Compare commits

..
2 Commits
Author SHA1 Message Date
github-actions[bot] 8fc6882c22 docs(release): prepare v1.24.2 notes (#8640)
Summary:
Generate a bilingual, product-focused draft from merged pull request metadata. Reuse the selected release-bound PR when one is available.

Verification:
Validate the catalog, citations, bilingual fields, and rendered GitHub release notes before committing.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
2026-08-12 23:40:54 +08:00
f53edb599a fix(session): fold recovery lineages into one ordinary list row / 恢复谱系折叠为单行逻辑会话 (#8599)
Fixes the 1.24.2 Catalog regression that re-projected every recovery JSONL
as its own ordinary sidebar topic after upgrade.

- Disposable session-catalog/v4.sqlite with logical projection
  (logical_topic_id, ordinary_visible, recovery roles/canonical)
- Re-anchor cross-topic recovery members onto the root logical topic;
  ordinary list stays one row per conversation
- Content-based covering/preferred selection; filename fallback for
  legacy *-recovery-<16hex>.jsonl
- Disable automatic physical recovery GC; keep explicit UI preview
  cleanup and `reasonix sessions cleanup --apply`
- History "Other saved versions" for inspection, default choice, and
  dry-run cleanup; externally removed open sessions move to one stable
  recovery lane without recreating the original path
- Bounded catalog metadata sync (30s) and sidebar reads (10s); live tabs
  stay visible while the projection lags, deduped by catalog projection;
  connection-pool regression guards (adopted from #8594)
- CLI sessions diagnose/cleanup and doctor sessions lineage metrics

Co-authored-by: yingziisme <[email protected]>
Co-authored-by: Cursor <[email protected]>
2026-08-12 21:36:31 +08:00
12269 changed files with 747052 additions and 1131414 deletions

