputs.crate != '' || inputs.filter != '') && format('ci-scoped-{0}', github.run_id)) || (github.ref != 'refs/heads/main' && format('ci-ref-{0}', github.ref)) || format('ci-{0}-{1}', github.workflow, github.sha) }}
+  cancel-in-progress: ${{ github.ref != 'refs/heads/main' && !(github.event_name == 'workflow_dispatch' && (inputs.crate != '' || inputs.filter != '')) }}
 
 env:
   CARGO_TERM_COLOR: always
@@ -45,6 +64,8 @@ jobs:
       mobile: ${{ steps.detect.outputs.mobile }}
       actions: ${{ steps.detect.outputs.actions }}
       trusted: ${{ steps.trust.outputs.trusted }}
+      scoped: ${{ steps.scope.outputs.scoped }}
+      scoped_os: ${{ steps.scope.outputs.scoped_os }}
     steps:
       - name: Classify event trust
         id: trust
@@ -64,6 +85,48 @@ jobs:
           else
             echo "trusted=false" >> "$GITHUB_OUTPUT"
           fi
+      - name: Classify scoped dispatch
+        id: scope
+        shell: bash
+        env:
+          EVENT_NAME: ${{ github.event_name }}
+          SCOPE_CRATE: ${{ inputs.crate }}
+          SCOPE_FILTER: ${{ inputs.filter }}
+          SCOPE_OS: ${{ inputs.os }}
+        run: |
+          set -euo pipefail
+          # A dispatch with a crate or a filter runs only the scoped test job.
+          # Inputs reach the shell only through env, and the crate and OS
+          # names are validated before anything uses them.
+          scoped=false
+          scoped_os='["ubuntu-latest"]'
+          if [[ "${EVENT_NAME}" == "workflow_dispatch" && ( -n "${SCOPE_CRATE}" || -n "${SCOPE_FILTER}" ) ]]; then
+            scoped=true
+            if [[ -n "${SCOPE_CRATE}" && ! "${SCOPE_CRATE}" =~ ^[A-Za-z0-9_-]+$ ]]; then
+              echo "::error::crate must be a package name (letters, digits, - and _)." >&2
+              exit 1
+            fi
+            oses=()
+            IFS=',' read -r -a requested <<< "${SCOPE_OS:-ubuntu-latest}"
+            for name in "${requested[@]}"; do
+              name="${name//[[:space:]]/}"
+              case "${name}" in
+                ubuntu|ubuntu-latest|linux) oses+=("ubuntu-latest") ;;
+                macos|macos-latest) oses+=("macos-latest") ;;
+                windows|windows-latest) oses+=("windows-latest") ;;
+                all) oses+=("ubuntu-latest" "macos-latest" "windows-latest") ;;
+                "") ;;
+                *)
+                  echo "::error::os entries must be ubuntu-latest, macos-latest, windows-latest or all." >&2
+                  exit 1
+                  ;;
+              esac
+            done
+            if [[ "${#oses[@]}" -eq 0 ]]; then oses=("ubuntu-latest"); fi
+            scoped_os="$(printf '%s\n' "${oses[@]}" | sort -u | jq -R . | jq -sc .)"
+          fi
+          echo "scoped=${scoped}" >> "${GITHUB_OUTPUT}"
+          echo "scoped_os=${scoped_os}" >> "${GITHUB_OUTPUT}"
       - uses: actions/checkout@v7
         with:
           fetch-depth: 0
@@ -197,6 +260,8 @@ jobs:
 
   versions:
     name: Version drift
+    needs: changes
+    if: needs.changes.outputs.scoped != 'true'
     timeout-minutes: 15
     runs-on: ubuntu-latest
     steps:
@@ -274,6 +339,8 @@ jobs:
 
   plugin-conversion:
     name: Plugin conversion
+    needs: changes
+    if: needs.changes.outputs.scoped != 'true'
     timeout-minutes: 5
     runs-on: ubuntu-latest
     steps:
@@ -293,6 +360,8 @@ jobs:
 
   integrations:
     name: Integrations
+    needs: changes
+    if: needs.changes.outputs.scoped != 'true'
     timeout-minutes: 15
     runs-on: ubuntu-latest
     steps:
@@ -349,7 +418,7 @@ jobs:
   extension-host-runtimes:
     name: Extension host runtimes (${{ matrix.os }})
     needs: changes
-    if: needs.changes.outputs.heavy == 'true'
+    if: needs.changes.outputs.heavy == 'true' && needs.changes.outputs.scoped != 'true'
     timeout-minutes: 15
     strategy:
       fail-fast: false
@@ -388,6 +457,8 @@ jobs:
 
   vscode-extension:
     name: VS Code extension
+    needs: changes
+    if: needs.changes.outputs.scoped != 'true'
     timeout-minutes: 15
     runs-on: ubuntu-latest
     defaults:
@@ -421,7 +492,7 @@ jobs:
   safety-gate:
     name: Safety gate
     needs: changes
-    if: github.event_name == 'workflow_dispatch' && needs.changes.outputs.heavy == 'true'
+    if: needs.changes.outputs.heavy == 'true' && needs.changes.outputs.scoped != 'true'
     timeout-minutes: 30
     runs-on: ubuntu-latest
     steps:
@@ -431,16 +502,18 @@ jobs:
           toolchain: stable
       - uses: mozilla-actions/[email protected]
         id: sccache
-        # The GitHub Actions cache backend is main-only, mirroring
-        # rust-cache's save-if: PR runs wrote thousands of refs/pull/N
-        # entries that pushed the repo past its 10 GiB cache cap.
-        if: github.ref == 'refs/heads/main'
+        # Every run reads the cache main wrote; only main writes it
+        # (SCCACHE_GHA_RW_MODE in Enable sccache), mirroring rust-cache's
+        # save-if. PR runs once wrote thousands of refs/pull/N entries that
+        # pushed the repo past its 10 GiB cache cap.
         continue-on-error: true
       - name: Enable sccache
         if: steps.sccache.outcome == 'success'
         shell: bash
         run: |
           echo "SCCACHE_GHA_ENABLED=true" >> "${GITHUB_ENV}"
+          if [ "${GITHUB_REF}" = refs/heads/main ]; then rw=READ_WRITE; else rw=READ_ONLY; fi
+          echo "SCCACHE_GHA_RW_MODE=${rw}" >> "${GITHUB_ENV}"
           echo "RUSTC_WRAPPER=sccache" >> "${GITHUB_ENV}"
           echo "SCCACHE_IGNORE_SERVER_IO_ERROR=1" >> "${GITHUB_ENV}"
       - name: Install Linux system dependencies
@@ -475,6 +548,7 @@ jobs:
   lint:
     name: Lint
     needs: changes
+    if: needs.changes.outputs.scoped != 'true'
     # Measured on 89ee629e with runner headroom: clippy 10 min, runtime-contract
     # budget 16 min, and persistence-backlog still running at the old 45-minute
     # cancellation. Commands, filters and budgets are unchanged.
@@ -494,13 +568,15 @@ jobs:
         # Cache bootstrap failures (e.g. GitHub 504s fetching the sccache
         # binary) degrade to an uncached build instead of failing product CI.
         continue-on-error: true
-        # Main-only GitHub Actions cache backend; see the Safety gate job.
-        if: needs.changes.outputs.heavy == 'true' && github.ref == 'refs/heads/main'
+        # Every run reads the GitHub Actions cache main wrote; only main writes (see Enable sccache).
+        if: needs.changes.outputs.heavy == 'true'
       - name: Enable sccache
         if: needs.changes.outputs.heavy == 'true' && steps.sccache.outcome == 'success'
         shell: bash
         run: |
           echo "SCCACHE_GHA_ENABLED=true" >> "${GITHUB_ENV}"
+          if [ "${GITHUB_REF}" = refs/heads/main ]; then rw=READ_WRITE; else rw=READ_ONLY; fi
+          echo "SCCACHE_GHA_RW_MODE=${rw}" >> "${GITHUB_ENV}"
           echo "RUSTC_WRAPPER=sccache" >> "${GITHUB_ENV}"
           echo "SCCACHE_IGNORE_SERVER_IO_ERROR=1" >> "${GITHUB_ENV}"
       - name: Install Linux system dependencies
@@ -725,7 +801,7 @@ jobs:
     # fail the build there while stable stays green (#6543).
     name: MSRV check (1.89)
     needs: changes
-    if: github.event_name == 'workflow_dispatch' && needs.changes.outputs.heavy == 'true'
+    if: needs.changes.outputs.heavy == 'true' && needs.changes.outputs.scoped != 'true'
     timeout-minutes: 45
     runs-on: ubuntu-latest
     steps:
@@ -746,7 +822,7 @@ jobs:
   workflow-rlm-cache:
     name: Workflow RLM cache
     needs: changes
-    if: needs.changes.outputs.workflow == 'true'
+    if: needs.changes.outputs.workflow == 'true' && needs.changes.outputs.scoped != 'true'
     timeout-minutes: 30
     runs-on: ubuntu-latest
     steps:
@@ -754,14 +830,15 @@ jobs:
       - uses: dtolnay/rust-toolchain@stable
       - uses: mozilla-actions/[email protected]
         id: sccache
-        # Main-only GitHub Actions cache backend; see the Safety gate job.
-        if: github.ref == 'refs/heads/main'
+        # Every run reads the GitHub Actions cache main wrote; only main writes (see Enable sccache).
         continue-on-error: true
       - name: Enable sccache
         if: steps.sccache.outcome == 'success'
         shell: bash
         run: |
           echo "SCCACHE_GHA_ENABLED=true" >> "${GITHUB_ENV}"
+          if [ "${GITHUB_REF}" = refs/heads/main ]; then rw=READ_WRITE; else rw=READ_ONLY; fi
+          echo "SCCACHE_GHA_RW_MODE=${rw}" >> "${GITHUB_ENV}"
           echo "RUSTC_WRAPPER=sccache" >> "${GITHUB_ENV}"
           echo "SCCACHE_IGNORE_SERVER_IO_ERROR=1" >> "${GITHUB_ENV}"
       - uses: Swatinem/rust-cache@v2
@@ -772,261 +849,64 @@ jobs:
         run: sh scripts/with-hermetic-test-home.sh cargo test -p codewhale-workflow --locked
 