No files matched your search

-5
View File
@@ -2,10 +2,8 @@
.gitignore text eol=lf
.npmrc text eol=lf
*.astro text eol=lf
*.bat text eol=crlf
*.css text eol=lf
*.go text eol=lf
*.golden text eol=lf
*.html text eol=lf
*.js text eol=lf
*.json text eol=lf
@@ -26,8 +24,5 @@
*.xml text eol=lf
*.yaml text eol=lf
*.yml text eol=lf
Makefile text eol=lf
.githooks/** text eol=lf
benchmarks/**/*.sh text eol=lf
internal/runtime/agent/testdata/perseveration/** -text
desktop/frontend-next/perf/media/delivery.pdf binary
+13 -43
View File
@@ -1,44 +1,14 @@
# Ownership for the studio line. Standard: docs/DOCS_STANDARD.md.
#
# Each rule lists exactly two owners: primary first, backup second.
# GitHub requests review from both. The last matching rule wins.
# repolint (codeowners, doc-owner) fails a rule for a missing path, a rule
# without a backup, and a document header that names different owners.
# Every pull request requests review from the maintainers below.
# Either maintainer can review and merge any change.
* @SivanCola @esengine
# Default: kernel, Studio shell, tooling, standing instructions.
* @esengine @SivanCola
# User-facing guides and references.
/docs/ @SivanCola @esengine
/docs/THEME_AUTHOR_GUIDE.md @esengine @SivanCola
/sdk/ @SivanCola @esengine
/README.md @SivanCola @esengine
/README.zh-CN.md @SivanCola @esengine
/SECURITY.md @SivanCola @esengine
# Contracts and processes the Studio line runs on.
/docs/DOCS_STANDARD.md @esengine @SivanCola
/docs/STUDIO_*.md @esengine @SivanCola
/docs/SESSION_MEMORY_RETRIEVAL.md @esengine @SivanCola
/docs/*_CONTRACT.md @esengine @SivanCola
/docs/TOOL_APPROVAL_MODES.md @esengine @SivanCola
/docs/CAPABILITY_DIAGNOSTICS.md @esengine @SivanCola
/docs/CHECKPOINTS.md @esengine @SivanCola
/docs/EXTENSION_*.md @esengine @SivanCola
/docs/SUBAGENT_*.md @esengine @SivanCola
/docs/ROADMAP.md @esengine @SivanCola
/docs/MIGRATING.md @esengine @SivanCola
/docs/design/ @esengine @SivanCola
/release-notes/ @esengine @SivanCola
# Release signing: provider policy, its contract and the workflow that uses it.
/.signpath/ @esengine @SivanCola
/.github/workflows/release-studio.yml @esengine @SivanCola
/.github/workflows/studio-certum-signing-smoke.yml @esengine @SivanCola
/.github/actions/setup-certum/ @esengine @SivanCola
/cmd/signpath-contract/ @esengine @SivanCola
/scripts/install-smoke.ps1 @esengine @SivanCola
/scripts/sign-certum.ps1 @esengine @SivanCola
/scripts/verify-windows-authenticode.ps1 @esengine @SivanCola
/scripts/verify-windows-package.ps1 @esengine @SivanCola
/scripts/windows-signing-lib.ps1 @esengine @SivanCola
# Keep the provider-side signing policy and its GitHub control plane under the
# same explicit review boundary as the machine-readable contract.
/.signpath/ @SivanCola @esengine
/.github/workflows/release-stable.yml @SivanCola @esengine
/.github/workflows/release-stable-trigger.yml @SivanCola @esengine
/.github/workflows/release-desktop.yml @SivanCola @esengine
/scripts/complete-signpath-request.ps1 @SivanCola @esengine
/scripts/package-windows-desktop.sh @SivanCola @esengine
/scripts/verify-windows-authenticode.ps1 @SivanCola @esengine
/cmd/signpath-contract/ @SivanCola @esengine
+5 -7
View File
@@ -7,25 +7,23 @@ body:
attributes:
value: |
Thanks for filing a bug. Pick which version line you're on below — it gets
labeled automatically, no maintainer action needed. Not sure which line
you run? See the [version roadmap](https://github.com/esengine/DeepSeek-Reasonix/discussions/10748).
labeled (`v1`/`v2`) automatically, no maintainer action needed.
- type: dropdown
id: version-line
attributes:
label: Version line
description: Which line are you running?
options:
- "2.x — Reasonix Studio, studio branch (active development)"
- "1.x — main-v2 branch (maintenance / stable)"
- "0.x — legacy TypeScript (maintenance only)"
- "v2 — Go rewrite (1.x), main-v2 (active development)"
- "v1 — Legacy TypeScript (0.x), maintenance only"
validations:
required: true
- type: input
id: version
attributes:
label: Exact version
description: Output of `reasonix --version`, or the version shown in Studio's settings.
placeholder: e.g. 2.20.0 or 1.39.0
description: Output of `reasonix --version`.
placeholder: e.g. 1.0.0
validations:
required: true
- type: textarea
-9
View File
@@ -1,14 +1,5 @@
blank_issues_enabled: false
contact_links:
- name: Version roadmap (1.x / 2.x)
url: https://github.com/esengine/DeepSeek-Reasonix/discussions/10748
about: Which line to use, which branch takes which changes, and where Reasonix is heading.
- name: Questions & discussions
url: https://github.com/esengine/DeepSeek-Reasonix/discussions
about: For usage questions, ideas, and general discussion — please use Discussions instead of opening an issue.
- name: Discord
url: https://discord.gg/XF78rEME2D
about: Bilingual community chat for setup help, workflow showcases, and feature ideas.
- name: QQ group (DeepSeek-Reasonix官方群, 1093562660)
url: https://qm.qq.com/q/i59b0z2R8s
about: Chinese-language community chat for setup help and everyday questions.
+2 -3
View File
@@ -9,9 +9,8 @@ body:
label: Version line
description: Which line is this for?
options:
- "2.x — Reasonix Studio, studio branch (active development)"
- "1.x — main-v2 branch (maintenance / stable)"
- "0.x — legacy TypeScript (maintenance only)"
- "v2 — Go rewrite (1.x), main-v2 (active development)"
- "v1 — Legacy TypeScript (0.x), maintenance only"
validations:
required: true
- type: textarea
-35
View File
@@ -1,35 +0,0 @@
name: Connect to Certum cloud signing
description: Verify the pinned SimplySign installer and connect to the configured certificate.
inputs:
username:
required: true
description: SimplySign account
otp-uri:
required: true
description: TOTP provisioning URI
thumbprint:
required: true
description: Expected SHA-1 certificate thumbprint
runs:
using: composite
steps:
- name: Verify SimplySign installer
shell: pwsh
run: |
$ErrorActionPreference = 'Stop'
Invoke-WebRequest 'https://files.certum.eu/software/SimplySignDesktop/Windows/9.4.3.90/SimplySignDesktop-9.4.3.90-64-bit-en.msi' -OutFile SimplySignDesktop.msi