   test:
-    name: Test
+    # Manual shared-process qualification only (workspace_test_mode =
+    # shared-process-twice): the GH6698 evidence of two libtest passes at one
+    # exact revision on one hosted Ubuntu runner. Every other run executes the
+    # suite through test-build, test-shard and test-suite below.
+    name: Shared-process tests
     needs: changes
-    if: github.event_name == 'workflow_dispatch'
-    # Contexts "Test (ubuntu-latest)" / "Test (macos-latest)" /
-    # "Test (windows-latest)" derive from job name + matrix.os and are
-    # independent of runs-on. For light changes the macOS/Windows legs only
-    # echo a skip line, so run them on ubuntu instead of queueing for scarce
-    # macOS/Windows runners. Pull requests run only the Linux leg here; the
-    # required "Test (windows-latest)" context comes from test-windows-pr.
-    # The ternary is safe: matrix.os is always a non-empty literal, so
-    # runs-on can never evaluate to empty.
-    # A cold GitHub-hosted Mac spent 77-80 minutes in Test on 2026-09-23
-    # (fork PRs #6431, #6417), too close to the old 90-minute limit.
-    # On 2026-10-04 a PR leg spent 111 minutes compiling with no cache and
-    # was cancelled at 120 with 12,356 tests passed and none failed. The
-    # limit stays a literal (release-workflows.test.js, #5496); 210 bounds
-    # cold compilation and the complete suite without disabling test deadlines.
+    if: github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice' && needs.changes.outputs.heavy == 'true' && needs.changes.outputs.scoped != 'true'
+    # A cold GitHub-hosted runner spent 111 minutes compiling with no cache on
+    # 2026-10-04. The limit stays a literal (release-workflows.test.js, #5496);
+    # 210 bounds cold compilation and both complete passes.
     timeout-minutes: 210
-    # macOS legs go to the self-hosted Mac ONLY when all three hold: the
-    # change is heavy, the event is trusted (not a fork PR), and the
-    # CW_SELF_HOSTED_MAC repo variable is 'true'. That variable is the kill
-    # switch: unset it and every leg falls back to GitHub-hosted runners
-    # immediately, with no commit — important because an offline
-    # self-hosted runner queues jobs forever, which is worse than a slow one.
-    runs-on: ${{ needs.changes.outputs.heavy != 'true' && 'ubuntu-latest' || (matrix.os == 'macos-latest' && needs.changes.outputs.trusted == 'true' && vars.CW_SELF_HOSTED_MAC == 'true' && fromJSON('["self-hosted","macOS","ARM64","codewhale-mac"]')) || matrix.os }}
-    strategy:
-      # A failure on one desktop platform must not erase evidence from the
-      # other one. We need both conclusions to diagnose and release safely.
-      fail-fast: false
-      matrix:
-        # Pull requests run the Linux workspace suite only; the macOS and
-        # Windows suites run in final validation. The explicit manual
-        # libtest qualification uses one hosted Ubuntu job, never a fork PR or
-        # self-hosted runner; normal dispatches keep all three legs.
-        os: ${{ fromJSON((github.event_name == 'pull_request' || (github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice')) && '["ubuntu-latest"]' || '["ubuntu-latest","macos-latest","windows-latest"]') }}
+    runs-on: ubuntu-latest
     steps:
-      - name: Skip tests for light change
-        if: needs.changes.outputs.heavy != 'true'
-        run: echo "No executable Rust changes detected; preserving required Test context."
       - uses: actions/checkout@v7
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
-      - name: Test Windows installer PATH helper
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'windows-latest'
-        shell: pwsh
-        run: ./scripts/installer/update-user-path.tests.ps1
-      - name: Install NSIS for Windows installer regression
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'windows-latest'
-        shell: pwsh
-        run: ./scripts/installer/install-nsis.ps1
-      - name: Test Windows installer PATH regression
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'windows-latest'
-        shell: pwsh
-        run: ./scripts/installer/installer-path-regression.tests.ps1 -AllowUserPathMutation
       - uses: dtolnay/rust-toolchain@stable
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
-      - name: Install macOS LLVM tools
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'macos-latest'
-        run: rustup component add llvm-tools --toolchain stable
-      - name: Omit macOS build debug metadata
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'macos-latest'
-        shell: bash
-        # Match the existing full-workspace CNB gate. Keep debug assertions and
-        # every test; only disposable build debug tables change on this lane.
-        run: printf '%s\n' 'CARGO_PROFILE_DEV_DEBUG=0' 'CARGO_PROFILE_TEST_DEBUG=0' >> "$GITHUB_ENV"
       - uses: mozilla-actions/[email protected]
         id: sccache
         continue-on-error: true
-        # Main-only GitHub Actions cache backend; see the Safety gate job.
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request') && github.ref == 'refs/heads/main'
+        # Every run reads the GitHub Actions cache main wrote; only main writes (see Enable sccache).
       - name: Enable sccache
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request') && steps.sccache.outcome == 'success'
+        if: steps.sccache.outcome == 'success'
         shell: bash
         run: |
           echo "SCCACHE_GHA_ENABLED=true" >> "${GITHUB_ENV}"
+          if [ "${GITHUB_REF}" = refs/heads/main ]; then rw=READ_WRITE; else rw=READ_ONLY; fi
+          echo "SCCACHE_GHA_RW_MODE=${rw}" >> "${GITHUB_ENV}"
           echo "RUSTC_WRAPPER=sccache" >> "${GITHUB_ENV}"
           echo "SCCACHE_IGNORE_SERVER_IO_ERROR=1" >> "${GITHUB_ENV}"
       - name: Install Linux system dependencies
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'ubuntu-latest' && (github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
         run: |
           bash scripts/ci-apt-install.sh libdbus-1-dev pkg-config bubblewrap apparmor-profiles
           sh scripts/prepare-linux-test-sandbox.sh
       - name: Ubuntu build headroom
         # Reuse the same fixed hosted-runner provisioning as Safety and Lint.
         # Exit 143 is a shutdown observation, not proof of OOM or timeout.
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'ubuntu-latest' && (github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
         shell: bash
         run: bash scripts/prepare-ubuntu-build-headroom.sh
       - uses: Swatinem/rust-cache@v2
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
         with:
           cache-bin: false
           save-if: ${{ github.ref == 'refs/heads/main' }}
-      - uses: taiki-e/install-action@nextest
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
-      - name: Check portable config policy boundary
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'ubuntu-latest' && (github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
-        shell: bash
-        run: |
-          python3 -B scripts/test_check_command_config_proof.py
-          sh scripts/check-portable-config.sh
-      - name: Check portable core navigation boundary
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'ubuntu-latest' && (github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
-        shell: bash
-        run: |
-          python3 -B scripts/test_check_command_core_navigation_proof.py
-          sh scripts/check-portable-core-navigation.sh
       - uses: actions/[email protected]
         # The extension-host integration tests spawn the real bundle under
         # Node >= 22.19; CODEWHALE_EXT_HOST_TESTS below makes a missing Node a
         # failure instead of a silent skip.
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
         with:
           node-version: 22
       - uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
         with:
           bun-version: 1.4.0
       - name: Build matching compiled-host input for required Native containment receipts
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
         shell: bash
-        run: |
-          set -euo pipefail
-          bun_binary="$(bun -p 'process.execPath')"
-          host_binary="$(node -p "require('path').join(process.env.RUNNER_TEMP, 'codewhale-extension-host' + (process.platform === 'win32' ? '.exe' : ''))")"
-          compiled_info="$(node crates/engine/extension-host/compile-host.mjs --bun "$bun_binary" --output "$host_binary")"
-          printf 'CODEWHALE_COMPILED_HOST_TEST_BINARY=%s\n' "$host_binary" >> "$GITHUB_ENV"
-          node - "$host_binary" "$bun_binary" "$compiled_info" >> "$GITHUB_ENV" <<'JS'
-          const fs = require('node:fs'), crypto = require('node:crypto');
-          const digest = (file) => crypto.createHash('sha256').update(fs.readFileSync(file)).digest('hex');
-          const info = JSON.parse(process.argv[4]);
-          if (!info.platform || !info.arch || info.platform !== info.compiler?.platform || info.arch !== info.compiler?.arch) throw new Error('image/compiler identity mismatch');
-          console.log('CODEWHALE_COMPILED_HOST_TEST_PLATFORM=' + info.platform);
-          console.log('CODEWHALE_COMPILED_HOST_TEST_ARCH=' + info.arch);
-          console.log('CODEWHALE_COMPILED_HOST_TEST_SHA256=' + digest(process.argv[2]));
-          console.log('CODEWHALE_COMPILED_HOST_TEST_BUN=' + process.argv[3]);
-          console.log('CODEWHALE_COMPILED_HOST_TEST_RUNTIME_SHA256=' + digest(process.argv[3]));
-          console.log('CODEWHALE_COMPILED_HOST_TEST_BUNDLE_SHA256=' + digest('crates/engine/extension-host/dist/codewhale-extension-host.mjs'));
-          JS
-          # The required hermetic full invocation executes the shared actual
-          # compiled Native secret/history/write scenario on each OS; Windows
-          # additionally requires Node/Bun/compiled LPAC owner/reparse probes.
-          # Missing requested compiled input or a failed sandbox is red.
-      - name: Hold this machine's build lock (self-hosted)
-        # The self-hosted Mac is also a developer machine: local agents build
-        # through scripts/dev-cargo.sh, which holds this lock, and two Cargo
-        # builds at once exhaust its memory. Opt-in: the runner's `.env` names
-        # the same file as CODEWHALE_BUILD_LOCK_FILE. The holder is a
-        # background process; the runner kills orphans when the job ends, so a
-        # cancelled or crashed job cannot leave the lock held.
-        # (`env.*` in `if:` cannot see the runner's `.env`, so the opt-in is
-        # checked in the script.)
-        if: needs.changes.outputs.heavy == 'true' && runner.environment == 'self-hosted'
-        shell: bash
-        run: |
-          if [ -z "${CODEWHALE_BUILD_LOCK_FILE:-}" ]; then
-            echo "CODEWHALE_BUILD_LOCK_FILE is not set on this runner; building without the machine lock."
-            exit 0
-          fi
-          hold="$RUNNER_TEMP/cw-build-lock.hold"
-          ready="$RUNNER_TEMP/cw-build-lock.ready"
-          log="$RUNNER_TEMP/cw-build-lock.log"
-          touch "$hold"
-          rm -f "$ready"
-          # The single quotes are deliberate: the inner sh expands $1/$2.
-          # shellcheck disable=SC2016
-          nohup python3 scripts/build-lock.py "$CODEWHALE_BUILD_LOCK_FILE" -- \
-            sh -c 'touch "$1"; while [ -e "$2" ]; do sleep 2; done' _ "$ready" "$hold" \
-            >"$log" 2>&1 &
-          shown=0
-          until [ -e "$ready" ]; do
-            if [ "$shown" -eq 0 ] && [ -s "$log" ]; then cat "$log"; shown=1; fi
-            sleep 2
-          done
-          cat "$log"
-          # Scripts inside this job already run under the lock.
-          echo "CODEWHALE_BUILD_LOCK_HELD=1" >> "$GITHUB_ENV"
+        run: bash scripts/ci/compiled-host-test-input.sh
       - name: Build canonical executable for acceptance tests
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
         shell: bash
         run: cargo build -p codewhale-cli --bin codewhale --all-features --locked
-        env:
-          RUSTC_WRAPPER: ${{ matrix.os != 'windows-latest' && env.RUSTC_WRAPPER || '' }}
-      - name: Windows Terminal clipboard-to-draft acceptance
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'windows-latest'
-        timeout-minutes: 5
-        shell: pwsh
-        run: node scripts/release/windows-terminal-clipboard.mjs target/debug/codewhale.exe "${{ runner.temp }}/windows-clipboard"
-      - name: Retain Windows Terminal clipboard-to-draft evidence
-        if: ${{ !cancelled() && matrix.os == 'windows-latest' && needs.changes.outputs.heavy == 'true' }}
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
-        with:
-          name: windows-clipboard-${{ github.sha }}-${{ github.run_attempt }}
-          path: ${{ runner.temp }}/windows-clipboard/
-          if-no-files-found: error
-          retention-days: 7
-      - name: Run tests
-        # Same test binaries as `cargo test`, run by cargo-nextest: one
-        # process per test, all runner cores busy, slow tests named instead
-        # of stalling the binary. `.config/nextest.toml` serializes the PTY
-        # binary and bounds the integration binary that spawns the real
-        # executable; retries are off, so a flake is a red run, not a hidden
-        # one. nextest does not run doctests — the next step keeps them.
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request') && !(github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice')
-        shell: bash
-        run: |
-          monitor=
-          if [ "$RUNNER_OS" = Linux ]; then
-            # Headroom trace: if the runner is terminated again, the last lines
-            # say whether memory or disk ran out.
-            ( while sleep 30; do
-                echo "[headroom] $(free -m | awk '/^Mem:/{print "mem used " $3 "/" $2 " MiB"} /^Swap:/{print "swap used " $3 "/" $2 " MiB"}' | paste -sd ' ') disk free $(df -h / | awk 'NR==2{print $4}')"
-              done ) &
-            monitor=$!
-          fi
-          status=0
-          sh scripts/with-hermetic-test-home.sh cargo nextest run --workspace --all-features --locked --profile ci || status=$?
-          if [ -n "$monitor" ]; then kill "$monitor" 2>/dev/null || true; fi
-          exit "$status"
-        env:
-          CARGO_BUILD_JOBS: '1'
-          QA_PTY_DIAGNOSTICS_DIR: ${{ runner.temp }}/pty-failures
-          CODEWHALE_EXT_HOST_TESTS: '1'
-          CODEWHALE_EXT_HOST_BUN_TESTS: '1'
-          # sccache 0.17 panics resolving its config directory under the
-          # isolated Windows home before Cargo can compile or run any test.
-          # Bypass only that optional cache; keep the full suite and isolation.
-          RUSTC_WRAPPER: ${{ matrix.os != 'windows-latest' && env.RUSTC_WRAPPER || '' }}
-          # Give test threads the stack the product gives itself. main.rs runs
-          # the owner thread and every tokio worker at
-          # CODEWHALE_MAIN_STACK_BYTES (32 MiB) because the engine and
-          # runtime-thread futures are genuinely deep. `#[tokio::test]` builds
-          # its own runtime and never sees that, so tests ran the same code on
-          # ~2 MiB (~1 MiB on Windows) — a configuration that never ships.
-          # That gap is what aborted the whole Windows test binary with
-          # STATUS_STACK_OVERFLOW in start_turn_accepts_dynamic_tools_and_
-          # environment_id, masking every other Windows result (78afd8d3d4
-          # Box::pin'd that one frame; the mismatch itself remained). std reads
-          # this for any thread spawned without an explicit size, which covers
-          # both libtest's per-test threads and tokio's workers. Test threads
-          # hold 16 MiB: the engine-only chains they run measured a ~2.5 MiB
-          # debug high-water, while the full ~16.5 MiB UI-loop chain that
-          # forced 32 MiB lives in spawned binaries, which size their own
-          # stacks explicitly and never read this variable.
-          RUST_MIN_STACK: '16777216'
-      - name: Retain sealed terminal failure diagnostics
-        if: failure()
-        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
-        with:
-          name: terminal-failures-${{ runner.os }}-${{ github.sha }}-${{ github.run_attempt }}
-          path: ${{ runner.temp }}/pty-failures/
-          include-hidden-files: true
-          if-no-files-found: ignore
-          retention-days: 7
       - name: Shared-process workspace qualification
-        if: github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice' && matrix.os == 'ubuntu-latest'
+        if: github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice'
         shell: bash
         env:
           CARGO_BUILD_JOBS: '1'
@@ -1128,83 +1008,193 @@ jobs:
           cat "$evidence/results.txt"
           exit "$failed"
       - name: Retain shared-process qualification evidence
-        if: always() && github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice' && matrix.os == 'ubuntu-latest'
+        if: always() && github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice'
         uses: actions/upload-artifact@v7
         with:
           name: shared-process-workspace-${{ github.sha }}-${{ github.run_attempt }}
           path: ${{ runner.temp }}/shared-process-workspace/
           if-no-files-found: error
           retention-days: 30
-      - name: Export exact compiled-image Native containment receipt
-        # JUnit is from the same full invocation: no second compile/test run.
-        # The official entire CI run must be green before release can consume
-        # this artifact. Missing/changed inputs or silent test skips are red.
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request') && !(github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice')
+
+  # One sharded pipeline per OS (ci-test-suite.yml). Shard counts follow the
+  # measured serial test time of each OS (2026-11-12, run 43216596942: Linux 19
+  # min, macOS 24 min, Windows 25 min of test execution, on top of 30 to 65
+  # minutes of single-job compilation), so a shard is five to eight minutes of
+  # tests plus setup. Hosted macOS is the scarce pool, so it gets three shards.
+  # The caller names ("Sharded tests (...)") never equal the required
+  # "Test (windows-latest)" context, which test-windows-pr still produces on
+  # every pull request.
+  test-ubuntu:
+    name: Sharded tests (ubuntu-latest)
+    needs: changes
+    if: >-
+      needs.changes.outputs.heavy == 'true' &&
+      needs.changes.outputs.scoped != 'true' &&
+      (github.event_name == 'pull_request' || github.event_name == 'workflow_dispatch') &&
+      !(github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice')
+    uses: ./.github/workflows/ci-test-suite.yml
+    with:
+      os: ubuntu-latest
+      shards: 4
+      shard_list: '[1,2,3,4]'
+      build_jobs: '2'
+      debug: line-tables-only
+
+  test-macos:
+    name: Sharded tests (macos-latest)
+    needs: changes
+    if: >-
+      needs.changes.outputs.heavy == 'true' &&
+      needs.changes.outputs.scoped != 'true' &&
+      (github.event_name == 'pull_request' || github.event_name == 'workflow_dispatch') &&
+      !(github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice')
+    uses: ./.github/workflows/ci-test-suite.yml
+    with:
+      os: macos-latest
+      shards: 3
+      shard_list: '[1,2,3]'
+      build_jobs: '1'
+      debug: '0'
+
+  test-windows:
+    name: Sharded tests (windows-latest)
+    needs: changes
+    if: >-
+      needs.changes.outputs.heavy == 'true' &&
+      needs.changes.outputs.scoped != 'true' &&
+      (github.event_name == 'pull_request' || github.event_name == 'workflow_dispatch') &&
+      !(github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice')
+    uses: ./.github/workflows/ci-test-suite.yml
+    with:
+      os: windows-latest
+      shards: 4
+      shard_list: '[1,2,3,4]'
+      build_jobs: '2'
+      debug: '0'
+
+  portable-boundaries:
+    name: Portable boundaries
+    needs: changes
+    if: >-
+      needs.changes.outputs.heavy == 'true' &&
+      needs.changes.outputs.scoped != 'true' &&
+      (github.event_name == 'pull_request' || github.event_name == 'workflow_dispatch')
+    timeout-minutes: 45
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v7
+      - uses: dtolnay/rust-toolchain@stable
+      - uses: mozilla-actions/[email protected]
+        id: sccache
+        continue-on-error: true
+        # Every run reads the GitHub Actions cache main wrote; only main writes (see Enable sccache).
+      - name: Enable sccache
+        if: steps.sccache.outcome == 'success'
+        shell: bash
+        run: |
+          echo "SCCACHE_GHA_ENABLED=true" >> "${GITHUB_ENV}"
+          if [ "${GITHUB_REF}" = refs/heads/main ]; then rw=READ_WRITE; else rw=READ_ONLY; fi
+          echo "SCCACHE_GHA_RW_MODE=${rw}" >> "${GITHUB_ENV}"
+          echo "RUSTC_WRAPPER=sccache" >> "${GITHUB_ENV}"
+          echo "SCCACHE_IGNORE_SERVER_IO_ERROR=1" >> "${GITHUB_ENV}"
+      - name: Install Linux system dependencies
+        run: |
+          bash scripts/ci-apt-install.sh libdbus-1-dev pkg-config bubblewrap apparmor-profiles
+          sh scripts/prepare-linux-test-sandbox.sh
+      - uses: Swatinem/rust-cache@v2
+        with:
+          cache-bin: false
+          save-if: ${{ github.ref == 'refs/heads/main' }}
+      - uses: taiki-e/install-action@nextest
+      - uses: actions/[email protected]
+        with:
+          node-version: 22
+      - uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2
+        with:
+          bun-version: 1.4.0
+      - name: Check portable config policy boundary
+        shell: bash
+        run: |
+          python3 -B scripts/test_check_command_config_proof.py
+          sh scripts/check-portable-config.sh
+      - name: Check portable core navigation boundary
+        shell: bash
+        run: |
+          python3 -B scripts/test_check_command_core_navigation_proof.py
+          sh scripts/check-portable-core-navigation.sh
+
+  scoped-test:
+    # workflow_dispatch with `crate` and/or `filter`: one package or one nextest
+    # filterset on the chosen OSes (default ubuntu-latest) instead of the full
+    # suite. `changes` skips every other job for such a run. The build uses the
+    # same flags as the full build; only `-p` and `-E` narrow what runs.
+    name: Scoped test (${{ matrix.os }})
+    needs: changes
+    if: needs.changes.outputs.scoped == 'true'
+    timeout-minutes: 90
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: ${{ fromJSON(needs.changes.outputs.scoped_os) }}
+    env:
+      CARGO_BUILD_JOBS: ${{ matrix.os == 'macos-latest' && '1' || '2' }}
+      CARGO_PROFILE_DEV_DEBUG: ${{ matrix.os == 'ubuntu-latest' && 'line-tables-only' || '0' }}
+      CARGO_PROFILE_TEST_DEBUG: ${{ matrix.os == 'ubuntu-latest' && 'line-tables-only' || '0' }}
+      CODEWHALE_EXT_HOST_TESTS: '1'
+      CODEWHALE_EXT_HOST_BUN_TESTS: '1'
+      RUST_MIN_STACK: '16777216'
+      SCOPE_CRATE: ${{ inputs.crate }}
+      SCOPE_FILTER: ${{ inputs.filter }}
+    steps:
+      - uses: actions/checkout@v7
+      - name: Point terminal failure diagnostics at the runner temp directory
+        shell: bash
+        run: echo "QA_PTY_DIAGNOSTICS_DIR=${RUNNER_TEMP}/pty-failures" >> "${GITHUB_ENV}"
+      - uses: dtolnay/rust-toolchain@stable
+      - name: Install Linux system dependencies
+        if: matrix.os == 'ubuntu-latest'
+        run: |
+          bash scripts/ci-apt-install.sh libdbus-1-dev pkg-config bubblewrap apparmor-profiles
+          sh scripts/prepare-linux-test-sandbox.sh
+      - uses: Swatinem/rust-cache@v2
+        with:
+          cache-bin: false
+          save-if: false
+      - uses: taiki-e/install-action@nextest
+      - uses: actions/[email protected]
+        with:
+          node-version: 22
+      - uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2
+        with:
+          bun-version: 1.4.0
+      - name: Build matching compiled-host input for required Native containment receipts
+        shell: bash
+        run: bash scripts/ci/compiled-host-test-input.sh
+      - name: Build canonical executable for acceptance tests
+        # Acceptance tests in the TUI, engine and CLI crates, and any run that
+        # names no crate, look for the codewhale binary next to the test
+        # binaries; a narrow run on another crate does not need it.
+        if: env.SCOPE_CRATE == '' || env.SCOPE_CRATE == 'codewhale-tui' || env.SCOPE_CRATE == 'codewhale-engine' || env.SCOPE_CRATE == 'codewhale-cli'
+        shell: bash
+        run: cargo build -p codewhale-cli --bin codewhale --all-features --locked
+      - name: Run the scoped tests
         shell: bash
         run: |
           set -euo pipefail
-          test "$(git rev-parse HEAD)" = "$GITHUB_SHA"
-          git diff --exit-code
-          git diff --cached --exit-code
-          python3 scripts/compiled-host-native-receipt.py \
-            --junit target/nextest/ci/junit.xml \
-            --compiled-image "$CODEWHALE_COMPILED_HOST_TEST_BINARY" \
-            --bundle crates/engine/extension-host/dist/codewhale-extension-host.mjs \
-            --bun "$CODEWHALE_COMPILED_HOST_TEST_BUN" --source-sha "$GITHUB_SHA" \
-            --image-platform "$CODEWHALE_COMPILED_HOST_TEST_PLATFORM" --image-arch "$CODEWHALE_COMPILED_HOST_TEST_ARCH" \
-            --host-sha256 "$CODEWHALE_COMPILED_HOST_TEST_SHA256" \
-            --runtime-sha256 "$CODEWHALE_COMPILED_HOST_TEST_RUNTIME_SHA256" \
-            --bundle-sha256 "$CODEWHALE_COMPILED_HOST_TEST_BUNDLE_SHA256" \
-            --output "$RUNNER_TEMP/native-compiled-host"
-      - name: Retain exact compiled image and Native receipt
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request') && !(github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice')
+          args=(--all-features --locked --profile ci --no-tests=fail)
+          if [ -n "${SCOPE_CRATE}" ]; then args+=(-p "${SCOPE_CRATE}"); else args+=(--workspace); fi
+          if [ -n "${SCOPE_FILTER}" ]; then args+=(-E "${SCOPE_FILTER}"); fi
+          sh scripts/with-hermetic-test-home.sh cargo nextest run "${args[@]}"
+      - name: Retain sealed terminal failure diagnostics
+        if: failure()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
         with:
-          name: native-compiled-host-${{ runner.os }}-${{ runner.arch }}
-          path: ${{ runner.temp }}/native-compiled-host/
-          if-no-files-found: error
-          retention-days: 30
-      - name: Run doctests
-        # The shared-process command already includes every workspace doctest.
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request') && !(github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice')
-        shell: bash
-        run: sh scripts/with-hermetic-test-home.sh cargo test --workspace --all-features --locked --doc
-        env:
-          RUSTC_WRAPPER: ${{ matrix.os != 'windows-latest' && env.RUSTC_WRAPPER || '' }}
-          RUST_MIN_STACK: '16777216'
-      # The Ubuntu lint lane validates non-RSS backlog fields. Run the same
-      # source-bound measurement on macOS so loss or growth of RSS evidence
-      # fails closed instead of becoming an unsupported-field skip.
-      # Only on the self-hosted Mac: there it reuses the warm build. Every
-      # hosted Mac leg (fork PRs, or any PR with CW_SELF_HOSTED_MAC off) runs
-      # it in the separate `macos-budget` job with its own timeout, because
-      # on a cold hosted Mac it pushed Test past 90 minutes.
-      - name: Check persistence-backlog RSS budget
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'macos-latest' && needs.changes.outputs.trusted == 'true' && vars.CW_SELF_HOSTED_MAC == 'true'
-        run: python3 scripts/check-persistence-backlog-budget.py
-      - name: Lockfile drift guard
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
-        run: git diff --exit-code -- Cargo.lock
-      - name: Run Offline Eval Harness
-        # The eval harness is OS-independent prompt/composition checking;
-        # running it once (on the faster macOS leg, warm from the test build)
-        # instead of once per desktop OS keeps the coverage while taking
-        # ~2min off the Windows critical path. Self-hosted Mac only; hosted
-        # Mac legs run it in `macos-budget` (see the RSS step above).
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'macos-latest' && needs.changes.outputs.trusted == 'true' && vars.CW_SELF_HOSTED_MAC == 'true'
-        run: cargo run -p codewhale-cli --bin codewhale --all-features -- eval
-      - name: sccache stats
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request') && steps.sccache.outcome == 'success'
-        continue-on-error: true
-        shell: bash
-        run: sccache --show-stats
-      - name: Linux test location (CNB)
-        if: needs.changes.outputs.heavy == 'true' && matrix.os == 'ubuntu-latest' && github.event_name != 'workflow_dispatch' && github.event_name != 'pull_request'
-        run: echo "Linux workspace tests run on CNB for non-PR release/main pushes; pull requests run directly on Ubuntu."
-      - name: Release this machine's build lock (self-hosted)
-        if: always() && runner.environment == 'self-hosted'
-        shell: bash
-        run: rm -f "$RUNNER_TEMP/cw-build-lock.hold"
+          name: terminal-failures-scoped-${{ runner.os }}-${{ github.sha }}-${{ github.run_attempt }}
+          path: ${{ runner.temp }}/pty-failures/
+          include-hidden-files: true
+          if-no-files-found: ignore
+          retention-days: 7
 