$hash = (Get-FileHash -Algorithm SHA256 SimplySignDesktop.msi).Hash
if ($hash -ne 'CE2B38EF0124A574BC08E5558726099CAAC79A7C58B8D3D712F629371FE1BCB7') {
throw "SimplySign installer SHA-256 $hash does not match the pinned 9.4.3.90 build"
}
$signature = Get-AuthenticodeSignature SimplySignDesktop.msi
if ($signature.Status -ne 'Valid' -or $signature.SignerCertificate.Subject -notmatch 'Asseco Data Systems') {
throw 'SimplySign installer signature or publisher verification failed'
}
- name: Authenticate cloud certificate
uses: dismine/windows-app-signing-setup-action@89ae3b032d4bc7a5b98d1a42a34e61ecb6faad64
with:
certum-username: ${{ inputs.username }}
certum-otp-uri: ${{ inputs.otp-uri }}
certum-key-id: ${{ inputs.thumbprint }}
capture-diagnostics: 'false'
+2 -61
View File
@@ -1,6 +1,5 @@
version: 2
updates:
# 2.x (studio): the development line.
- package-ecosystem: "gomod"
directory: "/"
schedule:
@@ -22,7 +21,7 @@ updates:
update-types: ["minor", "patch"]
- package-ecosystem: "npm"
directory: "/desktop/frontend-next"
directory: "/desktop/frontend"
schedule:
interval: "weekly"
open-pull-requests-limit: 5
@@ -32,7 +31,7 @@ updates:
update-types: ["minor", "patch"]
- package-ecosystem: "npm"
directory: "/desktop/electron"
directory: "/site"
schedule:
interval: "weekly"
open-pull-requests-limit: 5
@@ -48,61 +47,3 @@ updates:
groups:
actions:
patterns: ["*"]
# 1.x (main-v2): maintenance. Dependabot reads this file from the default
# branch only, so the 1.x line is named here; it takes patch updates alone.
- package-ecosystem: "gomod"
directory: "/"
target-branch: "main-v2"
schedule:
interval: "weekly"
open-pull-requests-limit: 3
ignore:
- dependency-name: "*"
update-types: ["version-update:semver-major", "version-update:semver-minor"]
groups:
go-1x:
patterns: ["*"]
update-types: ["patch"]
- package-ecosystem: "gomod"
directory: "/desktop"
target-branch: "main-v2"
schedule:
interval: "weekly"
open-pull-requests-limit: 3
ignore:
- dependency-name: "*"
update-types: ["version-update:semver-major", "version-update:semver-minor"]
groups:
go-1x:
patterns: ["*"]
update-types: ["patch"]
- package-ecosystem: "npm"
directory: "/desktop/frontend"
target-branch: "main-v2"
schedule:
interval: "weekly"
open-pull-requests-limit: 3
ignore:
- dependency-name: "*"
update-types: ["version-update:semver-major", "version-update:semver-minor"]
groups:
npm-1x:
patterns: ["*"]
update-types: ["patch"]
- package-ecosystem: "npm"
directory: "/site"
target-branch: "main-v2"
schedule:
interval: "weekly"
open-pull-requests-limit: 3
ignore:
- dependency-name: "*"
update-types: ["version-update:semver-major", "version-update:semver-minor"]
groups:
npm-1x:
patterns: ["*"]
update-types: ["patch"]
+9 -9
View File
@@ -6,36 +6,36 @@
agent:
- changed-files:
- any-glob-to-any-file:
- 'internal/runtime/agent/**'
- 'internal/session/control/**'
- 'internal/agent/**'
- 'internal/control/**'
mcp:
- changed-files:
- any-glob-to-any-file:
- 'internal/ext/plugin/**'
- 'internal/plugin/**'
- 'internal/codegraph/**'
config:
- changed-files:
- any-glob-to-any-file:
- 'internal/contract/config/**'
- 'internal/assembly/boot/**'
- 'internal/config/**'
- 'internal/boot/**'
provider:
- changed-files:
- any-glob-to-any-file:
- 'internal/contract/provider/**'
- 'internal/provider/**'
skills:
- changed-files:
- any-glob-to-any-file:
- 'internal/ext/skill/**'
- 'internal/contract/tool/**'
- 'internal/skill/**'
- 'internal/tool/**'
tui:
- changed-files:
- any-glob-to-any-file:
- 'internal/frontend/cli/**'
- 'internal/cli/**'
desktop:
- changed-files:
-7
View File
@@ -1,7 +0,0 @@
# Logins allowed to open pull requests, one per line, matched case-insensitively.
# Repository owners, members and collaborators, and bots, are always allowed.
lancy69
chuiwenwei
xxoingr
kevin9327
imcreazy
+20 -39
View File
@@ -1,44 +1,7 @@
<!--
Target branch: 2.x features and fixes go to `studio`. 1.x takes bug fixes,
provider/API compatibility, release/updater, security and platform stability
only, on `main-v2`.
Pull requests are limited to maintainers and invited contributors; one from
anyone else is closed automatically. Open an issue instead.
-->
## Summary
-
## Cause
-
## Blast radius
-
## Neighbouring behaviours tested
-
## Why this layer
-
## UI changes
<!--
Delete this section if the PR touches nothing under `desktop/frontend-next/src`
or the website pages.
-->
- Screenshots, before and after: light and dark, a wide window and a phone-width one (about 390 px). Attach them to this description.
- Entries moved, hidden or removed: none / list each one.
- Hiding or removing a visible control needs a linked issue and a release-notes line naming it. Do not retire UI with `display: none !important` or a feature flag.
- If a row of `docs/STUDIO_PARITY.md` changes, update it here.
## Issues
<!--
@@ -58,6 +21,24 @@ cancellation, reduced-motion behavior, and failure fallback. Verification must
assert that the target API was actually called; a mock that silently skips it
does not count.
## Disclosure
## Documentation impact
- [ ] AI assistance was used for this change (optional; for transparency only, it does not change how the change is reviewed)
Documentation-impact: TODO
For changes to user-visible CLI, Desktop, configuration, provider, permission,
or tool behavior, use one of:
- `Documentation-impact: updated - <what changed>` and update `docs/*.md`.
- `Documentation-impact: none - <why the embedded documentation remains correct>`.
## Cache impact
Cache-impact: TODO
Cache-guard: TODO
System-prompt-review: N/A
For cache-sensitive changes, fill these lines before requesting review:
- `Cache-impact`: `none`, `low`, `medium`, or `high`, plus the reason.
- `Cache-guard`: the focused guard test/command added or run, or why an existing guard covers the change.
- `System-prompt-review`: required reviewer/approval note when provider-visible system prompt, memory prefix, output style, or skill index behavior changes.
Binary file not shown.