   test-windows-pr:
     name: Test (windows-latest)
@@ -1239,7 +1229,7 @@ jobs:
             echo "## Test (windows-latest) on a pull request"
             echo
             echo "Compiled every workspace target for Windows with warnings denied and ran the installer PATH helper tests."
-            echo "The Windows and macOS test suites run in final validation, before a release, not on pull requests."
+            echo "The full Windows test suite runs beside this check as Sharded tests (windows-latest); it is not a required check."
           } >> "$GITHUB_STEP_SUMMARY"
 
   macos-budget:
@@ -1252,15 +1242,15 @@ jobs:
     # macOS leg runs on the self-hosted Mac (trusted event and
     # CW_SELF_HOSTED_MAC == 'true'), it runs these two steps on the warm
     # build instead. The name keeps "(fork PR)" so check contexts stay stable.
-    # Hosted macOS allows only five concurrent jobs per account, and this job
-    # doubled every pull request's claim on them, queueing the required Test
-    # (macos-latest) leg for hours. So pull requests skip it and every push to
-    # main still runs it; a regression shows up on the merge that caused it.
+    # Hosted macOS allows only five concurrent jobs per account, and the macOS
+    # test build and shards already take most of them, so pull requests skip
+    # this job and every dispatch (main-landing issues one per push to main)
+    # still runs it; a regression shows up on the merge that caused it. The
+    # sharded macOS suite never runs on the self-hosted Mac, so these gates no
+    # longer depend on CW_SELF_HOSTED_MAC.
     name: macOS budget and eval (fork PR)
     needs: changes
-    # The explicit Ubuntu-only qualification has no warm Test macOS leg, so
-    # retain these gates on the hosted fallback even when self-hosting is on.
-    if: needs.changes.outputs.heavy == 'true' && github.event_name != 'pull_request' && ((github.event_name == 'workflow_dispatch' && inputs.workspace_test_mode == 'shared-process-twice') || !(needs.changes.outputs.trusted == 'true' && vars.CW_SELF_HOSTED_MAC == 'true'))
+    if: needs.changes.outputs.heavy == 'true' && needs.changes.outputs.scoped != 'true' && github.event_name != 'pull_request'
     # Cold hosted builds consumed 73 minutes before eval could start (F run
     # 37497498829). Match Test's compile allowance; RSS/eval limits stay intact.
     timeout-minutes: 210
@@ -1282,9 +1272,10 @@ jobs:
   npm-wrapper-smoke:
     name: npm wrapper smoke
     needs: changes
-    if: github.event_name == 'workflow_dispatch'
-    # Same ternary rationale as the Test job: light legs only echo, so keep
-    # them off macOS/Windows runners. On pull_request the matrix is
+    if: needs.changes.outputs.scoped != 'true' && (github.event_name == 'pull_request' || github.event_name == 'workflow_dispatch')
+    # Light legs only echo, so keep them off macOS/Windows runners. The
+    # ternary is safe: matrix.os is always a non-empty literal. On
+    # pull_request the matrix is
     # ubuntu-only, so the required "npm wrapper smoke (ubuntu-latest)"
     # context is unaffected. Heavy pull requests execute the Ubuntu smoke
     # here; their branches may not be mirrored to CNB.
@@ -1310,13 +1301,15 @@ jobs:
       - uses: mozilla-actions/[email protected]
         id: sccache
         continue-on-error: true
-        # Main-only GitHub Actions cache backend; see the Safety gate job.
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request') && github.ref == 'refs/heads/main'
+        # Every run reads the GitHub Actions cache main wrote; only main writes (see Enable sccache).
+        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request')
       - name: Enable sccache
-        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request') && steps.sccache.outcome == 'success'
+        if: needs.changes.outputs.heavy == 'true' && (matrix.os != 'ubuntu-latest' || github.event_name == 'workflow_dispatch' || github.event_name == 'pull_request') && steps.sccache.outcome == 'success' && matrix.os != 'windows-latest'
         shell: bash
         run: |
           echo "SCCACHE_GHA_ENABLED=true" >> "${GITHUB_ENV}"
+          if [ "${GITHUB_REF}" = refs/heads/main ]; then rw=READ_WRITE; else rw=READ_ONLY; fi
+          echo "SCCACHE_GHA_RW_MODE=${rw}" >> "${GITHUB_ENV}"
           echo "RUSTC_WRAPPER=sccache" >> "${GITHUB_ENV}"
           echo "SCCACHE_IGNORE_SERVER_IO_ERROR=1" >> "${GITHUB_ENV}"
       - uses: actions/[email protected]
@@ -1358,8 +1351,14 @@ jobs:
   mobile-smoke:
     name: Mobile runtime smoke
     needs: changes
-    # Manual (workflow_dispatch) only; the release candidate builds android-arm64.
-    if: github.event_name == 'workflow_dispatch' && needs.changes.outputs.heavy == 'true'
+    # Not a required PR context. Pull requests run it only when the mobile
+    # runtime surface changed (see the `mobile` filter above); every dispatch,
+    # which main-landing issues for each push to main, runs it unconditionally
+    # as the pre-release safety net.
+    if: >-
+      needs.changes.outputs.scoped != 'true' &&
+      needs.changes.outputs.heavy == 'true' &&
+      (github.event_name != 'pull_request' || needs.changes.outputs.mobile == 'true')
     timeout-minutes: 30
     runs-on: ubuntu-latest
     steps:
@@ -1367,14 +1366,15 @@ jobs:
       - uses: dtolnay/rust-toolchain@stable
       - uses: mozilla-actions/[email protected]
         id: sccache
-        # Main-only GitHub Actions cache backend; see the Safety gate job.
-        if: github.ref == 'refs/heads/main'
+        # Every run reads the GitHub Actions cache main wrote; only main writes (see Enable sccache).
         continue-on-error: true
       - name: Enable sccache
         if: steps.sccache.outcome == 'success'
         shell: bash
         run: |
           echo "SCCACHE_GHA_ENABLED=true" >> "${GITHUB_ENV}"
+          if [ "${GITHUB_REF}" = refs/heads/main ]; then rw=READ_WRITE; else rw=READ_ONLY; fi
+          echo "SCCACHE_GHA_RW_MODE=${rw}" >> "${GITHUB_ENV}"
           echo "RUSTC_WRAPPER=sccache" >> "${GITHUB_ENV}"
           echo "SCCACHE_IGNORE_SERVER_IO_ERROR=1" >> "${GITHUB_ENV}"
       - name: Install Linux system dependencies
@@ -1401,7 +1401,7 @@ jobs:
   actionlint:
     name: Workflow lint
     needs: changes
-    if: needs.changes.outputs.actions == 'true'
+    if: needs.changes.outputs.actions == 'true' && needs.changes.outputs.scoped != 'true'
     timeout-minutes: 15
     runs-on: ubuntu-latest
     steps:
@@ -1418,7 +1418,8 @@ jobs:
   # Check documentation builds without warnings
   docs:
     name: Documentation
-    if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
+    needs: changes
+    if: needs.changes.outputs.scoped != 'true' && (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch')
     timeout-minutes: 60
     runs-on: ubuntu-latest
     steps:
@@ -1439,17 +1440,17 @@ jobs:
   cargo-deny:
     name: cargo-deny
     needs: changes
-    if: github.event_name == 'workflow_dispatch'
+    if: github.event_name == 'workflow_dispatch' && needs.changes.outputs.scoped != 'true'
     uses: ./.github/workflows/cargo-deny.yml
 
   security-audit:
     name: Security audit
     needs: changes
-    if: github.event_name == 'workflow_dispatch'
+    if: github.event_name == 'workflow_dispatch' && needs.changes.outputs.scoped != 'true'
     uses: ./.github/workflows/security-audit.yml
 
   ohos:
     name: OpenHarmony
     needs: changes
-    if: github.event_name == 'workflow_dispatch'
+    if: github.event_name == 'workflow_dispatch' && needs.changes.outputs.scoped != 'true'
     uses: ./.github/workflows/ohos.yml

.github/workflows/nightly.yml

@@ -73,7 +73,7 @@ jobs:
         id: sccache
         continue-on-error: true
       - name: Enable sccache
-        if: steps.sccache.outcome == 'success'
+        if: steps.sccache.outcome == 'success' && runner.os != 'Windows'
         shell: bash
         run: |
           {

.github/workflows/release-artifacts.yml

@@ -178,7 +178,7 @@ jobs:
         id: sccache
         continue-on-error: true
       - name: Enable sccache
-        if: steps.sccache.outcome == 'success'
+        if: steps.sccache.outcome == 'success' && runner.os != 'Windows'
         shell: bash
         run: |
           {

.github/workflows/release-candidate.yml

@@ -68,8 +68,6 @@ jobs:
 
           echo "sha=${actual}" >> "${GITHUB_OUTPUT}"
           echo "version=${workspace_version}" >> "${GITHUB_OUTPUT}"
-      - name: Check the first-party catalog snapshot is current
-        run: ./scripts/refresh-first-party-marketplace.sh --check
       - name: Check version and OHOS release contracts
         # --require-dated-release, as auto-tag.yml and release.yml run it:
         # release.yml only accepts an RC receipt for the exact tag SHA, so an

CHANGELOG.md

@@ -11,8 +11,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **Non-ASCII file names:** `/diff`, the `@git` and `@diff` mentions and the Runtime API workspace diff now show paths with Chinese, emoji and other non-ASCII characters as written instead of as quoted octal escapes, and the turn context no longer lists the workspace root as an empty "Active paths" entry. Thanks [@luochen211](https://github.com/luochen211) (PR 7317).
 - **Session names after a crash, and keep-forever snapshot retention:** a session name saved right after a crash that cut off the previous session-index line is no longer lost, because the new entry now starts on its own line; and a very large `[snapshots] max_age_days` (such as `9223372036854775807` for "keep forever") no longer wraps around and deletes every undo snapshot at session start. Thanks [@luochen211](https://github.com/luochen211) (PR 7316).
 - **Blank environment variables:** exporting a variable empty (for example `CODEWHALE_MODEL=` from a compose `${VAR:-}` template) no longer hides its legacy alias. `DEEPSEEK_MODEL`, `DEEPSEEK_BASE_URL`, `KIMI_BASE_URL`, `DEEPSEEK_FORCE_HTTP1`, `DEEPSEEK_STREAM_OPEN_TIMEOUT_SECS` and the other provider alias variables now apply as they do when the primary is unset. Thanks [@luochen211](https://github.com/luochen211) (PR 7315).
-- **Qwen Code as an external agent:** Codewhale now finds an installed Qwen Code CLI (`qwen`, 0.6.1 or newer) and can run it as a session over ACP with `qwen --acp`. `GET /v1/agents/external` lists it, an older install is reported as needing `npm i -g @qwen-code/qwen-code@latest`, and sign-in stays inside Qwen Code (`/auth`). Thanks [@MannXo](https://github.com/MannXo) (PR 7293).
-- **MCP and language servers that send requests mid-call:** a tool call no longer fails with "neither a result nor an error" when an MCP server sends a `ping` or other request that reuses the call's id while the call is running, and a language server's `workspace/configuration` or similar request no longer ends a pending LSP request. Codewhale now answers `ping`, answers other server requests with "Method not found" (it never advertised them), keeps waiting for the real reply, and a refused answer over Streamable HTTP no longer drops the connection to legacy SSE. Thanks [@luochen211](https://github.com/luochen211) (PR 7319).
+- **Qwen Code as an external agent:** Codewhale now finds an installed Qwen Code CLI (`qwen`, 0.6.1 or newer) and can run it as a session over ACP with `qwen --acp`. `GET /v1/agents/external` lists it, an older install is reported as needing `npm i -g @qwen-code/qwen-code@latest`, and sign-in stays inside Qwen Code (`/auth`). Thanks [@MannXo](https://github.com/MannXo) (#7293).
 - **README and contributor credit:** the repository front page now opens the way the website does: the line "The model can be anyone's. The agent should be yours.", one sentence on what Codewhale is, the install command and one picture, with reference detail moved into the linked guides; the 18 translated READMEs follow the same structure and say they are machine-drafted. The contributor wall and the website now count the same 433 people (407 GitHub accounts and the commit authors whose email is not linked to an account), where the website had counted accounts only. `DEPENDENCY_SECURITY.md` moved into `docs/dependency-maintenance.md`, and `.github/SUPPORT.md` says where to ask for help.
 - **Website Plugins page:** The website's Plugins page now describes what the released version does, and a labelled section says what 0.10.3 adds: one Plugins place, Standard and Creator mode, and the Permissions control.
 - **Plugin docs and plugin-creator:** Docs now say everything you add is a plugin. The bundled plugin-creator skill no longer tells an agent never to trust or enable a plugin (in Full Access it may, for one it wrote and installed from its project), points code and screen plugins at cordis-plugin-development, and an untouched copy from 0.10.2 now upgrades. The Chinese extension and Runtime API docs now say editor ACP sessions get plugin tools and configured MCP servers.

CONTRIBUTING.md

@@ -165,7 +165,7 @@ before starting a change; preserve any local work when updating your branch.
 
 - Write tests for new functionality
 - Run the tests near your change (see [Fast local loop](#fast-local-loop));
-  your pull request runs the Linux workspace suite and a Windows build for
+  your pull request runs the full test suite on Linux, macOS and Windows for
   you (see [What happens after you open a pull request](#what-happens-after-you-open-a-pull-request))
 - Colocate unit tests beside the code they cover (standard Rust `#[cfg(test)]`
   modules), and add integration tests under the owning crate's `tests/`
@@ -174,8 +174,8 @@ before starting a change; preserve any local work when updating your branch.
 
 ### Pre-push verification
 
-These are the Rust commands your pull request runs on Linux (the `Lint` and
-`Test (ubuntu-latest)` checks). You do not need all of them before every
+These are the Rust commands your pull request runs on Linux (the `Lint` check
+and the Linux shards of `Sharded tests (ubuntu-latest)`). You do not need all of them before every
 push: run `cargo fmt`, then check and test the crates you touched (see
 [Fast local loop](#fast-local-loop)). Run the full set locally when a change
 spans many crates, or let the pull request run it for you:
@@ -215,7 +215,7 @@ cargo install cargo-nextest --locked      # once
 scripts/dev-test.sh tui
 scripts/dev-cargo.sh nextest run --workspace --all-features --locked
 
-# 4. The Linux suite your PR runs:
+# 4. The suite your PR runs on Linux (hosted CI runs it as parallel shards):
 cargo test --workspace --all-features --locked
 ```
 
@@ -340,6 +340,9 @@ maintainer to approve the first run for a brand-new GitHub account.
 | --- | --- | --- |
 | `Lint` (required) | `cargo fmt --check`, clippy on every target, and the fast drift checks (locales, READMEs, provider registry, dead-code and blocking-call budgets) | Run `cargo fmt --all` and the clippy command in [Pre-push verification](#pre-push-verification); the log names the file and the check |
 | `Test (windows-latest)` (required) | Compiles every workspace target for Windows with warnings denied, and the installer PATH helper tests | Usually a `cfg(windows)` path or an import only Windows sees; the log has the compiler error |
+| `Sharded tests (ubuntu-latest)`, `(macos-latest)`, `(windows-latest)` | The full workspace test suite, built once per OS and run as parallel shards of one nextest archive, plus the doctests; each OS ends with a `Suite check` that fails if the shards did not run every test exactly once | Open the failing `Shard k/N` job; the log names the test. The same test reproduces locally with `cargo nextest run -E 'test(name)'` |
+| `Safety gate`, `MSRV check (1.89)`, `npm wrapper smoke`, `Portable boundaries` | The hermetic safety and authorization tests, a build on the minimum Rust version, the npm wrapper install smoke and the portable-core boundary checks | The log names the failing test or the compiler error |
+| `Mobile runtime smoke` | The `codewhale serve --mobile` HTTP and event-stream smoke, only when the mobile runtime surface changed | The log shows the failing request |
 | `Version drift` | Version files, the `feat:` release-note rule, release helper tests and contributor credit | The credit step only reports: a maintainer adds your credit when the PR merges |
 
 Smaller checks cover the VS Code extension, integrations, the extension host
@@ -347,11 +350,15 @@ and plugin conversion, plus workflow lint when you change `.github/workflows`.
 A documentation-only change skips the Rust builds but still reports the
 required checks, so it can merge.
 
-Your pull request does not run the test suites, the safety gate, MSRV, the npm
-wrapper smoke, the slow runtime-contract and persistence-backlog measurements,
-the dependency audits or the OpenHarmony build. Maintainers run those once per
-release, on the exact commit being released. If that run finds a problem your
-change introduced, we fix it on `main`; we will not ask you to chase it.
+Only `Lint` and `Test (windows-latest)` are required. Every other job above is
+informational for a contributor's pull request: a red one tells you and the
+maintainers something, and never blocks the merge. Your pull request does not
+run the slow runtime-contract and persistence-backlog measurements, the macOS
+budget and eval job, the dependency audits or the OpenHarmony build; those run
+on `main` and for each release, on the exact commit. If one of them finds a
+problem your change introduced, we fix it on `main`; we will not ask you to
+chase it. A newer push to your branch cancels the older run of the same pull
+request, so the runners always work on your latest commit.
 
 A red check that your change did not cause is not your job to fix. We rerun it
 or fix it on `main`, and we never close a PR because CI is slow or flaky. Your

Cargo.lock

@@ -1179,7 +1179,6 @@ dependencies = [
  "agent-client-protocol",
  "agent-client-protocol-schema",
  "anyhow",
- "arboard",
  "async-stream",
  "async-trait",
  "axum",

crates/engine/Cargo.toml

@@ -219,9 +219,6 @@ libc = "0.2"
 windows-sys = { version = "0.61.2", features = ["Wdk_Storage_FileSystem", "Win32_System_IO", "Win32_Foundation", "Win32_Security", "Win32_Security_Authorization", "Win32_Security_Isolation", "Win32_Storage_FileSystem", "Win32_System_LibraryLoader", "Win32_System_Pipes", "Win32_System_Threading", "Win32_System_SystemServices", "Win32_System_WindowsProgramming"] }
 
 
-[target.'cfg(any(target_os = "macos", target_os = "windows", all(target_os = "linux", not(target_env = "ohos"))))'.dependencies]
-arboard = { version = "3.4", features = ["wayland-data-control"] }
-
 [target.'cfg(not(target_env = "ohos"))'.dependencies]
 portable-pty = "0.9"
 

crates/engine/src/core/authority/auto_review.rs

@@ -2152,10 +2152,7 @@ fn git_push_refspec_is_protected(refspec: &str) -> bool {
     // Tag-like names (`v1`, `v0.9.1`): git resolves branch-vs-tag on the
     // server, so treat them as publishes.
     let mut chars = lower.chars();
-    if chars.next() == Some('v') && chars.next().is_some_and(|ch| ch.is_ascii_digit()) {
-        return true;
-    }
-    false
+    chars.next() == Some('v') && chars.next().is_some_and(|ch| ch.is_ascii_digit())
 }
 
 fn git_tag_tokens_are_publish_like(tokens: &[&str]) -> bool {

crates/engine/src/extension_host/dsh_workspace.rs

@@ -203,10 +203,13 @@ impl Drop for TestOverlay {
 
 fn overlay_granted() -> bool {
     #[cfg(test)]
-    if TEST_OVERLAY.with(std::cell::Cell::get) {
-        return true;
+    {
+        TEST_OVERLAY.with(std::cell::Cell::get)
+    }
+    #[cfg(not(test))]
+    {
+        false
     }
-    false
 }
 
 fn caller(call: &BrokerCall, tool_name: &'static str) -> Result<ToolContext, BrokerRefusal> {

crates/engine/src/fleet/manager.rs

@@ -1748,11 +1748,7 @@ impl FleetManager {
                 return Err(start_error);
             }
         };
-        if !started {
-            return Ok(false);
-        }
-
-        Ok(true)
+        Ok(started)
     }
 
     fn start_leased_workers(

crates/engine/src/mcp.rs

@@ -1669,10 +1669,6 @@ pub trait McpTransport: Send + Sync {
 
     async fn recv(&mut self) -> Result<Vec<u8>>;
 
-    async fn send_reply(&mut self, msg: Vec<u8>) -> Result<()> {
-        self.send(msg).await
-    }
-
     /// Record the protocol revision negotiated at `initialize`. Only the
     /// Streamable HTTP transport uses it (the `MCP-Protocol-Version` header on
     /// subsequent requests); stdio and legacy SSE have no header channel, so
@@ -3129,30 +3125,6 @@ impl McpConnection {
         result
     }
 
-    async fn answer_server_request(&mut self, reply: &serde_json::Value) {
-        let result = match serde_json::to_vec(reply) {
-            Ok(bytes) => {
-                let cancel_token = self.cancel_token.clone();
-                tokio::select! {
-                    biased;
-                    _ = cancel_token.cancelled() => {
-                        Err(anyhow::anyhow!("MCP connection was cancelled"))
-                    }
-                    result = self.transport.send_reply(bytes) => result,
-                }
-            }
-            Err(err) => Err(err.into()),
-        };
-        if let Err(err) = result {
-            tracing::debug!(
-                target: "mcp",
-                server = %self.name,
-                error = %err,
-                "could not answer an MCP server request"
-            );
-        }
-    }
-
     /// Handshake and discovery receive: each frame wait is bounded by the read
     /// knob, and a server that stays silent past it is treated as dead.
     async fn recv(&mut self, expected_id: String) -> Result<serde_json::Value> {
@@ -3230,22 +3202,6 @@ impl McpConnection {
             {
                 self.catalog_stale = true;
             }
-            if let Some(method) = value.get("method") {
-                if let Some(id) = value.get("id") {
-                    let method = method.as_str().unwrap_or_default();
-                    let reply = if method == "ping" {
-                        serde_json::json!({"jsonrpc": "2.0", "id": id, "result": {}})
-                    } else {
-                        serde_json::json!({
-                            "jsonrpc": "2.0",
-                            "id": id,
-                            "error": {"code": -32601, "message": format!("Method not found: {method}")}
-                        })
-                    };
-                    self.answer_server_request(&reply).await;
-                }
-                continue;
-            }
             if response_id_matches(value.get("id"), &expected_id) {
                 // Marks the connection stale so it is rebuilt, but this is a
                 // reply to the request, so it never qualifies for a replay.

crates/engine/src/mcp/http.rs

@@ -177,21 +177,6 @@ impl McpTransport for HttpTransport {
         }
     }
 
-    async fn send_reply(&mut self, msg: Vec<u8>) -> Result<()> {
-        match &mut self.mode {
-            HttpTransportMode::Streamable(transport) => {
-                transport.send(msg).await.map_err(|err| match err {
-                    StreamableSendError::Incompatible(detail)
-                    | StreamableSendError::StaleSession(detail) => anyhow::anyhow!(
-                        "MCP Streamable HTTP refused a reply to a server request: {detail}"
-                    ),
-                    StreamableSendError::Other(err) => err,
-                })
-            }
-            HttpTransportMode::Sse(transport) => transport.send(msg).await,
-        }
-    }
-
     async fn recv(&mut self) -> Result<Vec<u8>> {
         match &mut self.mode {
             HttpTransportMode::Streamable(transport) => transport.recv().await,

crates/engine/src/mcp/tests.rs

@@ -3314,80 +3314,6 @@ async fn call_method_skips_notifications_and_unmatched_responses() {
     assert_eq!(sent[0]["method"], "tools/call");
 }
 
-#[tokio::test]
-async fn call_method_answers_server_requests_instead_of_taking_them_as_the_reply() {
-    let sent = Arc::new(Mutex::new(Vec::new()));
-    let transport = ScriptedValueTransport {
-        sent: Arc::clone(&sent),
-        responses: VecDeque::from([
-            json_frame(serde_json::json!({"jsonrpc": "2.0", "id": 1, "method": "ping"})),
-            json_frame(serde_json::json!({"jsonrpc": "2.0", "id": "1", "method": "roots/list"})),
-            json_frame(serde_json::json!({"jsonrpc": "2.0", "id": 1, "result": {"ok": true}})),
-        ]),
-    };
-    let mut conn = test_connection(Box::new(transport));
-
-    let result = conn
-        .call_method("tools/call", serde_json::json!({"name": "echo"}), 1)
-        .await
-        .unwrap();
-
-    assert_eq!(result, serde_json::json!({"ok": true}));
-    let sent = sent.lock().unwrap();
-    assert_eq!(sent.len(), 3);
-    assert_eq!(
-        sent[1],
-        serde_json::json!({"jsonrpc": "2.0", "id": 1, "result": {}})
-    );
-    assert_eq!(sent[2]["id"], "1");
-    assert_eq!(sent[2]["error"]["code"], -32601);
-}
-
-struct ReplyRefusingTransport {
-    sent: Arc<Mutex<Vec<serde_json::Value>>>,
-    responses: VecDeque<Vec<u8>>,
-}
-
-#[async_trait::async_trait]
-impl McpTransport for ReplyRefusingTransport {
-    async fn send(&mut self, msg: Vec<u8>) -> Result<()> {
-        let msg: serde_json::Value = serde_json::from_slice(&msg)?;
-        anyhow::ensure!(msg.get("method").is_some(), "status=405 Method Not Allowed");
-        self.sent.lock().unwrap().push(msg);
-        Ok(())
-    }
-
-    async fn recv(&mut self) -> Result<Vec<u8>> {
-        self.responses
-            .pop_front()
-            .context("scripted transport exhausted")
-    }
-}
-
-#[tokio::test]
-async fn call_method_keeps_its_reply_when_answering_a_server_request_fails() {
-    let sent = Arc::new(Mutex::new(Vec::new()));
-    let transport = ReplyRefusingTransport {
-        sent: Arc::clone(&sent),
-        responses: VecDeque::from([
-            json_frame(serde_json::json!({"jsonrpc": "2.0", "id": 1, "method": "ping"})),
-            json_frame(serde_json::json!({"jsonrpc": "2.0", "id": 1, "result": {"ok": true}})),
-        ]),
-    };
-    let mut conn = test_connection(Box::new(transport));
-
-    let result = conn
-        .call_method("tools/call", serde_json::json!({"name": "echo"}), 1)
-        .await
-        .unwrap();
-
-    assert_eq!(result, serde_json::json!({"ok": true}));
-    assert_eq!(conn.state, ConnectionState::Ready);
-    let sent = sent.lock().unwrap();
-    assert_eq!(sent.len(), 1);
-    assert_eq!(sent[0]["method"], "tools/call");
-}
-
 #[tokio::test]
 async fn call_method_invalid_json_includes_server_output_preview() {
     let sent = Arc::new(Mutex::new(Vec::new()));
@@ -6844,60 +6770,6 @@ async fn custom_headers_applied_to_get_preflight() {
     );
 }
 
-#[tokio::test]
-async fn refused_reply_to_a_server_request_does_not_fall_back_to_sse() {
-    use tokio::io::{AsyncReadExt, AsyncWriteExt};
-    use tokio::net::TcpListener;
-
-    let _env = crate::test_support::lock_test_env();
-    let _lock = lock_mcp_loopback_tests().await;
-    let _no_proxy = crate::test_support::EnvVarGuard::set("NO_PROXY", "*,127.0.0.1");
-    let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
-    let addr = listener.local_addr().unwrap();
-    let requests = Arc::new(Mutex::new(Vec::new()));
-    let requests_srv = Arc::clone(&requests);
-
-    let server = tokio::spawn(async move {
-        loop {
-            let (mut socket, _) = listener.accept().await.unwrap();
-            let mut buf = [0u8; 4096];
-            let n = socket.read(&mut buf).await.unwrap();
-            let request_line = String::from_utf8_lossy(&buf[..n])
-                .lines()
-                .next()
-                .unwrap_or_default()
-                .to_string();
-            requests_srv.lock().unwrap().push(request_line);
-            socket
-                .write_all(
-                    b"HTTP/1.1 405 Method Not Allowed\r\nConnection: close\r\nContent-Length: 0\r\n\r\n",
-                )
-                .await
-                .unwrap();
-        }
-    });
-
-    let url = format!("http://{addr}/mcp");
-    let mut transport = HttpTransport::new(
-        test_mcp_http_client(&url),
-        url,
-        tokio_util::sync::CancellationToken::new(),
-        Duration::from_secs(2),
-    );
-
-    let result = transport
-        .send_reply(json_frame(
-            serde_json::json!({"jsonrpc": "2.0", "id": 1, "result": {}}),
-        ))
-        .await;
-
-    server.abort();
-    assert!(result.is_err(), "a refused reply reports the refusal");
-    let requests = requests.lock().unwrap();
-    assert_eq!(requests.len(), 1, "requests seen: {requests:?}");
-    assert!(requests[0].starts_with("POST "), "{requests:?}");
-}
-
 // === add_runtime_server_config conflict tests ===
 
 #[test]

crates/engine/src/tools/subagent/coord/ledger.rs

@@ -133,14 +133,9 @@ impl WriteScopeClaim {
                 return true;
             }
         }
-        if self
-            .contracts
+        self.contracts
             .iter()
             .any(|contract| other.contracts.iter().any(|other| other == contract))
-        {
-            return true;
-        }
-        false
     }
 
     #[must_use]

crates/lsp/src/client.rs

@@ -255,12 +255,11 @@ impl StdioLspTransport {
         let child = Arc::new(AsyncMutex::new(Some(child)));
         let dispatcher_child = child.clone();
         let dispatcher_pending = pending.clone();
-        let dispatcher_outbound = tx_outbound.clone();
         let dispatcher_task = spawn_supervised(
             "lsp-dispatcher",
             std::panic::Location::caller(),
             async move {
-                dispatcher_task(rx_inbound, tx_diag, dispatcher_outbound, dispatcher_pending).await;
+                dispatcher_task(rx_inbound, tx_diag, dispatcher_pending).await;
                 // EOF, malformed frames, or an overflowing diagnostics queue
                 // terminate the producer too, so its pipes cannot stay stuck.
                 if let Some(mut child) = dispatcher_child.lock().await.take() {
@@ -537,19 +536,15 @@ impl LspTransport for StdioLspTransport {
 
 /// Send a JSON value as one Content-Length-framed JSON-RPC message.
 async fn send_message(tx: &mpsc::Sender<Vec<u8>>, value: &Value) -> Result<()> {
-    tx.send(frame_message(value)?)
-        .await
-        .map_err(|_| anyhow!("LSP outbound channel closed"))?;
-    Ok(())
-}
-
-fn frame_message(value: &Value) -> Result<Vec<u8>> {
     let body = serde_json::to_vec(value).context("serialize LSP message")?;
     let header = format!("Content-Length: {}\r\n\r\n", body.len());
     let mut frame = Vec::with_capacity(header.len() + body.len());
     frame.extend_from_slice(header.as_bytes());
     frame.extend_from_slice(&body);
-    Ok(frame)
+    tx.send(frame)
+        .await
+        .map_err(|_| anyhow!("LSP outbound channel closed"))?;
+    Ok(())
 }
 
 /// Background task that drains the outbound queue and writes each frame to
@@ -643,7 +638,6 @@ fn parse_header(buf: &[u8]) -> Result<Option<(usize, usize)>> {
 async fn dispatcher_task(
     mut rx: mpsc::Receiver<Value>,
     tx_diag: mpsc::Sender<DiagnosticMessage>,
-    tx_outbound: mpsc::Sender<Vec<u8>>,
     pending: Arc<AsyncMutex<HashMap<i64, oneshot::Sender<Value>>>>,
 ) {
     while let Some(value) = rx.recv().await {
@@ -660,19 +654,6 @@ async fn dispatcher_task(
             }
             continue;
         }
-        if let Some(method) = method {
-            if let Some(id) = value.get("id") {
-                let reply = json!({
-                    "jsonrpc": "2.0",
-                    "id": id,
-                    "error": {"code": -32601, "message": format!("unsupported method {method}")},
-                });
-                if let Ok(frame) = frame_message(&reply) {
-                    let _ = tx_outbound.try_send(frame);
-                }
-            }
-            continue;
-        }
         // Replies have an `id` and a `result` or `error`.
         if let Some(id) = value.get("id").and_then(|v| v.as_i64()) {
             let mut map = pending.lock().await;
@@ -799,31 +780,6 @@ pub(super) mod tests {
         assert_eq!(diags[0].message, "missing semicolon");
     }
 
-    #[tokio::test]
-    async fn server_request_sharing_a_pending_id_is_answered_not_taken_as_the_reply() {
-        let (tx_in, rx_in) = mpsc::channel(4);
-        let (tx_out, mut rx_out) = mpsc::channel::<Vec<u8>>(4);
-        let (tx_diag, _rx_diag) = mpsc::channel(4);
-        let pending = Arc::new(AsyncMutex::new(HashMap::new()));
-        let (reply_tx, reply_rx) = oneshot::channel();
-        pending.lock().await.insert(2, reply_tx);
-        let dispatcher = tokio::spawn(dispatcher_task(rx_in, tx_diag, tx_out, pending));
-        tx_in
-            .send(json!({"jsonrpc":"2.0","id":2,"method":"workspace/configuration","params":{"items":[]}}))
-            .await
-            .unwrap();
-        tx_in