Before

Width:  |  Height:  |  Size: 138 KiB

-19
View File
@@ -1,19 +0,0 @@
<?xml version="1.0" encoding="UTF-8"?>
<svg id="_图层_2" data-name="图层 2" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 145.55 113.29">
<defs>
<style>
.cls-1 {
fill: #ffffff;
stroke: #ffffff;
stroke-miterlimit: 10;
}
</style>
</defs>
<g id="_图层_1-2" data-name="图层 1">
<g>
<path class="cls-1" d="M144.68,38.49l-.06-.23c-.88-3.28-2.5-5.94-4.58-8.06.14,5.65-2.96,11.02-6.22,16.66l-.39.68c-2.26,3.91-4.66,7.94-6.57,11.1l-.86,1.38c-3.36,5.44-6.27,10.14-12.18,14.18-8.34,5.87-18.2,5.81-26.92,5.76-2.81-.02-5.48-.03-7.95.14l-.35.02c-3.22.06-5.96,1.57-8.17,4.49l-.14.18c-.86,1.06-1.7,2.26-2.58,3.54-3.43,4.92-7.69,11.04-16.17,12.42-4.37.86-9.98.84-14.94.83h-1.95c-.52,0-1.06,0-1.61,0-6.95,0-16.08-.89-21.94-6.55-.15-.15-.3-.3-.44-.45.5,4.73,2.33,8.64,5.44,11.65,5.86,5.66,14.98,6.55,21.94,6.55.55,0,1.09,0,1.61-.01h1.93c4.96.02,10.58.04,14.96-.83,8.48-1.37,12.74-7.5,16.17-12.42.88-1.27,1.72-2.48,2.58-3.54l.14-.18c2.21-2.92,4.95-4.43,8.17-4.49l.35-.02c2.47-.17,5.13-.16,7.95-.14,8.72.05,18.58.11,26.91-5.76,5.92-4.03,8.82-8.73,12.19-14.17l.86-1.39c1.91-3.15,4.3-7.18,6.57-11.09l.39-.69c3.81-6.6,7.41-12.84,5.86-19.57ZM120.9,23.02c-.28,0-.56,0-.83,0-9.68-.19-24.03-.09-35.57-.01l-2.04.02c-.93.01-1.82.02-2.67.03-8.36.08-14.4.13-23.33,3.82l-.27.12c-10.76,4.68-16.91,12.16-22.83,21.95-5.53,8.76-12.62,20.57-16.32,26.79-.41.69-.77,1.3-1.09,1.84l-.49.84c-1.49,2.53-3.23,5.49-4.21,8.95,3.98,1.9,8.52,2.65,12.71,2.91.27-.77.64-1.56,1.07-2.38.48-.94,1.05-1.9,1.63-2.89l.49-.84c.96-1.62,2.38-3.99,4.05-6.78,3.82-6.36,8.98-14.88,13.19-21.55l.07-.11c5.02-8.31,9.2-13.45,16.91-16.81l.11-.05c6.57-2.7,10.54-2.74,18.43-2.81.87-.01,1.78-.02,2.69-.03l1.99-.02c9.17-.06,20.13-.14,29.01-.07,2.26.01,4.39.04,6.32.08h.12s.11,0,.11,0c1.25-.01,2.91.11,4.61.46,1.03.22,2.08.52,3.05.93.21-.35.41-.71.62-1.06l.39-.68c1.92-3.33,3.79-6.57,4.97-9.82-4.08-1.92-8.7-2.76-12.89-2.82Z"/>
<path class="cls-1" d="M115.07,11.81c-9.7-.19-24.07-.09-35.62-.01h-1.99c-.93.03-1.82.04-2.67.04-8.36.08-14.4.14-23.33,3.83l-.27.12c-10.76,4.67-16.91,12.16-22.83,21.95-6.13,9.71-14.19,23.21-17.41,28.63l-.5.84c-2.08,3.55-4.68,7.97-4.94,13.39v.2s0,.21,0,.21c.01.81.06,1.61.15,2.38.14.15.29.3.44.45,1.53,1.47,3.28,2.63,5.15,3.52,3.98,1.9,8.52,2.65,12.71,2.91,1.41.09,2.78.12,4.08.12.55,0,1.09-.01,1.61-.01h1.95c4.95.01,10.57.03,14.94-.83,8.48-1.38,12.74-7.5,16.17-12.42.88-1.28,1.72-2.48,2.58-3.54l.14-.18c2.21-2.92,4.95-4.44,8.17-4.49l.35-.02c2.47-.17,5.14-.16,7.95-.14,8.71.05,18.58.1,26.92-5.76,5.91-4.04,8.82-8.74,12.18-14.18l.86-1.39c1.74-2.86,3.87-6.45,5.95-10.03.21-.35.41-.71.62-1.06l.39-.68c1.92-3.33,3.79-6.57,4.97-9.82.82-2.26,1.31-4.53,1.25-6.84-5.18-5.29-13.22-7.25-19.97-7.19ZM122.56,40.37l-.39.67c-2.21,3.81-4.55,7.77-6.39,10.79l