-            .send(json!({"jsonrpc":"2.0","id":2,"result":"definition"}))
-            .await
-            .unwrap();
-        assert_eq!(reply_rx.await.unwrap()["result"], "definition");
-        let answer = next_document(&mut rx_out).await;
-        assert_eq!(answer["id"], 2);
-        assert_eq!(answer["error"]["code"], -32601);
-        drop(tx_in);
-        dispatcher.await.unwrap();
-    }
-
     #[test]
     fn round_trips_uri_path() {
         let path = fixture_path("example/foo.rs");
@@ -1353,12 +1309,11 @@ while True:
             let (mut producer, reader) = tokio::io::duplex(MAX_LSP_HEADER_BYTES * 2);
             let (tx, rx) = mpsc::channel(8);
             let (diag, _diagnostics) = mpsc::channel(8);
-            let (outbound, _outbound_rx) = mpsc::channel(8);
             let pending = Arc::new(AsyncMutex::new(HashMap::new()));
             let (reply, receiver) = oneshot::channel();
             pending.lock().await.insert(1, reply);
             let read_task = tokio::spawn(reader_task(reader, tx));
-            let dispatch = tokio::spawn(dispatcher_task(rx, diag, outbound, pending.clone()));
+            let dispatch = tokio::spawn(dispatcher_task(rx, diag, pending.clone()));
             producer.write_all(&frame).await.unwrap();
             assert!(
                 timeout(Duration::from_secs(1), receiver)

crates/plugins/src/marketplace/store.rs

@@ -105,9 +105,7 @@ impl MarketplaceStore {
         };
         // Every surface consumes this same projection. Browsing is offline
         // and read-only; installation and updates use the reviewed installer.
-        // A user's local catalog or removal always takes precedence; only a
-        // saved copy of an older embedded snapshot is superseded.
-        replace_superseded_embedded_catalog(&mut state)?;
+        // A user's local catalog or removal always takes precedence.
         if !state.first_party_removed && !state.catalogs.contains_key("codewhale") {
             state
                 .catalogs
@@ -237,7 +235,6 @@ impl MarketplaceStore {
             .write()
             .map_err(|e| format!("failed to lock marketplace state for update: {e}"))?;
         let mut next = self.load_unlocked()?;
-        replace_superseded_embedded_catalog(&mut next)?;
         let result = mutate(&mut next)?;
         save_state_with_hardener(&self.path, &next, harden_plugin_state_file)?;
         Ok(result)
@@ -316,57 +313,17 @@ fn collapse_same_source_catalogs(mut state: MarketplaceState) -> MarketplaceStat
     state
 }
 
-#[derive(Deserialize)]
-struct EmbeddedSnapshot {
-    repository: String,
-    revision: String,
-    catalog: serde_json::Value,
-}
-
-impl EmbeddedSnapshot {
-    fn embedded() -> Result<Self, String> {
-        serde_json::from_str(include_str!("../../assets/first-party-marketplace.json"))
-            .map_err(|error| format!("invalid bundled marketplace: {error}"))
-    }
-
-    fn source(&self) -> String {
-        format!("{}/tree/{}", self.repository, self.revision)
-    }
-
-    fn supersedes(&self, source_path: &str) -> bool {
-        source_path
-            .strip_prefix(&format!("{}/tree/", self.repository))
-            .is_some_and(|revision| {
-                revision != self.revision
-                    && revision.len() == 40
-                    && revision.bytes().all(|byte| byte.is_ascii_hexdigit())
-            })
-    }
-}
-
-/// A saved `codewhale` catalog whose source is an older revision of the
-/// embedded snapshot is replaced by the embedded one. A catalog added from any
-/// other source (the desktop's bundled file, a person's own document or URL)
-/// is never touched.
-fn replace_superseded_embedded_catalog(state: &mut MarketplaceState) -> Result<(), String> {
-    if state.first_party_removed {
-        return Ok(());
-    }
-    let Some(stored) = state.catalogs.get("codewhale") else {
-        return Ok(());
-    };
-    let snapshot = EmbeddedSnapshot::embedded()?;
-    if snapshot.supersedes(&stored.source_path) {
-        state
-            .catalogs
-            .insert("codewhale".into(), first_party_catalog()?);
-    }
-    Ok(())
-}
-
 fn first_party_catalog() -> Result<StoredMarketplaceCatalog, String> {
-    let snapshot = EmbeddedSnapshot::embedded()?;
-    let source = snapshot.source();
+    #[derive(Deserialize)]
+    struct Snapshot {
+        repository: String,
+        revision: String,
+        catalog: serde_json::Value,
+    }
+    let snapshot: Snapshot =
+        serde_json::from_str(include_str!("../../assets/first-party-marketplace.json"))
+            .map_err(|error| format!("invalid bundled marketplace: {error}"))?;
+    let source = format!("{}/tree/{}", snapshot.repository, snapshot.revision);
     let mut catalog = super::parsers::parse_catalog(super::parsers::MarketplaceDocument {
         catalog_id: MarketplaceCatalogId::new("codewhale"),
         format: super::types::MarketplaceFormat::Codewhale,
@@ -414,8 +371,10 @@ mod tests {
                 panic!("uninstallable candidate")
             };
             if candidate.name == "computer-use" {
-                let (source, sha256) = snapshot_entry_pin("computer-use");
-                assert_eq!(spec, format!("{source}#sha256={sha256}"));
+                assert_eq!(
+                    spec,
+                    "https://github.com/codewhale-hq/codewhale-cu-plugin/releases/download/v0.16.0/computer-use.tar.gz#sha256=11a4f5c97c32717fa5648d10c080aa0db48a4798ccdf0d17112d1ae5b9133ae4"
+                );
             } else {
                 assert!(spec.starts_with(
                     "https://codeload.github.com/codewhale-hq/codewhale-plugin-marketplace/"
@@ -442,122 +401,6 @@ mod tests {
         );
     }
 
-    fn snapshot_plugins() -> Vec<serde_json::Value> {
-        let snapshot: serde_json::Value =
-            serde_json::from_str(include_str!("../../assets/first-party-marketplace.json"))
-                .unwrap();
-        snapshot["catalog"]["plugins"].as_array().unwrap().clone()
-    }
-
-    fn snapshot_entry_pin(name: &str) -> (String, String) {
-        let entry = snapshot_plugins()
-            .into_iter()
-            .find(|entry| entry["name"] == name)
-            .unwrap();
-        (
-            entry["source"].as_str().unwrap().to_string(),
-            entry["sha256"].as_str().unwrap().to_string(),
-        )
-    }
-
-    #[test]
-    fn every_remote_snapshot_entry_carries_a_sha256() {
-        let snapshot = EmbeddedSnapshot::embedded().unwrap();
-        let pinned_bundle = format!(
-            "https://codeload.github.com/codewhale-hq/codewhale-plugin-marketplace/tar.gz/{}#",
-            snapshot.revision
-        );
-        let mut remote = 0;
-        for entry in snapshot_plugins() {
-            let source = entry["source"].as_str().unwrap();
-            assert!(
-                source.starts_with("https://"),
-                "{source} is not a pinned remote source"
-            );
-            if source.starts_with(&pinned_bundle) {
-                continue;
-            }
-            remote += 1;
-            let sha256 = entry["sha256"].as_str().unwrap_or_default();
-            assert!(
-                sha256.len() == 64
-                    && sha256
-                        .bytes()
-                        .all(|b| matches!(b, b'0'..=b'9' | b'a'..=b'f')),
-                "{} has no sha256",
-                entry["name"]
-            );
-            assert!(
-                !source.contains("/latest/"),
-                "{} follows a floating release",
-                entry["name"]
-            );
-        }
-        assert!(remote >= 1, "the computer-use release asset is remote");
-    }
-
-    fn write_state(store: &MarketplaceStore, source_path: &str) {
-        let mut catalog = first_party_catalog().unwrap();
-        catalog.source_path = source_path.to_string();
-        catalog.catalog.candidates.clear();
-        let mut state = MarketplaceState::default();
-        state.catalogs.insert("codewhale".into(), catalog);
-        ensure_private_plugin_state_directory(store.path().parent().unwrap()).unwrap();
-        fs::write(store.path(), serde_json::to_string(&state).unwrap()).unwrap();
-    }
-
-    #[test]
-    fn a_saved_copy_of_an_older_embedded_snapshot_is_replaced() {
-        let root = tempfile::tempdir().unwrap();
-        let store = MarketplaceStore::open(Some(&root.path().join("plugins/state.json"))).unwrap();
-        let snapshot = EmbeddedSnapshot::embedded().unwrap();
-        let older = format!("{}/tree/{}", snapshot.repository, "0".repeat(40));
-        write_state(&store, &older);
-
-        let loaded = store.load().unwrap();
-        let kept = loaded.get("codewhale").unwrap();
-        assert_eq!(kept.source_path, snapshot.source());
-        assert!(kept.catalog.total_candidates() > 0);
-
-        store
-            .add(
-                &MarketplaceCatalogId::new("codewhale"),
-                first_party_catalog().unwrap(),
-            )
-            .unwrap();
-        let persisted: MarketplaceState =
-            serde_json::from_str(&fs::read_to_string(store.path()).unwrap()).unwrap();
-        assert_eq!(
-            persisted.get("codewhale").unwrap().source_path,
-            snapshot.source()
-        );
-    }
-
-    #[test]
-    fn desktop_bundled_and_person_added_catalogs_are_kept() {
-        let snapshot = EmbeddedSnapshot::embedded().unwrap();
-        let bundled_file = Path::new("resources")
-            .join("marketplace")
-            .join("marketplace.json")
-            .display()
-            .to_string();
-        let moving_branch = format!("{}/tree/main", snapshot.repository);
-        for source in [
-            bundled_file.as_str(),
-            "https://example.com/team/marketplace.json",
-            moving_branch.as_str(),
-        ] {
-            let root = tempfile::tempdir().unwrap();
-            let store =
-                MarketplaceStore::open(Some(&root.path().join("plugins/state.json"))).unwrap();
-            write_state(&store, source);
-            let loaded = store.load().unwrap();
-            let kept = loaded.get("codewhale").unwrap();
-            assert_eq!(kept.source_path, source);
-            assert_eq!(kept.catalog.total_candidates(), 0);
-        }
-    }
-
     #[test]
     fn malformed_store_is_not_hidden_by_bundled_catalog() {
         let root = tempfile::tempdir().unwrap();

crates/tui/CHANGELOG.md

@@ -11,8 +11,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - **Non-ASCII file names:** `/diff`, the `@git` and `@diff` mentions and the Runtime API workspace diff now show paths with Chinese, emoji and other non-ASCII characters as written instead of as quoted octal escapes, and the turn context no longer lists the workspace root as an empty "Active paths" entry. Thanks [@luochen211](https://github.com/luochen211) (PR 7317).
 - **Session names after a crash, and keep-forever snapshot retention:** a session name saved right after a crash that cut off the previous session-index line is no longer lost, because the new entry now starts on its own line; and a very large `[snapshots] max_age_days` (such as `9223372036854775807` for "keep forever") no longer wraps around and deletes every undo snapshot at session start. Thanks [@luochen211](https://github.com/luochen211) (PR 7316).
 - **Blank environment variables:** exporting a variable empty (for example `CODEWHALE_MODEL=` from a compose `${VAR:-}` template) no longer hides its legacy alias. `DEEPSEEK_MODEL`, `DEEPSEEK_BASE_URL`, `KIMI_BASE_URL`, `DEEPSEEK_FORCE_HTTP1`, `DEEPSEEK_STREAM_OPEN_TIMEOUT_SECS` and the other provider alias variables now apply as they do when the primary is unset. Thanks [@luochen211](https://github.com/luochen211) (PR 7315).
-- **Qwen Code as an external agent:** Codewhale now finds an installed Qwen Code CLI (`qwen`, 0.6.1 or newer) and can run it as a session over ACP with `qwen --acp`. `GET /v1/agents/external` lists it, an older install is reported as needing `npm i -g @qwen-code/qwen-code@latest`, and sign-in stays inside Qwen Code (`/auth`). Thanks [@MannXo](https://github.com/MannXo) (PR 7293).
-- **MCP and language servers that send requests mid-call:** a tool call no longer fails with "neither a result nor an error" when an MCP server sends a `ping` or other request that reuses the call's id while the call is running, and a language server's `workspace/configuration` or similar request no longer ends a pending LSP request. Codewhale now answers `ping`, answers other server requests with "Method not found" (it never advertised them), keeps waiting for the real reply, and a refused answer over Streamable HTTP no longer drops the connection to legacy SSE. Thanks [@luochen211](https://github.com/luochen211) (PR 7319).
+- **Qwen Code as an external agent:** Codewhale now finds an installed Qwen Code CLI (`qwen`, 0.6.1 or newer) and can run it as a session over ACP with `qwen --acp`. `GET /v1/agents/external` lists it, an older install is reported as needing `npm i -g @qwen-code/qwen-code@latest`, and sign-in stays inside Qwen Code (`/auth`). Thanks [@MannXo](https://github.com/MannXo) (#7293).
 - **README and contributor credit:** the repository front page now opens the way the website does: the line "The model can be anyone's. The agent should be yours.", one sentence on what Codewhale is, the install command and one picture, with reference detail moved into the linked guides; the 18 translated READMEs follow the same structure and say they are machine-drafted. The contributor wall and the website now count the same 433 people (407 GitHub accounts and the commit authors whose email is not linked to an account), where the website had counted accounts only. `DEPENDENCY_SECURITY.md` moved into `docs/dependency-maintenance.md`, and `.github/SUPPORT.md` says where to ask for help.
 - **Website Plugins page:** The website's Plugins page now describes what the released version does, and a labelled section says what 0.10.3 adds: one Plugins place, Standard and Creator mode, and the Permissions control.
 - **Plugin docs and plugin-creator:** Docs now say everything you add is a plugin. The bundled plugin-creator skill no longer tells an agent never to trust or enable a plugin (in Full Access it may, for one it wrote and installed from its project), points code and screen plugins at cordis-plugin-development, and an untouched copy from 0.10.2 now upgrades. The Chinese extension and Runtime API docs now say editor ACP sessions get plugin tools and configured MCP servers.

crates/tui/src/tui/context_menu.rs

@@ -284,8 +284,8 @@ impl ContextMenuView {
         if start > 0 {
             rows.push(MenuRow::MoreAbove(start));
         }
-        for idx in start..end {
-            if idx > start && self.entries[idx].section_start {
+        for (idx, entry) in self.entries.iter().enumerate().take(end).skip(start) {
+            if idx > start && entry.section_start {
                 rows.push(MenuRow::Divider);
             }
             rows.push(MenuRow::Entry(idx));

crates/tui/src/tui/transcript/tests.rs

@@ -1441,8 +1441,8 @@ fn prose_cells_fill_full_width_on_ultrawide_by_default() {
     let mut cache = TranscriptViewCache::new();
     cache.ensure_filtered(&refs, &revisions, 220, options, &HashMap::new(), None, None);
 
-    for idx in 0..3 {
-        let width = max_line_width(&cache.per_cell[idx].lines);
+    for (idx, cell) in cache.per_cell.iter().enumerate().take(3) {
+        let width = max_line_width(&cell.lines);
         assert!(
             width > RETIRED_RAIL_MEASURE,
             "prose cell {idx} wrapped to {width} columns — still on the retired \
@@ -1495,8 +1495,8 @@ fn transcript_prose_measure_caps_prose_but_not_tools() {
     let mut cache = TranscriptViewCache::new();
     cache.ensure_filtered(&refs, &revisions, 220, options, &HashMap::new(), None, None);
 
-    for idx in 0..3 {
-        let width = max_line_width(&cache.per_cell[idx].lines);
+    for (idx, cell) in cache.per_cell.iter().enumerate().take(3) {
+        let width = max_line_width(&cell.lines);
         assert!(
             width > 105,
             "prose cell {idx} wrapped to {width} columns — still on the retired \

crates/tui/src/tui/ui/terminal.rs

@@ -1110,7 +1110,7 @@ mod screen_mode_tests {
                         viewport: ratatui::Viewport::Inline(3),
                     },
                 )
-                .map_err(|err| io::Error::other(err.to_string()))
+                .map_err(|err| match err {})
             },
         )
         .expect("a successful probe must switch");

docs/CONTRIBUTORS.md

@@ -44,6 +44,7 @@ changelog or release notes, and this contributor record.
 
 **Contributions**
 
+- **[MannXo](https://github.com/MannXo)** — added Qwen Code to the external agent registry so an installed Qwen Code CLI runs as a session over ACP ([#7293](https://github.com/codewhale-hq/Codewhale/pull/7293)).
 - **[SamhandsomeLee](https://github.com/SamhandsomeLee)** — drove Responses request, URL and stream behavior from one checked dialect profile table ([#7261](https://github.com/codewhale-hq/Codewhale/pull/7261)).
 - **[gaord](https://github.com/gaord)** — ran the read-only `/system`, `/tokens` and `/cost` builtins over the Runtime API through one gated terminal-host catalog ([#7277](https://github.com/codewhale-hq/Codewhale/pull/7277)).
 

docs/RELEASE_RUNBOOK.md

@@ -244,7 +244,30 @@ The manual `ci.yml` path verifies that the dispatch resolved to
 `expected_sha`, disables light-change shortcuts, and forces the heavy Rust,
 workflow, mobile, Actions, Linux, macOS, Windows, npm-wrapper, and documentation
 gates. A mismatch fails before those gates start; it never silently tests a
-different head.
+different head. The Linux, macOS and Windows suites run as `Sharded tests (<os>)`
+(`ci-test-suite.yml`): one test build per OS, parallel nextest shards of that
+archive, and a `Suite check` job that fails unless the shards together ran
+exactly the tests the archive lists; read that job, not only the shard jobs.
+
+`ci.yml` also takes optional scoped-run inputs for a quick answer from an agent
+or a maintainer. With `crate` (one workspace package, for example
+`codewhale-tui`) and/or `filter` (a nextest filterset such as
+`test(auto_review)`) set, only the `Scoped test` job runs, on the operating
+systems named by `os` (default `ubuntu-latest`; a comma-separated list or
+`all`), and the full suite and every other gate are skipped. `expected_sha`
+stays required and is still verified:
+
+```bash
+gh workflow run ci.yml --ref ci/my-task \
+  -f expected_sha="$(git rev-parse HEAD)" \
+  -f crate=codewhale-tui -f filter='test(auto_review)'
+```
+
+A scoped run proves a slice, never a release; release evidence is a full
+dispatch with the scoped inputs empty. A newer dispatch on the same non-main
+branch, and a newer push to a pull request, cancels the older run of the same
+workflow; runs on `main` are keyed by SHA and never cancel each other, and a
+scoped run never cancels a full one.
 
 Verify beforehand that clean local `HEAD`, `main` and `origin/main` agree.
 Record both run URLs, resolved SHAs and required job results; inspect executed

scripts/ci/compiled-host-test-input.sh

@@ -0,0 +1,18 @@
+#!/usr/bin/env bash
+set -euo pipefail
+bun_binary="$(bun -p 'process.execPath')"
+host_binary="$(node -p "require('path').join(process.env.RUNNER_TEMP, 'codewhale-extension-host' + (process.platform === 'win32' ? '.exe' : ''))")"
+compiled_info="$(node crates/engine/extension-host/compile-host.mjs --bun "$bun_binary" --output "$host_binary")"
+printf 'CODEWHALE_COMPILED_HOST_TEST_BINARY=%s\n' "$host_binary" >> "$GITHUB_ENV"
+node - "$host_binary" "$bun_binary" "$compiled_info" >> "$GITHUB_ENV" <<'JS'
+const fs = require('node:fs'), crypto = require('node:crypto');
+const digest = (file) => crypto.createHash('sha256').update(fs.readFileSync(file)).digest('hex');
+const info = JSON.parse(process.argv[4]);
+if (!info.platform || !info.arch || info.platform !== info.compiler?.platform || info.arch !== info.compiler?.arch) throw new Error('image/compiler identity mismatch');
+console.log('CODEWHALE_COMPILED_HOST_TEST_PLATFORM=' + info.platform);
+console.log('CODEWHALE_COMPILED_HOST_TEST_ARCH=' + info.arch);
+console.log('CODEWHALE_COMPILED_HOST_TEST_SHA256=' + digest(process.argv[2]));
+console.log('CODEWHALE_COMPILED_HOST_TEST_BUN=' + process.argv[3]);
+console.log('CODEWHALE_COMPILED_HOST_TEST_RUNTIME_SHA256=' + digest(process.argv[3]));
+console.log('CODEWHALE_COMPILED_HOST_TEST_BUNDLE_SHA256=' + digest('crates/engine/extension-host/dist/codewhale-extension-host.mjs'));
+JS

scripts/ci/nextest-shard-count.py

@@ -0,0 +1,83 @@
+#!/usr/bin/env python3
+"""Count nextest tests so a sharded run can prove it ran every test once.
+
+  list LISTING_JSON         tests the archive selects (nextest list --message-format json)
+  junit JUNIT_XML           tests one nextest invocation ran
+  check DIRECTORY           per OS, the shard files must cover the expected total exactly
+
+The sharded CI suite splits one archive across jobs. A shard that is skipped,
+runs a different filter, or double-counts a partition changes the sum, so the
+suite gate fails instead of reporting a smaller run as a pass.
+"""
+from __future__ import annotations
+
+import json
+import sys
+import xml.etree.ElementTree as ET
+from pathlib import Path
+
+
+def listed(path: str) -> int:
+    suites = json.loads(Path(path).read_text(encoding="utf-8"))["rust-suites"]
+    return sum(
+        1
+        for suite in suites.values()
+        for case in suite["testcases"].values()
+        if case["filter-match"]["status"] == "matches"
+    )
+
+
+def ran(path: str) -> int:
+    return int(ET.parse(path).getroot().get("tests", "0"))
+
+
+def check(directory: str) -> int:
+    files = sorted(Path(directory).rglob("*.json"))
+    expected = {}
+    shards: dict[str, dict[int, int]] = {}
+    for file in files:
+        record = json.loads(file.read_text(encoding="utf-8"))
+        if file.name.startswith("expected-"):
+            expected[record["os"]] = record
+        elif file.name.startswith("shard-"):
+            slot = shards.setdefault(record["os"], {})
+            if record["shard"] in slot:
+                print(f"{record['os']}: shard {record['shard']} reported twice")
+                return 1
+            slot[record["shard"]] = record["tests"]
+    if not expected:
+        print("no expected-<os>.json found")
+        return 1
+    failed = False
+    for os_name, record in sorted(expected.items()):
+        got = shards.get(os_name, {})
+        want = set(range(1, record["shards"] + 1))
+        total = sum(got.values())
+        print(f"{os_name}: archive lists {record['tests']}, shards ran {total} {dict(sorted(got.items()))}")
+        if set(got) != want:
+            print(f"{os_name}: expected shards {sorted(want)}, found {sorted(got)}")
+            failed = True
+        if total != record["tests"]:
+            print(f"{os_name}: shard total {total} differs from the archive's {record['tests']}")
+            failed = True
+        if any(count <= 0 for count in got.values()):
+            print(f"{os_name}: a shard ran no tests")
+            failed = True
+    return 1 if failed else 0
+
+
+def main(argv: list[str]) -> int:
+    if len(argv) == 3 and argv[1] == "list":
+        print(listed(argv[2]))
+        return 0
+    if len(argv) == 3 and argv[1] == "junit":
+        print(ran(argv[2]))
+        return 0
+    if len(argv) == 3 and argv[1] == "check":
+        return check(argv[2])
+    print(__doc__, file=sys.stderr)
+    return 2
+
+
+if __name__ == "__main__":
+    raise SystemExit(main(sys.argv))

scripts/refresh-first-party-marketplace.sh

@@ -1,87 +0,0 @@
-#!/usr/bin/env bash
-# Refresh or check the first-party catalog snapshot embedded in the Engine
-# (crates/plugins/assets/first-party-marketplace.json) against the head of the
-# public marketplace repository.
-#
-#   scripts/refresh-first-party-marketplace.sh            rewrite the snapshot from the marketplace head
-#   scripts/refresh-first-party-marketplace.sh --check    fail when the snapshot is older than the head
-#   --marketplace DIR                                     use this checkout instead of cloning the head
-#
-# Nothing is installed, trusted or executed. "Older" means the files the
-# catalog installs from (marketplace.json, plugins/, skills/) differ between the
-# snapshot's revision and the head; unrelated commits do not make it stale.
-set -euo pipefail
-
-root="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-snapshot="${root}/crates/plugins/assets/first-party-marketplace.json"
-repo_url="https://github.com/codewhale-hq/codewhale-plugin-marketplace"
-mode=refresh
-checkout=""
-
-while [[ $# -gt 0 ]]; do
-  case "$1" in
-    --check) mode=check; shift ;;
-    --marketplace)
-      [[ $# -ge 2 ]] || { echo "--marketplace needs a directory" >&2; exit 2; }
-      checkout="$2"; shift 2 ;;
-    -h|--help) sed -n '2,12p' "${BASH_SOURCE[0]}"; exit 0 ;;
-    *) echo "unknown argument: $1" >&2; exit 2 ;;
-  esac
-done
-
-if [[ -z "${checkout}" ]]; then
-  tmp="$(mktemp -d)"
-  trap 'rm -rf "${tmp}"' EXIT
-  git clone --quiet --filter=blob:none --no-checkout "${repo_url}" "${tmp}/marketplace"
-  checkout="${tmp}/marketplace"
-fi
-
-head_rev="$(git -C "${checkout}" rev-parse HEAD)"
-
-if [[ "${mode}" == refresh ]]; then
-  validate="$(cat <<'PY'
-import json, re, sys
-catalog = json.load(sys.stdin)
-problems = []
-for plugin in catalog["plugins"]:
-    name, source = plugin["name"], plugin["source"]
-    if source.startswith("path:") or re.fullmatch(r"github:[A-Za-z0-9_.-]+/[A-Za-z0-9_.-]+", source):
-        continue
-    if "/latest/" in source:
-        problems.append(f"{name}: follows a floating release ({source})")
-    if not re.fullmatch(r"[0-9a-f]{64}", plugin.get("sha256", "")):
-        problems.append(f"{name}: remote source has no sha256")
-    for field in ("version", "platforms"):
-        if not plugin.get(field):
-            problems.append(f"{name}: remote source has no {field}")
-if problems:
-    sys.exit("marketplace head is not publishable as a snapshot:\n  " + "\n  ".join(problems))
-PY
-)"
-  git -C "${checkout}" show HEAD:marketplace.json | python3 -c "${validate}"
-  python3 "${root}/scripts/sync-marketplace.py" --marketplace "${checkout}"
-  exit 0
-fi
-
-snapshot_rev="$(python3 -c 'import json,sys; print(json.load(open(sys.argv[1]))["revision"])' "${snapshot}")"
-if [[ "${snapshot_rev}" == "${head_rev}" ]]; then
-  echo "First-party catalog snapshot is at the marketplace head ${head_rev}."
-  exit 0
-fi
-if ! git -C "${checkout}" cat-file -e "${snapshot_rev}^{commit}" 2>/dev/null; then
-  echo "::error::Snapshot revision ${snapshot_rev} is not in the marketplace history; run scripts/refresh-first-party-marketplace.sh." >&2
-  exit 1
-fi
-tree_ids() {
-  local rev="$1" path
-  for path in marketplace.json plugins skills; do
-    git -C "${checkout}" rev-parse "${rev}:${path}" 2>/dev/null || echo "missing:${path}"
-  done
-}
-if [[ "$(tree_ids "${snapshot_rev}")" == "$(tree_ids "${head_rev}")" ]]; then
-  echo "First-party catalog snapshot (${snapshot_rev}) is current: nothing it installs from changed up to ${head_rev}."
-  exit 0
-fi
-echo "::error::The first-party catalog snapshot (${snapshot_rev}) is older than the marketplace head (${head_rev}). Run scripts/refresh-first-party-marketplace.sh, review the diff and commit it." >&2
-git -C "${checkout}" diff --stat "${snapshot_rev}" "${head_rev}" -- marketplace.json plugins skills >&2 || true
-exit 1

scripts/test_ci_migration_wiring.py

@@ -273,7 +273,7 @@ class CiWiringTests(unittest.TestCase):
         block = ci[start:next_step]
         self.assertIn(
             "sh scripts/with-hermetic-test-home.sh cargo nextest run "
-            "-p codewhale-tui -p codewhale-runtime --lib --locked --no-tests=fail "
+            "-p codewhale-engine -p codewhale-tui -p codewhale-runtime --lib --locked --no-tests=fail "
             "-E 'test(auto_review) | test(authority) | test(sandbox)'", block
         )
         self.assertIn(
@@ -290,27 +290,27 @@ class CiWiringTests(unittest.TestCase):
         self.assertIn("check-command-migration-manifest.py", ci)
         self.assertIn("needs.changes.outputs.heavy == 'true'", ci)
 
-    def test_shared_process_mode_is_manual_hosted_and_preserves_default_runner(self) -> None:
+    def test_shared_process_mode_is_manual_hosted_ubuntu_only_and_replaces_the_sharded_suite(self) -> None:
         ci = load_ci()
         self.assertIn("default: nextest", ci)
-        matrix = ci.split("  test:\n", 1)[1].split("    steps:\n", 1)[0]
-        self.assertIn(
-            SHARED_PROCESS_MODE + " && '[\"ubuntu-latest\"]' || "
-            "'[\"ubuntu-latest\",\"macos-latest\",\"windows-latest\"]'", matrix,
-        )
-        self.assertIn("needs.changes.outputs.trusted == 'true'", matrix)
+        job = ci.split("  test:\n", 1)[1].split("    steps:\n", 1)[0]
+        self.assertIn("runs-on: ubuntu-latest", job)
+        self.assertIn(SHARED_PROCESS_MODE, job)
+        self.assertNotIn("pull_request", job)
+        for name in ("test-ubuntu", "test-macos", "test-windows"):
+            caller = ci.split(f"  {name}:\n", 1)[1].split("    uses:", 1)[0]
+            self.assertIn(f"!({SHARED_PROCESS_MODE})", caller)
         qualification = named_step(ci, "Shared-process workspace qualification")
-        self.assertIn("if: " + SHARED_PROCESS_MODE + " && matrix.os == 'ubuntu-latest'", qualification)
-        for name, command in (
-            ("Run tests", "cargo nextest run --workspace --all-features --locked --profile ci"),
-            ("Run doctests", "cargo test --workspace --all-features --locked --doc"),
-        ):
-            step = named_step(ci, name)
-            self.assertIn(command, step)
-            self.assertIn(f"!({SHARED_PROCESS_MODE})", step)
+        self.assertIn("if: " + SHARED_PROCESS_MODE, qualification)
         upload = named_step(ci, "Retain shared-process qualification evidence")
         self.assertIn("if: always() && " + SHARED_PROCESS_MODE, upload)
         self.assertIn("if-no-files-found: error", upload)
+        suite = (ROOT / ".github" / "workflows" / "ci-test-suite.yml").read_text(encoding="utf-8")
+        for name, command in (
+            ("Run tests", "cargo nextest run"),
+            ("Run doctests", "cargo test --workspace --all-features --locked --doc"),
+        ):
+            self.assertIn(command, named_step(suite, name))
 
     def test_shared_process_runs_exact_command_twice_with_fresh_hermetic_homes(self) -> None:
         proof = run_shared_process_fixture()
@@ -332,31 +332,24 @@ class CiWiringTests(unittest.TestCase):
             for identity in ISSUE_6698_TESTS:
                 self.assertIn(f"test {identity} ... ok", proof["evidence"][f"run-{run}.log"])
 
-    def test_shared_process_keeps_macos_budget_when_self_hosted_test_leg_is_absent(self) -> None:
+    def test_macos_budget_runs_on_every_heavy_dispatch_and_never_on_pull_requests(self) -> None:
         job = load_ci().split("  macos-budget:\n", 1)[1].split("    steps:\n", 1)[0]
         condition = next(line.strip()[4:] for line in job.splitlines() if line.strip().startswith("if: "))
         self.assertIn("runs-on: macos-latest", job)
         for event in ("pull_request", "push", "schedule", "workflow_dispatch"):
-            for mode in ("nextest", "shared-process-twice"):
-                for heavy, trusted, self_hosted in (
-                    (True, True, True), (True, True, False),
-                    (True, False, True), (False, True, True),
-                ):
-                    with self.subTest(event=event, mode=mode, heavy=heavy, trusted=trusted, self_hosted=self_hosted):
+            for heavy in (True, False):
+                for scoped in (True, False):
+                    with self.subTest(event=event, heavy=heavy, scoped=scoped):
                         expression = condition
                         for name, value in {
                             "github.event_name": event,
-                            "inputs.workspace_test_mode": mode,
                             "needs.changes.outputs.heavy": str(heavy).lower(),
-                            "needs.changes.outputs.trusted": str(trusted).lower(),
-                            "vars.CW_SELF_HOSTED_MAC": str(self_hosted).lower(),
+                            "needs.changes.outputs.scoped": str(scoped).lower(),
                         }.items():
                             expression = expression.replace(name, repr(value))
                         expression = re.sub(r"!(?!=)", "not ", expression.replace("&&", "and").replace("||", "or"))
                         actual = eval(expression, {"__builtins__": {}}, {})
-                        previous = heavy and event != "pull_request" and not (trusted and self_hosted)
-                        qualification = heavy and event == "workflow_dispatch" and mode == "shared-process-twice"
-                        self.assertEqual(actual, previous or qualification)
+                        self.assertEqual(actual, heavy and not scoped and event != "pull_request")
 
     def test_shared_process_retains_failed_first_run_and_still_runs_second_once(self) -> None:
         proof = run_shared_process_fixture("first_failure")

web/data/agent-docs.generated.json

@@ -334,8 +334,8 @@
     "sources": [
       {
         "path": "docs/RUNTIME_API.md",
-        "sha256": "c4f348c27735a7df47a79faf11a6d5e3284a1e67a76c9372e76a8809236af45a",
-        "content": "# Codewhale Engine Runtime API & Integration Contract\n\n> 阅读简体中文版：[zh_hans/RUNTIME_API.md](zh_hans/RUNTIME_API.md)。\n\n`codewhale app-server --http` exposes the [Codewhale Engine](ARCHITECTURE.md) through\nits canonical local Runtime API and control plane.\nLocal SDKs, mobile/remote-control clients, and editor integrations talk to it\ninstead of screen-scraping terminal output. It serves the full HTTP/SSE runtime\nAPI (`/v1/*`), a JSON-RPC control transport over stdio, and the phone-friendly\nmobile page. `codewhale doctor --json` provides machine-readable health, and\n`codewhale serve --acp` speaks the Agent Client Protocol over stdio for editors\nsuch as Zed.\n\n`codewhale serve --http` / `serve --mobile` remain as **compatibility aliases**\nfor `codewhale app-server --http` / `--mobile`; both launch the identical\nserver. Desktop, web, SDK and custom integrations should target\n`app-server --http`. `--stdio` and `--socket` expose compatibility controls.\nBare `app-server` selects initialized JSONL control over stdio. Local stdio,\nsocket and ACP owners do not bind a hidden TCP listener. Explicit `--http`\nretains HTTP/SSE and compatibility routes on the same Runtime owner.\n\n`codewhale exec` is the separate one-shot headless worker path (stream-json,\nfleet worker subprocess, CI primitive). It is not part of this API, but it\nshares the same runtime, provider/model resolution, permission profiles, and\nevent vocabulary.\n\nThis document is the stable integration contract for native workbench\napplications (and other local supervisors) that embed the Codewhale Engine.\n\n## Architecture\n\n```\nlocal supervisor / SDK / automation harness\n        │\n        ├─ codewhale app-server --http     → HTTP/SSE runtime API (/v1/*)        [canonical]\n        ├─ codewhale app-server --mobile   → runtime API + mobile control page\n        ├─ codewhale app-server --stdio    → JSON-RPC control transport over stdio\n        ├─ codewhale app-server --socket   → same JSON-RPC over a unix domain socket (desktop daemon)\n        ├─ codewhale doctor --json         → machine-readable health & capability\n        ├─ codewhale serve --acp           → ACP stdio agent for editors such as Zed\n        ├─ codewhale serve --mcp           → MCP stdio server\n        ├─ codewhale serve --http/--mobile → legacy aliases for `app-server --http/--mobile`\n        └─ codewhale exec [args]           → one-shot headless worker (stream-json)\n```\n\nThe engine runs as a local-only process. All APIs bind to `localhost` by\ndefault. No hosted relay, no provider-token custody, no secret leakage.\n\nFor the read-only record of what a thread or turn did, see\n[`docs/RECEIPTS.md`](RECEIPTS.md): `codewhale receipts` on the CLI and the\n`/receipt` routes under **Threads** below.\n\n## Runtime API entrypoints\n\n| Entry | Transport | Use |\n|---|---|---|\n| `codewhale web [--port 7878]` | HTTP/SSE on `127.0.0.1:7878` + embedded client | First-class loopback-only browser client; opens the default browser |\n| `codewhale app-server --http` | HTTP/SSE on `127.0.0.1:7878` | Full `/v1/*` runtime API (canonical) |\n| `codewhale app-server --mobile` | HTTP/SSE on loopback + `/mobile` | Runtime API + local mobile control page |\n| `codewhale app-server --stdio` | Initialized JSONL over stdio | Compatibility controls and explicit tool-free turns through the held manager; no hidden TCP listener or full Codex interaction channel. |\n| `codewhale app-server --socket [--socket-path P]` | RPC text messages over WebSocket on a `0600` Unix domain socket (macOS/Linux) or an owner-only named pipe (Windows) | Desktop daemon: multi-client, kernel peer identity checked, `daemon/attach` claim handshake. Windows native-app qualification remains open. |\n| `codewhale app-server` | Initialized JSONL over stdio | Same selection as `--stdio`; HTTP requires explicit `--http`. |\n| `codewhale serve --http` / `--mobile` | same server as `app-server --http`/`--mobile` | Compatibility aliases |\n\n`app-server --http` and `--mobile` launch the same mature runtime API server\nhistorically reached through `serve --http` — no routes or behavior changed, so\nevery endpoint documented below is identical across both entrypoints. The\nruntime API token is read from `--auth-token`, then `CODEWHALE_RUNTIME_TOKEN`,\nthen `DEEPSEEK_RUNTIME_TOKEN`; use `--insecure-no-auth` only with a loopback\nbind. The `serve` compatibility aliases keep their `--insecure` flag.\nBare `app-server` does not open an HTTP port. Selected attached HTTP listeners\nretain their explicit bearer and non-loopback security checks.\n\nInitialized stdio/private-socket controls support `thread/loaded/list` with\n`{cursor?: string | null, limit?: number | null}`. The result is\n`{data: string[], nextCursor: string | null}` from the held manager's loaded\nengine snapshot, including idle and archived engines still in memory. It dLine truncated