-.84,1.36c-3.01,4.87-4.83,7.81-8.48,10.29l-.1.07c-4.94,3.49-11.95,3.45-19.38,3.41-2.88-.02-5.87-.04-8.79.16-7.1.19-13.51,3.58-18.07,9.57-1.13,1.4-2.12,2.83-3.08,4.21-2.92,4.18-4.71,6.57-7.64,7.02l-.31.06c-3.09.63-8.28.61-12.45.6h-2.16c-3.85.07-7.01-.16-9.45-.69-2.24-.48-3.87-1.22-4.89-2.2-.66-.64-1.54-1.81-1.63-4.65.11-1.35.71-2.82,1.52-4.35.48-.94,1.05-1.9,1.63-2.89l.49-.84c3.17-5.33,11.19-18.75,17.24-28.33l.07-.11c5.02-8.32,9.2-13.45,16.92-16.81l.1-.05c6.57-2.7,10.54-2.74,18.43-2.82.87,0,1.78,0,2.69-.03h1.94c11.52-.09,25.86-.19,35.38,0h.23c1.25-.01,2.92.11,4.61.46,3.15.66,6.4,2.12,7.27,5.02.2,1.2-.89,3.62-2.27,6.18-.7,1.31-1.48,2.65-2.2,3.9ZM120.07,23.01c-9.68-.19-24.03-.09-35.57-.01l-2.04.02c-.93.01-1.82.02-2.67.03-8.36.08-14.4.13-23.33,3.82l-.27.12c-10.76,4.68-16.91,12.16-22.83,21.95-5.53,8.76-12.62,20.57-16.32,26.79-.78-.35-1.41-.77-1.9-1.24-.66-.64-1.54-1.81-1.63-4.65.18-2.18,1.62-4.64,3.15-7.25l.49-.83c3.17-5.32,11.18-18.73,17.24-28.33l.07-.12c5.02-8.31,9.2-13.45,16.92-16.81l.1-.04c6.57-2.7,10.54-2.74,18.43-2.82.87-.01,1.78-.01,2.69-.03h1.99c11.5-.09,25.82-.19,35.33,0h.23c3.57-.04,10.55,1.02,11.88,5.48.14.84-.35,2.27-1.13,3.93-.28,0-.56,0-.83,0Z"/>
<path class="cls-1" d="M134.68,16.09l-.06-.23c-3.07-11.45-15.08-15.33-24.55-15.25-9.68-.19-24.03-.09-35.57-.01h-2.04c-.93.03-1.82.04-2.67.04-8.36.08-14.4.14-23.33,3.83l-.27.12c-10.76,4.67-16.91,12.16-22.83,21.95-6.14,9.73-14.2,23.22-17.41,28.62l-.49.85c-2.09,3.55-4.69,7.96-4.95,13.39v.2s0,.21,0,.21c.09,5.57,1.82,10.13,5.15,13.58.14.15.29.3.44.45,1.53,1.47,3.28,2.63,5.15,3.52,3.98,1.9,8.52,2.65,12.71,2.91,1.41.09,2.78.12,4.08.12.55,0,1.09-.01,1.61-.01h1.95c4.95.01,10.57.03,14.94-.83,8.48-1.38,12.74-7.5,16.17-12.42.88-1.28,1.72-2.48,2.58-3.54l.14-.18c2.21-2.92,4.95-4.44,8.17-4.49l.35-.02c2.47-.17,5.14-.16,7.95-.14,8.71.05,18.58.1,26.92-5.76,5.91-4.04,8.82-8.74,12.18-14.18l.86-1.39c1.74-2.86,3.87-6.45,5.95-10.03.21-.35.41-.71.62-1.06l.39-.68c1.92-3.33,3.79-6.57,4.97-9.82.82-2.26,1.31-4.53,1.25-6.84-.02-.96-.14-1.93-.36-2.91ZM119.76,25.27c-.7,1.31-1.48,2.65-2.2,3.9l-.39.67c-1.19,2.04-2.41,4.13-3.57,6.09-1.01,1.7-1.97,3.31-2.82,4.7l-.84,1.36c-3.01,4.87-4.83,7.81-8.48,10.29l-.1.07c-4.94,3.49-11.96,3.45-19.38,3.41-2.89-.02-5.87-.04-8.79.16-7.1.18-13.51,3.58-18.07,9.57-1.13,1.4-2.12,2.83-3.08,4.21-2.92,4.18-4.71,6.57-7.64,7.02l-.31.06c-3.09.63-8.27.61-12.45.6h-2.16c-3.85.07-7.01-.16-9.45-.69-1.16-.25-2.16-.57-2.99-.96-.78-.35-1.41-.77-1.9-1.24-.66-.64-1.54-1.81-1.63-4.65.18-2.18,1.62-4.64,3.15-7.25l.49-.83c3.17-5.32,11.18-18.73,17.24-28.33l.07-.12c5.02-8.31,9.2-13.45,16.92-16.81l.1-.04c6.57-2.7,10.54-2.74,18.43-2.82.87-.01,1.78-.01,2.69-.03h1.99c11.5-.09,25.82-.19,35.33,0h.23c3.57-.04,10.55,1.02,11.88,5.48.14.84-.35,2.27-1.13,3.93-.34.72-.73,1.48-1.14,2.25Z"/>
</g>
</g>
</svg>