+        "sha256": "6f3b8885978ef2e1b33d0fdf66ad6ccf5c25464c1951b59658aec00745390db2",
+        "content": "# Codewhale Engine Runtime API & Integration Contract\n\n> 阅读简体中文版：[zh_hans/RUNTIME_API.md](zh_hans/RUNTIME_API.md)。\n\n`codewhale app-server --http` exposes the [Codewhale Engine](ARCHITECTURE.md) through\nits canonical local Runtime API and control plane.\nLocal SDKs, mobile/remote-control clients, and editor integrations talk to it\ninstead of screen-scraping terminal output. It serves the full HTTP/SSE runtime\nAPI (`/v1/*`), a JSON-RPC control transport over stdio, and the phone-friendly\nmobile page. `codewhale doctor --json` provides machine-readable health, and\n`codewhale serve --acp` speaks the Agent Client Protocol over stdio for editors\nsuch as Zed.\n\n`codewhale serve --http` / `serve --mobile` remain as **compatibility aliases**\nfor `codewhale app-server --http` / `--mobile`; both launch the identical\nserver. Desktop, web, SDK and custom integrations should target\n`app-server --http`. `--stdio` and `--socket` expose compatibility controls.\nBare `app-server` selects initialized JSONL control over stdio. Local stdio,\nsocket and ACP owners do not bind a hidden TCP listener. Explicit `--http`\nretains HTTP/SSE and compatibility routes on the same Runtime owner.\n\n`codewhale exec` is the separate one-shot headless worker path (stream-json,\nfleet worker subprocess, CI primitive). It is not part of this API, but it\nshares the same runtime, provider/model resolution, permission profiles, and\nevent vocabulary.\n\nThis document is the stable integration contract for native workbench\napplications (and other local supervisors) that embed the Codewhale Engine.\n\n## Architecture\n\n```\nlocal supervisor / SDK / automation harness\n        │\n        ├─ codewhale app-server --http     → HTTP/SSE runtime API (/v1/*)        [canonical]\n        ├─ codewhale app-server --mobile   → runtime API + mobile control page\n        ├─ codewhale app-server --stdio    → JSON-RPC control transport over stdio\n        ├─ codewhale app-server --socket   → same JSON-RPC over a unix domain socket (desktop daemon)\n        ├─ codewhale doctor --json         → machine-readable health & capability\n        ├─ codewhale serve --acp           → ACP stdio agent for editors such as Zed\n        ├─ codewhale serve --mcp           → MCP stdio server\n        ├─ codewhale serve --http/--mobile → legacy aliases for `app-server --http/--mobile`\n        └─ codewhale exec [args]           → one-shot headless worker (stream-json)\n```\n\nThe engine runs as a local-only process. All APIs bind to `localhost` by\ndefault. No hosted relay, no provider-token custody, no secret leakage.\n\nFor the read-only record of what a thread or turn did, see\n[`docs/RECEIPTS.md`](RECEIPTS.md): `codewhale receipts` on the CLI and the\n`/receipt` routes under **Threads** below.\n\n## Runtime API entrypoints\n\n| Entry | Transport | Use |\n|---|---|---|\n| `codewhale web [--port 7878]` | HTTP/SSE on `127.0.0.1:7878` + embedded client | First-class loopback-only browser client; opens the default browser |\n| `codewhale app-server --http` | HTTP/SSE on `127.0.0.1:7878` | Full `/v1/*` runtime API (canonical) |\n| `codewhale app-server --mobile` | HTTP/SSE on loopback + `/mobile` | Runtime API + local mobile control page |\n| `codewhale app-server --stdio` | Initialized JSONL over stdio | Compatibility controls and explicit tool-free turns through the held manager; no hidden TCP listener or full Codex interaction channel. |\n| `codewhale app-server --socket [--socket-path P]` | RPC text messages over WebSocket on a `0600` Unix domain socket (macOS/Linux) or an owner-only named pipe (Windows) | Desktop daemon: multi-client, kernel peer identity checked, `daemon/attach` claim handshake. Windows native-app qualification remains open. |\n| `codewhale app-server` | Initialized JSONL over stdio | Same selection as `--stdio`; HTTP requires explicit `--http`. |\n| `codewhale serve --http` / `--mobile` | same server as `app-server --http`/`--mobile` | Compatibility aliases |\n\n`app-server --http` and `--mobile` launch the same mature runtime API server\nhistorically reached through `serve --http` — no routes or behavior changed, so\nevery endpoint documented below is identical across both entrypoints. The\nruntime API token is read from `--auth-token`, then `CODEWHALE_RUNTIME_TOKEN`,\nthen `DEEPSEEK_RUNTIME_TOKEN`; use `--insecure-no-auth` only with a loopback\nbind. The `serve` compatibility aliases keep their `--insecure` flag.\nBare `app-server` does not open an HTTP port. Selected attached HTTP listeners\nretain their explicit bearer and non-loopback security checks.\n\nInitialized stdio/private-socket controls support `thread/loaded/list` with\n`{cursor?: string | null, limit?: number | null}`. The result is\n`{data: string[], nextCursor: string | null}` from the held manager's loaded\nengine snapshot, including idle and archived engines still in memory. It dLine truncated
       }
     ]
   },
@@ -377,8 +377,8 @@
     "sources": [
       {
         "path": "CONTRIBUTING.md",
-        "sha256": "7b3e523585eb5535555b1d8aaa66153ce0451b1cedc165dfba212e372dd71d41",
-        "content": "# Contributing to Codewhale\n\nCodewhale is what it is because of the people who used it, reported what broke,\ntranslated it, added their providers and sent fixes. Thank you for being here.\nEveryone is welcome, in English or 中文, with or without code.\n\nAll feedback is a gift to this project, including \"this looked wrong\" or \"I got\nlost here\" with no fix attached. The [roadmap](docs/ROADMAP.md) says where\nCodewhale is going and where help counts most.\n\n## Where things go\n\n| You have | Best place |\n| --- | --- |\n| A bug, a crash, or behavior that surprised you | [Bug report](https://github.com/codewhale-hq/CodeWhale/issues/new?template=bug_report.yml) |\n| An idea or a feature you want | [Feature request](https://github.com/codewhale-hq/CodeWhale/issues/new?template=feature_request.yml), or [Ideas](https://github.com/codewhale-hq/CodeWhale/discussions/categories/ideas) to talk it through first |\n| Feedback on how something looks, reads or feels | [Feedback](https://github.com/codewhale-hq/CodeWhale/issues/new?template=feedback.yml) |\n| A question | [Q&A](https://github.com/codewhale-hq/CodeWhale/discussions/categories/q-a) or [Discord](https://discord.gg/37gfS3ksug) |\n| A security problem | A [private advisory](https://github.com/codewhale-hq/CodeWhale/security/advisories/new), never a public issue ([policy](.github/SECURITY.md)) |\n| A fix for an open issue | A pull request that links the issue |\n| A translation, a provider or model, a docs fix | A pull request, no issue needed ([standing projects](#standing-projects)) |\n| A new feature, already written or not | An issue first ([why](#new-features-start-as-an-issue)) |\n\nIf you are not sure, open an issue. Nothing is filed in the wrong place for\nlong: a maintainer moves it.\n\n## New features start as an issue\n\nIssues are the best place for feature requests, ideas and bugs. A pull request\nis the best place for work we have already agreed on.\n\nA feature that first appears as a pull request takes longer to land than one\nthat starts as an issue, even when we want it. The questions that decide a\nfeature are about the product, not the diff: does it belong in the engine or in\na plugin, what does it do in each permission mode, how does it show up in\nreceipts, what is it called in 19 languages. Answering those in a review means\nrewriting code you already finished. Answering them in an issue costs you a few\nsentences.\n\nSo for a new feature:\n\n1. Open a [feature request](https://github.com/codewhale-hq/CodeWhale/issues/new?template=feature_request.yml).\n   Say what you are trying to do and what gets in the way. A proposed design is\n   welcome and optional.\n2. Tick \"I would like to open a pull request\" if you want to build it.\n3. A maintainer answers with the shape that fits, or with what is still\n   undecided. Then the pull request is small and merges quickly.\n\nA feature pull request that arrives without an issue is still read, and it is\nnever closed for that reason. We will most likely ask the same questions there\nand it will take longer.\n\n## Standing projects\n\nThese are always open, and a pull request needs no issue first:\n\n- **Translations.** You can improve the terminal language packs, the translated\n  READMEs in `docs/i18n/`, the website, and the translated docs in\n  `docs/zh_hans/` and `docs/id/`. The [localization matrix](docs/LOCALIZATION.md)\n  shows what each language has and how to add one, and the\n  [Chinese localization group](https://github.com/codewhale-hq/CodeWhale/issues/6804)\n  is recruiting.\n- **Providers and models.** You can add a missing provider or a new model, or\n  correct a price, a context size or a provider quirk.\n  [Providers](docs/PROVIDERS.md) lists what ships, and\n  `python3 scripts/check-provider-registry.py` checks your change.\n- **Documentation.** You can fix a typo, an unclear step, a missing example or\n  an instruction that no longer matches the product.\n- **Bug fixes.** If there is no issue yet, describe the bug in the pull request\n  by saying what you ran, what happened and what you expected.\n- **Planned work.** You can take anything in a\n  [milestone](docs/ROADMAP.md#releases) or anything labelled\n  [help wanted](https://github.com/codewhale-hq/CodeWhale/issues?q=is%3Aissue+is%3Aopen+label%3A%22help+wanted%22).\n  Comment on the issue to claim it so two people do not build the same thing.\n\nLarge changes to planned work are easier in layers; see\n[Layered and EPIC-Sized Work](#layered-and-epic-sized-work).\n\n### What a planned-work issue gives you\n\nEvery issue a maintainer opens for planned work, and every release tracking issue, has the same parts, so you can pick one up without asking first:\n\n- **Status:** what is on `main` already, with commit links, and what is still missing.\n- **Ways to help:** each piece open to contributors, with the files to read, the command to test it locally (usually `scripts/dev-test.sh <crate> <filter>Line truncated
+        "sha256": "096a609f93d82c8bcea9f967072d5306807cd4375390218f9d3784023df0dcb0",
+        "content": "# Contributing to Codewhale\n\nCodewhale is what it is because of the people who used it, reported what broke,\ntranslated it, added their providers and sent fixes. Thank you for being here.\nEveryone is welcome, in English or 中文, with or without code.\n\nAll feedback is a gift to this project, including \"this looked wrong\" or \"I got\nlost here\" with no fix attached. The [roadmap](docs/ROADMAP.md) says where\nCodewhale is going and where help counts most.\n\n## Where things go\n\n| You have | Best place |\n| --- | --- |\n| A bug, a crash, or behavior that surprised you | [Bug report](https://github.com/codewhale-hq/CodeWhale/issues/new?template=bug_report.yml) |\n| An idea or a feature you want | [Feature request](https://github.com/codewhale-hq/CodeWhale/issues/new?template=feature_request.yml), or [Ideas](https://github.com/codewhale-hq/CodeWhale/discussions/categories/ideas) to talk it through first |\n| Feedback on how something looks, reads or feels | [Feedback](https://github.com/codewhale-hq/CodeWhale/issues/new?template=feedback.yml) |\n| A question | [Q&A](https://github.com/codewhale-hq/CodeWhale/discussions/categories/q-a) or [Discord](https://discord.gg/37gfS3ksug) |\n| A security problem | A [private advisory](https://github.com/codewhale-hq/CodeWhale/security/advisories/new), never a public issue ([policy](.github/SECURITY.md)) |\n| A fix for an open issue | A pull request that links the issue |\n| A translation, a provider or model, a docs fix | A pull request, no issue needed ([standing projects](#standing-projects)) |\n| A new feature, already written or not | An issue first ([why](#new-features-start-as-an-issue)) |\n\nIf you are not sure, open an issue. Nothing is filed in the wrong place for\nlong: a maintainer moves it.\n\n## New features start as an issue\n\nIssues are the best place for feature requests, ideas and bugs. A pull request\nis the best place for work we have already agreed on.\n\nA feature that first appears as a pull request takes longer to land than one\nthat starts as an issue, even when we want it. The questions that decide a\nfeature are about the product, not the diff: does it belong in the engine or in\na plugin, what does it do in each permission mode, how does it show up in\nreceipts, what is it called in 19 languages. Answering those in a review means\nrewriting code you already finished. Answering them in an issue costs you a few\nsentences.\n\nSo for a new feature:\n\n1. Open a [feature request](https://github.com/codewhale-hq/CodeWhale/issues/new?template=feature_request.yml).\n   Say what you are trying to do and what gets in the way. A proposed design is\n   welcome and optional.\n2. Tick \"I would like to open a pull request\" if you want to build it.\n3. A maintainer answers with the shape that fits, or with what is still\n   undecided. Then the pull request is small and merges quickly.\n\nA feature pull request that arrives without an issue is still read, and it is\nnever closed for that reason. We will most likely ask the same questions there\nand it will take longer.\n\n## Standing projects\n\nThese are always open, and a pull request needs no issue first:\n\n- **Translations.** You can improve the terminal language packs, the translated\n  READMEs in `docs/i18n/`, the website, and the translated docs in\n  `docs/zh_hans/` and `docs/id/`. The [localization matrix](docs/LOCALIZATION.md)\n  shows what each language has and how to add one, and the\n  [Chinese localization group](https://github.com/codewhale-hq/CodeWhale/issues/6804)\n  is recruiting.\n- **Providers and models.** You can add a missing provider or a new model, or\n  correct a price, a context size or a provider quirk.\n  [Providers](docs/PROVIDERS.md) lists what ships, and\n  `python3 scripts/check-provider-registry.py` checks your change.\n- **Documentation.** You can fix a typo, an unclear step, a missing example or\n  an instruction that no longer matches the product.\n- **Bug fixes.** If there is no issue yet, describe the bug in the pull request\n  by saying what you ran, what happened and what you expected.\n- **Planned work.** You can take anything in a\n  [milestone](docs/ROADMAP.md#releases) or anything labelled\n  [help wanted](https://github.com/codewhale-hq/CodeWhale/issues?q=is%3Aissue+is%3Aopen+label%3A%22help+wanted%22).\n  Comment on the issue to claim it so two people do not build the same thing.\n\nLarge changes to planned work are easier in layers; see\n[Layered and EPIC-Sized Work](#layered-and-epic-sized-work).\n\n### What a planned-work issue gives you\n\nEvery issue a maintainer opens for planned work, and every release tracking issue, has the same parts, so you can pick one up without asking first:\n\n- **Status:** what is on `main` already, with commit links, and what is still missing.\n- **Ways to help:** each piece open to contributors, with the files to read, the command to test it locally (usually `scripts/dev-test.sh <crate> <filter>Line truncated
       },
       {
         "path": "docs/AGENT_ETHOS.md",
@@ -387,8 +387,8 @@
       },
       {
         "path": "docs/CONTRIBUTORS.md",
-        "sha256": "bedd1cc62969108856f75d823d13bde9f7217a936d4b0e7af3d2e6232cad5770",