Before

Width:  |  Height:  |  Size: 5.4 KiB

-19
View File
@@ -1,19 +0,0 @@
<?xml version="1.0" encoding="UTF-8"?>
<svg id="_图层_2" data-name="图层 2" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 145.55 113.29">
<defs>
<style>
.cls-1 {
fill: #111111;
stroke: #111111;
stroke-miterlimit: 10;
}
</style>
</defs>
<g id="_图层_1-2" data-name="图层 1">
<g>
<path class="cls-1" d="M144.68,38.49l-.06-.23c-.88-3.28-2.5-5.94-4.58-8.06.14,5.65-2.96,11.02-6.22,16.66l-.39.68c-2.26,3.91-4.66,7.94-6.57,11.1l-.86,1.38c-3.36,5.44-6.27,10.14-12.18,14.18-8.34,5.87-18.2,5.81-26.92,5.76-2.81-.02-5.48-.03-7.95.14l-.35.02c-3.22.06-5.96,1.57-8.17,4.49l-.14.18c-.86,1.06-1.7,2.26-2.58,3.54-3.43,4.92-7.69,11.04-16.17,12.42-4.37.86-9.98.84-14.94.83h-1.95c-.52,0-1.06,0-1.61,0-6.95,0-16.08-.89-21.94-6.55-.15-.15-.3-.3-.44-.45.5,4.73,2.33,8.64,5.44,11.65,5.86,5.66,14.98,6.55,21.94,6.55.55,0,1.09,0,1.61-.01h1.93c4.96.02,10.58.04,14.96-.83,8.48-1.37,12.74-7.5,16.17-12.42.88-1.27,1.72-2.48,2.58-3.54l.14-.18c2.21-2.92,4.95-4.43,8.17-4.49l.35-.02c2.47-.17,5.13-.16,7.95-.14,8.72.05,18.58.11,26.91-5.76,5.92-4.03,8.82-8.73,12.19-14.17l.86-1.39c1.91-3.15,4.3-7.18,6.57-11.09l.39-.69c3.81-6.6,7.41-12.84,5.86-19.57ZM120.9,23.02c-.28,0-.56,0-.83,0-9.68-.19-24.03-.09-35.57-.01l-2.04.02c-.93.01-1.82.02-2.67.03-8.36.08-14.4.13-23.33,3.82l-.27.12c-10.76,4.68-16.91,12.16-22.83,21.95-5.53,8.76-12.62,20.57-16.32,26.79-.41.69-.77,1.3-1.09,1.84l-.49.84c-1.49,2.53-3.23,5.49-4.21,8.95,3.98,1.9,8.52,2.65,12.71,2.91.27-.77.64-1.56,1.07-2.38.48-.94,1.05-1.9,1.63-2.89l.49-.84c.96-1.62,2.38-3.99,4.05-6.78,3.82-6.36,8.98-14.88,13.19-21.55l.07-.11c5.02-8.31,9.2-13.45,16.91-16.81l.11-.05c6.57-2.7,10.54-2.74,18.43-2.81.87-.01,1.78-.02,2.69-.03l1.99-.02c9.17-.06,20.13-.14,29.01-.07,2.26.01,4.39.04,6.32.08h.12s.11,0,.11,0c1.25-.01,2.91.11,4.61.46,1.03.22,2.08.52,3.05.93.21-.35.41-.71.62-1.06l.39-.68c1.92-3.33,3.79-6.57,4.97-9.82-4.08-1.92-8.7-2.76-12.89-2.82Z"/>
<path class="cls-1" d="M115.07,11.81c-9.7-.19-24.07-.09-35.62-.01h-1.99c-.93.03-1.82.04-2.67.04-8.36.08-14.4.14-23.33,3.83l-.27.12c-10.76,4.67-16.91,12.16-22.83,21.95-6.13,9.71-14.19,23.21-17.41,28.63l-.5.84c-2.08,3.55-4.68,7.97-4.94,13.39v.2s0,.21,0,.21c.01.81.06,1.61.15,2.38.14.15.29.3.44.45,1.53,1.47,3.28,2.63,5.15,3.52,3.98,1.9,8.52,2.65,12.71,2.91,1.41.09,2.78.12,4.08.12.55,0,1.09-.01,1.61-.01h1.95c4.95.01,10.57.03,14.94-.83,8.48-1.38,12.74-7.5,16.17-12.42.88-1.28,1.72-2.48,2.58-3.54l.14-.18c2.21-2.92,4.95-4.44,8.17-4.49l.35-.02c2.47-.17,5.14-.16,7.95-.14,8.71.05,18.58.1,26.92-5.76,5.91-4.04,8.82-8.74,12.18-14.18l.86-1.39c1.74-2.86,3.87-6.45,5.95-10.03.21-.35.41-.71.62-1.06l.39-.68c1.92-3.33,3.79-6.57,4.97-9.82.82-2.26,1.31-4.53,1.25-6.84-5.18-5.29-13.22-7.25-19.97-7.19ZM122.56,40.37l-.39.67c-2.21,3.81-4.55,7.77-6.39,10.79l-.84,1.36c-3.01,4.87-4.83,7.81-8.48,10.29l-.1.07c-4.94,3.49-11.95,3.45-19.38,3.41-2.88-.02-5.87-.04-8.79.16-7.1.19-13.51,3.58-18.07,9.57-1.13,1.4-2.12,2.83-3.08,4.21-2.92,4.18-4.71,6.57-7.64,7.02l-.31.06c-3.09.63-8.28.61-12.45.6h-2.16c-3.85.07-7.01-.16-9.45-.69-2.24-.48-3.87-1.22-4.89-2.2-.66-.64-1.54-1.81-1.63-4.65.11-1.35.71-2.82,1.52-4.35.48-.94,1.05-1.9,1.63-2.89l.49-.84c3.17-5.33,11.19-18.75,17.24-28.33l.07-.11c5.02-8.32,9.2-13.45,16.92-16.81l.1-.05c6.57-2.7,10.54-2.74,18.43-2.82.87,0,1.78,0,2.69-.03h1.94c11.52-.09,25.86-.19,35.38,0h.23c1.25-.01,2.92.11,4.61.46,3.15.66,6.4,2.12,7.27,5.02.2,1.2-.89,3.62-2.27,6.18-.7,1.31-1.48,2.65-2.2,3.9ZM120.07,23.01c-9.68-.19-24.03-.09-35.57-.01l-2.04.02c-.93.01-1.82.02-2.67.03-8.36.08-14.4.13-23.33,3.82l-.27.12c-10.76,4.68-16.91,12.16-22.83,21.95-5.53,8.76-12.62,20.57-16.32,26.79-.78-.35-1.41-.77-1.9-1.24-.66-.64-1.54-1.81-1.63-4.65.18-2.18,1.62-4.64,3.15-7.25l.49-.83c3.17-5.32,11.18-18.73,17.24-28.33l.07-.12c5.02-8.31,9.2-13.45,16.92-16.81l.1-.04c6.57-2.7,10.54-2.74,18.43-2.82.87-.01,1.78-.01,2.69-.03h1.99c11.5-.09,25.82-.19,35.33,0h.23c3.57-.04,10.55,1.02,11.88,5.48.14.84-.35,2.27-1.13,3.93-.28,0-.56,0-.83,0Z"/>
<path class="cls-1" d="M134.68,16.09l-.06-.23c-3.07-11.45-15.08-15.33-24.55-15.25-9.68-.19-24.03-.09-35.57-.01h-2.04c-.93.03-1.82.04-2.67.04-8.36.08-14.4.14-23.33,3.83l-.27.12c-10.76,4.67-16.91,12.16-22.83,21.95-6.14,9.73-14.2,23.22-17.41,28.62l-.49.85c-2.09,3.55-4.69,7.96-4.95,13.39v.2s0,.21,0,.21c.09,5.57,1.82,10.13,5.15,13.58.14.15.29.3.44.45,1.53,1.47,3.28,2.63,5.15,3.52,3.98,1.9,8.52,2.65,12.71,2.91,1.41.09,2.78.12,4.08.12.55,0,1.09-.01,1.61-.01h1.95c4.95.01,10.57.03,14.94-.83,8.48-1.38,12.74-7.5,16.17-12.42.88-1.28,1.72-2.48,2.58-3.54l.14-.18c2.21-2.92,4.95-4.44,8.17-4.49l.35-.02c2.47-.17,5.14-.16,7.95-.14,8.71.05,18.58.1,26.92-5.76,5.91-4.04,8.82-8.74,12.18-14.18l.86-1.39c1.74-2.86,3.87-6.45,5.95-10.03.21-.35.41-.71.62-1.06l.39-.68c1.92-3.33,3.79-6.57,4.97-9.82.82-2.26,1.31-4.53,1.25-6.84-.02-.96-.14-1.93-.36-2.91ZM119.76,25.27c-.7,1.31-1.48,2.65-2.2,3.9l-.39.67c-1.19,2.04-2.41,4.13-3.57,6.09-1.01,1.7-1.97,3.31-2.82,4.7l-.84,1.36c-3.01,4.87-4.83,7.81-8.48,10.29l-.1.07c-4.94,3.49-11.96,3.45-19.38,3.41-2.89-.02-5.87-.04-8.79.16-7.1.18-13.51,3.58-18.07,9.57-1.13,1.4-2.12,2.83-3.08,4.21-2.92,4.18-4.71,6.57-7.64,7.02l-.31.06c-3.09.63-8.27.61-12.45.6h-2.16c-3.85.07-7.01-.16-9.45-.69-1.16-.25-2.16-.57-2.99-.96-.78-.35-1.41-.77-1.9-1.24-.66-.64-1.54-1.81-1.63-4.65.18-2.18,1.62-4.64,3.15-7.25l.49-.83c3.17-5.32,11.18-18.73,17.24-28.33l.07-.12c5.02-8.31,9.2-13.45,16.92-16.81l.1-.04c6.57-2.7,10.54-2.74,18.43-2.82.87-.01,1.78-.01,2.69-.03h1.99c11.5-.09,25.82-.19,35.33,0h.23c3.57-.04,10.55,1.02,11.88,5.48.14.84-.35,2.27-1.13,3.93-.34.72-.73,1.48-1.14,2.25Z"/>
</g>
</g>
</svg>