-        "content": "# Contributors\n\nCodewhale is built in the open with a growing community of contributors. Every\nissue report and pull request is real project work — welcome at any experience\nlevel. This is the full per-PR contributor record, grouped into release/time\nbands with the newest band first so it stays scannable. Expand any band to see\neveryone.\n\nFor the live list, see the\n[GitHub contributors page](https://github.com/codewhale-hq/CodeWhale/graphs/contributors),\n[`AUTHOR_MAP`](https://github.com/codewhale-hq/CodeWhale/blob/main/.github/AUTHOR_MAP),\nand [CHANGELOG.md](../CHANGELOG.md).\n\n<!-- contributor-wall:start -->\n## Everyone\n\n[![433 people built Codewhale](assets/contributors.png)](#everyone)\n\n433 people have code, co-authored commits, carried-forward pull requests or credited reports in Codewhale. GitHub's contributor list shows only commit authors whose email is linked to their account, so it counts 212 of them. If your commits are missing there, add your commit email to your GitHub account and they will link.\n\n[0gl20shk0sbt36](https://github.com/0gl20shk0sbt36) · [1Git2Clone](https://github.com/1Git2Clone) · [20bytes](https://github.com/20bytes) · [47Cid](https://github.com/47Cid) · [7jrxt42BxFZo4iAnN4CX](https://github.com/7jrxt42BxFZo4iAnN4CX) · [8sero](https://github.com/8sero) · [A-Corner](https://github.com/A-Corner) · [aboimpinto](https://github.com/aboimpinto) · [Aboysky](https://github.com/Aboysky) · [adity982](https://github.com/adity982) · [AdityaVG13](https://github.com/AdityaVG13) · [AGSaturn](https://github.com/AGSaturn) · [ai-dengwsh](https://github.com/ai-dengwsh) · [aiapienthusiast](https://github.com/aiapienthusiast) · [Aitensa](https://github.com/Aitensa) · [AiurArtanis](https://github.com/AiurArtanis) · [alitvak69](https://github.com/alitvak69) · [all-lopezg](https://github.com/all-lopezg) · [alvin1](https://github.com/alvin1) · [amuthantamil](https://github.com/amuthantamil) · [AnaheimEX](https://github.com/AnaheimEX) · [Andrea-Bruno](https://github.com/Andrea-Bruno) · [Angel-Hair](https://github.com/Angel-Hair) · [angziii](https://github.com/angziii) · [AnonymousUser443](https://github.com/AnonymousUser443) · [antarikshraya](https://github.com/antarikshraya) · [Anyexyz](https://github.com/Anyexyz) · [Apeiron0w0](https://github.com/Apeiron0w0) · [aqilaziz](https://github.com/aqilaziz) · [AresNing](https://github.com/AresNing) · [ArronAI007](https://github.com/ArronAI007) · [Artenx](https://github.com/Artenx) · [asdfg314284230](https://github.com/asdfg314284230) · [Asteria1028](https://github.com/Asteria1028) · [asto18089](https://github.com/asto18089) · [atmosuwiryo](https://github.com/atmosuwiryo) · [axobase001](https://github.com/axobase001) · [aznikline](https://github.com/aznikline) · [baendlorel](https://github.com/baendlorel) · [banqii](https://github.com/banqii) · [Beltran12138](https://github.com/Beltran12138) · [bevis-wong](https://github.com/bevis-wong) · [BH8GCJ](https://github.com/BH8GCJ) · [BigBenLabs](https://github.com/BigBenLabs) · [Biilow](https://github.com/Biilow) · [bistack](https://github.com/bistack) · [BrathonBai](https://github.com/BrathonBai) · [bruce6135](https://github.com/bruce6135) · [BryonGo](https://github.com/BryonGo) · [buiducnhat](https://github.com/buiducnhat) · [buko](https://github.com/buko) · [BX166](https://github.com/BX166) · [c020627](https://github.com/c020627) · [cacdcaecawae](https://github.com/cacdcaecawae) · [caeserchen](https://github.com/caeserchen) · [caiyilian](https://github.com/caiyilian) · [CCChisato](https://github.com/CCChisato) · [ccomma](https://github.com/ccomma) · [cenab](https://github.com/cenab) · [ChaceLyee2101](https://github.com/ChaceLyee2101) · [chatman-media](https://github.com/chatman-media) · [chennest](https://github.com/chennest) · [chinaqy110](https://github.com/chinaqy110) · [chnjames](https://github.com/chnjames) · [chuntseevolving](https://github.com/chuntseevolving) · [ci4ic4](https://github.com/ci4ic4) · [cian-lien](https://github.com/cian-lien) · [cmdcorp6534](https://github.com/cmdcorp6534) · [cmyyy](https://github.com/cmyyy) · [codepgq](https://github.com/codepgq) · [Colorful-glassblock](https://github.com/Colorful-glassblock) · [ComeFromTheMars](https://github.com/ComeFromTheMars) · [CrepuscularIRIS](https://github.com/CrepuscularIRIS) · [ct-jaryn](https://github.com/ct-jaryn) · [ctxyao](https://github.com/ctxyao) · [cy2311](https://github.com/cy2311) · [cyq1017](https://github.com/cyq1017) · [czf0718](https://github.com/czf0718) · [dajiaohuang](https://github.com/dajiaohuang) · [DarrellThomas](https://github.com/DarrellThomas) · [ddlw](https://github.com/ddlw) · [demian-welt](https://github.com/demian-welt) · [Dev-next-gen](https://github.com/Dev-next-gen) · [dfwqdyl-ui](https://github.com/dfwqdyl-ui) · [DI-HUO-MING-YI](https://github.com/DI-HUO-MING-YI) · [DieMoe233](https://github.Line truncated
+        "sha256": "4368939ea860f6db9484fd86e613574b72eca71361f08a1130ec87c5f5c66c78",
+        "content": "# Contributors\n\nCodewhale is built in the open with a growing community of contributors. Every\nissue report and pull request is real project work — welcome at any experience\nlevel. This is the full per-PR contributor record, grouped into release/time\nbands with the newest band first so it stays scannable. Expand any band to see\neveryone.\n\nFor the live list, see the\n[GitHub contributors page](https://github.com/codewhale-hq/CodeWhale/graphs/contributors),\n[`AUTHOR_MAP`](https://github.com/codewhale-hq/CodeWhale/blob/main/.github/AUTHOR_MAP),\nand [CHANGELOG.md](../CHANGELOG.md).\n\n<!-- contributor-wall:start -->\n## Everyone\n\n[![433 people built Codewhale](assets/contributors.png)](#everyone)\n\n433 people have code, co-authored commits, carried-forward pull requests or credited reports in Codewhale. GitHub's contributor list shows only commit authors whose email is linked to their account, so it counts 212 of them. If your commits are missing there, add your commit email to your GitHub account and they will link.\n\n[0gl20shk0sbt36](https://github.com/0gl20shk0sbt36) · [1Git2Clone](https://github.com/1Git2Clone) · [20bytes](https://github.com/20bytes) · [47Cid](https://github.com/47Cid) · [7jrxt42BxFZo4iAnN4CX](https://github.com/7jrxt42BxFZo4iAnN4CX) · [8sero](https://github.com/8sero) · [A-Corner](https://github.com/A-Corner) · [aboimpinto](https://github.com/aboimpinto) · [Aboysky](https://github.com/Aboysky) · [adity982](https://github.com/adity982) · [AdityaVG13](https://github.com/AdityaVG13) · [AGSaturn](https://github.com/AGSaturn) · [ai-dengwsh](https://github.com/ai-dengwsh) · [aiapienthusiast](https://github.com/aiapienthusiast) · [Aitensa](https://github.com/Aitensa) · [AiurArtanis](https://github.com/AiurArtanis) · [alitvak69](https://github.com/alitvak69) · [all-lopezg](https://github.com/all-lopezg) · [alvin1](https://github.com/alvin1) · [amuthantamil](https://github.com/amuthantamil) · [AnaheimEX](https://github.com/AnaheimEX) · [Andrea-Bruno](https://github.com/Andrea-Bruno) · [Angel-Hair](https://github.com/Angel-Hair) · [angziii](https://github.com/angziii) · [AnonymousUser443](https://github.com/AnonymousUser443) · [antarikshraya](https://github.com/antarikshraya) · [Anyexyz](https://github.com/Anyexyz) · [Apeiron0w0](https://github.com/Apeiron0w0) · [aqilaziz](https://github.com/aqilaziz) · [AresNing](https://github.com/AresNing) · [ArronAI007](https://github.com/ArronAI007) · [Artenx](https://github.com/Artenx) · [asdfg314284230](https://github.com/asdfg314284230) · [Asteria1028](https://github.com/Asteria1028) · [asto18089](https://github.com/asto18089) · [atmosuwiryo](https://github.com/atmosuwiryo) · [axobase001](https://github.com/axobase001) · [aznikline](https://github.com/aznikline) · [baendlorel](https://github.com/baendlorel) · [banqii](https://github.com/banqii) · [Beltran12138](https://github.com/Beltran12138) · [bevis-wong](https://github.com/bevis-wong) · [BH8GCJ](https://github.com/BH8GCJ) · [BigBenLabs](https://github.com/BigBenLabs) · [Biilow](https://github.com/Biilow) · [bistack](https://github.com/bistack) · [BrathonBai](https://github.com/BrathonBai) · [bruce6135](https://github.com/bruce6135) · [BryonGo](https://github.com/BryonGo) · [buiducnhat](https://github.com/buiducnhat) · [buko](https://github.com/buko) · [BX166](https://github.com/BX166) · [c020627](https://github.com/c020627) · [cacdcaecawae](https://github.com/cacdcaecawae) · [caeserchen](https://github.com/caeserchen) · [caiyilian](https://github.com/caiyilian) · [CCChisato](https://github.com/CCChisato) · [ccomma](https://github.com/ccomma) · [cenab](https://github.com/cenab) · [ChaceLyee2101](https://github.com/ChaceLyee2101) · [chatman-media](https://github.com/chatman-media) · [chennest](https://github.com/chennest) · [chinaqy110](https://github.com/chinaqy110) · [chnjames](https://github.com/chnjames) · [chuntseevolving](https://github.com/chuntseevolving) · [ci4ic4](https://github.com/ci4ic4) · [cian-lien](https://github.com/cian-lien) · [cmdcorp6534](https://github.com/cmdcorp6534) · [cmyyy](https://github.com/cmyyy) · [codepgq](https://github.com/codepgq) · [Colorful-glassblock](https://github.com/Colorful-glassblock) · [ComeFromTheMars](https://github.com/ComeFromTheMars) · [CrepuscularIRIS](https://github.com/CrepuscularIRIS) · [ct-jaryn](https://github.com/ct-jaryn) · [ctxyao](https://github.com/ctxyao) · [cy2311](https://github.com/cy2311) · [cyq1017](https://github.com/cyq1017) · [czf0718](https://github.com/czf0718) · [dajiaohuang](https://github.com/dajiaohuang) · [DarrellThomas](https://github.com/DarrellThomas) · [ddlw](https://github.com/ddlw) · [demian-welt](https://github.com/demian-welt) · [Dev-next-gen](https://github.com/Dev-next-gen) · [dfwqdyl-ui](https://github.com/dfwqdyl-ui) · [DI-HUO-MING-YI](https://github.com/DI-HUO-MING-YI) · [DieMoe233](https://github.Line truncated
       },
       {
         "path": "docs/RELEASE_CHECKLIST.md",
@@ -403,8 +403,8 @@
     "sources": [
       {
         "path": "docs/RUNTIME_API.md",
-        "sha256": "c4f348c27735a7df47a79faf11a6d5e3284a1e67a76c9372e76a8809236af45a",
-        "content": "# Codewhale Engine Runtime API & Integration Contract\n\n> 阅读简体中文版：[zh_hans/RUNTIME_API.md](zh_hans/RUNTIME_API.md)。\n\n`codewhale app-server --http` exposes the [Codewhale Engine](ARCHITECTURE.md) through\nits canonical local Runtime API and control plane.\nLocal SDKs, mobile/remote-control clients, and editor integrations talk to it\ninstead of screen-scraping terminal output. It serves the full HTTP/SSE runtime\nAPI (`/v1/*`), a JSON-RPC control transport over stdio, and the phone-friendly\nmobile page. `codewhale doctor --json` provides machine-readable health, and\n`codewhale serve --acp` speaks the Agent Client Protocol over stdio for editors\nsuch as Zed.\n\n`codewhale serve --http` / `serve --mobile` remain as **compatibility aliases**\nfor `codewhale app-server --http` / `--mobile`; both launch the identical\nserver. Desktop, web, SDK and custom integrations should target\n`app-server --http`. `--stdio` and `--socket` expose compatibility controls.\nBare `app-server` selects initialized JSONL control over stdio. Local stdio,\nsocket and ACP owners do not bind a hidden TCP listener. Explicit `--http`\nretains HTTP/SSE and compatibility routes on the same Runtime owner.\n\n`codewhale exec` is the separate one-shot headless worker path (stream-json,\nfleet worker subprocess, CI primitive). It is not part of this API, but it\nshares the same runtime, provider/model resolution, permission profiles, and\nevent vocabulary.\n\nThis document is the stable integration contract for native workbench\napplications (and other local supervisors) that embed the Codewhale Engine.\n\n## Architecture\n\n```\nlocal supervisor / SDK / automation harness\n        │\n        ├─ codewhale app-server --http     → HTTP/SSE runtime API (/v1/*)        [canonical]\n        ├─ codewhale app-server --mobile   → runtime API + mobile control page\n        ├─ codewhale app-server --stdio    → JSON-RPC control transport over stdio\n        ├─ codewhale app-server --socket   → same JSON-RPC over a unix domain socket (desktop daemon)\n        ├─ codewhale doctor --json         → machine-readable health & capability\n        ├─ codewhale serve --acp           → ACP stdio agent for editors such as Zed\n        ├─ codewhale serve --mcp           → MCP stdio server\n        ├─ codewhale serve --http/--mobile → legacy aliases for `app-server --http/--mobile`\n        └─ codewhale exec [args]           → one-shot headless worker (stream-json)\n```\n\nThe engine runs as a local-only process. All APIs bind to `localhost` by\ndefault. No hosted relay, no provider-token custody, no secret leakage.\n\nFor the read-only record of what a thread or turn did, see\n[`docs/RECEIPTS.md`](RECEIPTS.md): `codewhale receipts` on the CLI and the\n`/receipt` routes under **Threads** below.\n\n## Runtime API entrypoints\n\n| Entry | Transport | Use |\n|---|---|---|\n| `codewhale web [--port 7878]` | HTTP/SSE on `127.0.0.1:7878` + embedded client | First-class loopback-only browser client; opens the default browser |\n| `codewhale app-server --http` | HTTP/SSE on `127.0.0.1:7878` | Full `/v1/*` runtime API (canonical) |\n| `codewhale app-server --mobile` | HTTP/SSE on loopback + `/mobile` | Runtime API + local mobile control page |\n| `codewhale app-server --stdio` | Initialized JSONL over stdio | Compatibility controls and explicit tool-free turns through the held manager; no hidden TCP listener or full Codex interaction channel. |\n| `codewhale app-server --socket [--socket-path P]` | RPC text messages over WebSocket on a `0600` Unix domain socket (macOS/Linux) or an owner-only named pipe (Windows) | Desktop daemon: multi-client, kernel peer identity checked, `daemon/attach` claim handshake. Windows native-app qualification remains open. |\n| `codewhale app-server` | Initialized JSONL over stdio | Same selection as `--stdio`; HTTP requires explicit `--http`. |\n| `codewhale serve --http` / `--mobile` | same server as `app-server --http`/`--mobile` | Compatibility aliases |\n\n`app-server --http` and `--mobile` launch the same mature runtime API server\nhistorically reached through `serve --http` — no routes or behavior changed, so\nevery endpoint documented below is identical across both entrypoints. The\nruntime API token is read from `--auth-token`, then `CODEWHALE_RUNTIME_TOKEN`,\nthen `DEEPSEEK_RUNTIME_TOKEN`; use `--insecure-no-auth` only with a loopback\nbind. The `serve` compatibility aliases keep their `--insecure` flag.\nBare `app-server` does not open an HTTP port. Selected attached HTTP listeners\nretain their explicit bearer and non-loopback security checks.\n\nInitialized stdio/private-socket controls support `thread/loaded/list` with\n`{cursor?: string | null, limit?: number | null}`. The result is\n`{data: string[], nextCursor: string | null}` from the held manager's loaded\nengine snapshot, including idle and archived engines still in memory. It dLine truncated
+        "sha256": "6f3b8885978ef2e1b33d0fdf66ad6ccf5c25464c1951b59658aec00745390db2",
+        "content": "# Codewhale Engine Runtime API & Integration Contract\n\n> 阅读简体中文版：[zh_hans/RUNTIME_API.md](zh_hans/RUNTIME_API.md)。\n\n`codewhale app-server --http` exposes the [Codewhale Engine](ARCHITECTURE.md) through\nits canonical local Runtime API and control plane.\nLocal SDKs, mobile/remote-control clients, and editor integrations talk to it\ninstead of screen-scraping terminal output. It serves the full HTTP/SSE runtime\nAPI (`/v1/*`), a JSON-RPC control transport over stdio, and the phone-friendly\nmobile page. `codewhale doctor --json` provides machine-readable health, and\n`codewhale serve --acp` speaks the Agent Client Protocol over stdio for editors\nsuch as Zed.\n\n`codewhale serve --http` / `serve --mobile` remain as **compatibility aliases**\nfor `codewhale app-server --http` / `--mobile`; both launch the identical\nserver. Desktop, web, SDK and custom integrations should target\n`app-server --http`. `--stdio` and `--socket` expose compatibility controls.\nBare `app-server` selects initialized JSONL control over stdio. Local stdio,\nsocket and ACP owners do not bind a hidden TCP listener. Explicit `--http`\nretains HTTP/SSE and compatibility routes on the same Runtime owner.\n\n`codewhale exec` is the separate one-shot headless worker path (stream-json,\nfleet worker subprocess, CI primitive). It is not part of this API, but it\nshares the same runtime, provider/model resolution, permission profiles, and\nevent vocabulary.\n\nThis document is the stable integration contract for native workbench\napplications (and other local supervisors) that embed the Codewhale Engine.\n\n## Architecture\n\n```\nlocal supervisor / SDK / automation harness\n        │\n        ├─ codewhale app-server --http     → HTTP/SSE runtime API (/v1/*)        [canonical]\n        ├─ codewhale app-server --mobile   → runtime API + mobile control page\n        ├─ codewhale app-server --stdio    → JSON-RPC control transport over stdio\n        ├─ codewhale app-server --socket   → same JSON-RPC over a unix domain socket (desktop daemon)\n        ├─ codewhale doctor --json         → machine-readable health & capability\n        ├─ codewhale serve --acp           → ACP stdio agent for editors such as Zed\n        ├─ codewhale serve --mcp           → MCP stdio server\n        ├─ codewhale serve --http/--mobile → legacy aliases for `app-server --http/--mobile`\n        └─ codewhale exec [args]           → one-shot headless worker (stream-json)\n```\n\nThe engine runs as a local-only process. All APIs bind to `localhost` by\ndefault. No hosted relay, no provider-token custody, no secret leakage.\n\nFor the read-only record of what a thread or turn did, see\n[`docs/RECEIPTS.md`](RECEIPTS.md): `codewhale receipts` on the CLI and the\n`/receipt` routes under **Threads** below.\n\n## Runtime API entrypoints\n\n| Entry | Transport | Use |\n|---|---|---|\n| `codewhale web [--port 7878]` | HTTP/SSE on `127.0.0.1:7878` + embedded client | First-class loopback-only browser client; opens the default browser |\n| `codewhale app-server --http` | HTTP/SSE on `127.0.0.1:7878` | Full `/v1/*` runtime API (canonical) |\n| `codewhale app-server --mobile` | HTTP/SSE on loopback + `/mobile` | Runtime API + local mobile control page |\n| `codewhale app-server --stdio` | Initialized JSONL over stdio | Compatibility controls and explicit tool-free turns through the held manager; no hidden TCP listener or full Codex interaction channel. |\n| `codewhale app-server --socket [--socket-path P]` | RPC text messages over WebSocket on a `0600` Unix domain socket (macOS/Linux) or an owner-only named pipe (Windows) | Desktop daemon: multi-client, kernel peer identity checked, `daemon/attach` claim handshake. Windows native-app qualification remains open. |\n| `codewhale app-server` | Initialized JSONL over stdio | Same selection as `--stdio`; HTTP requires explicit `--http`. |\n| `codewhale serve --http` / `--mobile` | same server as `app-server --http`/`--mobile` | Compatibility aliases |\n\n`app-server --http` and `--mobile` launch the same mature runtime API server\nhistorically reached through `serve --http` — no routes or behavior changed, so\nevery endpoint documented below is identical across both entrypoints. The\nruntime API token is read from `--auth-token`, then `CODEWHALE_RUNTIME_TOKEN`,\nthen `DEEPSEEK_RUNTIME_TOKEN`; use `--insecure-no-auth` only with a loopback\nbind. The `serve` compatibility aliases keep their `--insecure` flag.\nBare `app-server` does not open an HTTP port. Selected attached HTTP listeners\nretain their explicit bearer and non-loopback security checks.\n\nInitialized stdio/private-socket controls support `thread/loaded/list` with\n`{cursor?: string | null, limit?: number | null}`. The result is\n`{data: string[], nextCursor: string | null}` from the held manager's loaded\nengine snapshot, including idle and archived engines still in memory. It dLine truncated
       },
       {
         "path": "docs/RUNTIME_ROUTES.md",

web/lib/release-credits.ts

@@ -35,6 +35,7 @@ export const RELEASE_CONTRIBUTORS: string[] = [
   "@luochen211",
   "@gannp",
   "@Felixkw12",
+  "@MannXo",
 ];
 
 /**