Before

Width:  |  Height:  |  Size: 5.4 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 2.4 KiB

-37
View File
@@ -1,37 +0,0 @@
name: acp-conformance
# The agent is driven over stdio by the official ACP TypeScript SDK, and every
# message it writes is held to the schema that SDK ships. Not a required check:
# it runs when the ACP frontend or the harness changes, and on request.
on:
pull_request:
branches: [main-v2, studio]
paths:
- "internal/frontend/acp/**"
- "tools/acp-conformance/**"
- ".github/workflows/acp-conformance.yml"
workflow_dispatch:
permissions:
contents: read
concurrency:
group: acp-conformance-${{ github.ref }}
cancel-in-progress: true
jobs:
conformance:
runs-on: ubuntu-latest
timeout-minutes: 15
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
persist-credentials: false
- uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
with:
go-version-file: go.mod
cache: true
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
with:
node-version: "22"
- run: make acp-conformance
-293
View File
@@ -1,293 +0,0 @@
name: Android
# The Android shell under `android/` ships from the same tag the desktop line
# publishes, and lands on the same release page.
#
# It is a separate workflow rather than a job inside release-studio.yml for two
# reasons. The toolchains share nothing — Go, the Android SDK, a Gradle wrapper —
# so a runner carrying both would install three toolchains to build one artifact.
# And `signing-contract` pins which workflows may reach the SignPath credentials;
# this one never touches them, and keeping it out of that graph is what keeps the
# fingerprint stable.
#
# A failure here does not hold back the desktop release: the two are independent
# by construction, and an APK is worth less than the desktop builds it sits
# beside.
on:
push:
tags: ['studio-v*']
workflow_dispatch:
inputs:
tag:
description: "Existing studio tag to build for and attach to (for example studio-v2.51.0)"
required: false
type: string
ref:
description: "A branch to build as a trial run of this workflow; nothing is attached to any release"
required: false
type: string
permissions:
contents: write # upload the APK to the release the tag belongs to
concurrency:
group: android-${{ inputs.tag || inputs.ref || github.ref_name }}
cancel-in-progress: false
jobs:
resolve:
name: resolve the tag
runs-on: ubuntu-latest
permissions:
contents: read
outputs:
tag: ${{ steps.release.outputs.tag }}
version: ${{ steps.release.outputs.version }}
sha: ${{ steps.release.outputs.sha }}
candidate: ${{ steps.release.outputs.candidate }}
publish: ${{ steps.release.outputs.publish }}
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
fetch-depth: 0
persist-credentials: false
# The same tag grammar release-studio.yml enforces, spelled the same way:
# two workflows that disagree about which tags name a release would publish
# one artifact for a tag the other refused.
- name: Resolve tag and version
id: release
env:
IN_TAG: ${{ inputs.tag }}
IN_REF: ${{ inputs.ref }}
REF_NAME: ${{ github.ref_name }}
run: |
# A trial run builds a branch as a throwaway version and publishes nothing.
if [ -n "$IN_REF" ] && [ -z "$IN_TAG" ]; then
sha="$(git rev-parse --verify "origin/$IN_REF^{commit}")"
{ echo "tag="; echo "version=v0.0.0-dev.${sha:0:7}"; echo "sha=$sha"
echo "candidate=false"; echo "publish=false"; } >> "$GITHUB_OUTPUT"
echo "trial build of $IN_REF at $sha"
exit 0
fi
tag="${IN_TAG:-$REF_NAME}"
semver='^(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)(-([0-9A-Za-z-]+)(\.[0-9A-Za-z-]+)*)?$'
if [[ ! "$tag" =~ ^studio-v(.+)$ ]] || [[ ! "${BASH_REMATCH[1]}" =~ $semver ]]; then
echo "::error::studio release tag must be studio-vMAJOR.MINOR.PATCH[-PRERELEASE], got: $tag"
exit 1
fi
sha="$(git rev-parse --verify "refs/tags/$tag^{commit}")"
if ! git merge-base --is-ancestor "$sha" origin/studio; then
echo "::error::$tag is not on studio history"
exit 1
fi
version="${tag#studio-}"
candidate=false
if [[ "${version#v}" =~ $semver ]] && [ -n "${BASH_REMATCH[4]}" ]; then
candidate=true
fi
{ echo "tag=$tag"; echo "version=$version"; echo "sha=$sha"
echo "candidate=$candidate"; echo "publish=true"; } >> "$GITHUB_OUTPUT"
echo "resolved $tag -> $version at $sha (candidate=$candidate)"
build:
name: build the APK
needs: resolve
runs-on: ubuntu-latest
permissions:
contents: read
# `publish` decides whether to attach anything, and the only honest answer to
# "will this APK install" is whether a key was restored.
outputs:
signed: ${{ steps.key.outputs.signed }}
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
ref: ${{ needs.resolve.outputs.sha }}
fetch-depth: 0
persist-credentials: false
- uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
with:
go-version-file: go.mod
cache: true
cache-dependency-path: go.sum
# pnpm first: setup-node's pnpm cache needs the executable to exist.
- uses: pnpm/action-setup@ea17c68df8912ef543352723c149a84f56e3d413 # v4
with:
version: 10
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
with:
node-version: "24"
cache: pnpm
cache-dependency-path: desktop/frontend-next/pnpm-lock.yaml
- uses: actions/setup-java@b843ebf9b913146b38da2eaa7aeddf76074ae1ad # v6.1.0
with:
distribution: temurin
java-version: "21"
cache: gradle
# The SDK is the one thing nothing sets for you, per android/README.md. The
# runner image carries one; this asserts the two pieces the build names
# rather than re-installing them, so a missing package fails here by name
# instead of surfacing later as an unresolved compileSdk.
- name: Provide the Android SDK
run: |
if [ -z "${ANDROID_HOME:-}" ]; then
echo "::error::ANDROID_HOME is unset; the runner image is expected to provide the Android SDK"
exit 1
fi
for piece in "platforms/android-35" "build-tools/35.0.0"; do
if [ ! -d "$ANDROID_HOME/$piece" ]; then
echo "::error::the Android SDK at $ANDROID_HOME has no $piece; android/README.md names it as a requirement"
exit 1
fi
done
echo "ANDROID_SDK_ROOT=$ANDROID_HOME" >> "$GITHUB_ENV"
# A key that exists signs; a key that does not leaves the artifact
# unsigned, which is what a release with no key has always produced. The
# decision is the repository's, not this workflow's: it reads the four
# secrets android/README.md already names.
- name: Restore the signing key
id: key
env:
KEYSTORE_BASE64: ${{ secrets.REASONIX_KEYSTORE_BASE64 }}
STORE_PASSWORD: ${{ secrets.REASONIX_KEYSTORE_PASSWORD }}
KEY_PASSWORD: ${{ secrets.REASONIX_KEY_PASSWORD }}
KEY_ALIAS: ${{ secrets.REASONIX_KEY_ALIAS }}
run: |
if [ -z "$KEYSTORE_BASE64" ] || [ -z "$STORE_PASSWORD" ] || [ -z "$KEY_ALIAS" ]; then
echo "signed=false" >> "$GITHUB_OUTPUT"
echo "no signing key configured: the APK will be unsigned"
exit 0
fi
mkdir -p android/keystore
printf '%s' "$KEYSTORE_BASE64" | base64 --decode > android/keystore/release.jks
{
echo "storeFile=keystore/release.jks"
echo "storePassword=${STORE_PASSWORD}"
echo "keyAlias=${KEY_ALIAS}"
# PKCS#12 keystores use one password for both; only write the second
# when it is genuinely a different fact.
[ -n "$KEY_PASSWORD" ] && echo "keyPassword=${KEY_PASSWORD}"
} > android/keystore/keystore.properties
chmod 600 android/keystore/*
echo "signed=true" >> "$GITHUB_OUTPUT"
echo "signing key restored"
- name: Build the kernel and the page
working-directory: android
env:
VERSION: ${{ needs.resolve.outputs.version }}
run: |
export REASONIX_APP_VERSION="${VERSION#v}"
bash ./scripts/build-kernel.sh
bash ./scripts/sync-page.sh
# The one end-to-end check that needs no phone: the kernel the shell starts,
# serving the page the APK carries. Running it here means an APK that would
# open blank fails the build instead of the install.
- name: Verify the packaged page
working-directory: android
run: |
(cd .. && go build -o reasonix-verify ./cmd/reasonix)
KERNEL_BIN="$PWD/../reasonix-verify" bash ./scripts/verify-page.sh
- name: Assemble
working-directory: android
env:
VERSION: ${{ needs.resolve.outputs.version }}
run: |
# The version comes from the tag, so the About screen and the update
# check name this release. versionCode keeps its existing shape, which
# is why the build digit stays 1: it is a single decimal place, and a
# number that grows past it would collide with the next patch.
./gradlew assembleRelease --no-daemon "-PreasonixVersion=${VERSION#v}"
- name: Name the artifact
id: apk
working-directory: android/app/build/outputs/apk/release
run: |
src="$(ls ./*.apk | head -n 1)"
if [ -z "$src" ]; then
echo "::error::assembleRelease produced no APK"
exit 1
fi
case "$src" in
*-unsigned.apk) name="ReasonixMobile-${{ needs.resolve.outputs.version }}-unsigned.apk" ;;
*) name="ReasonixMobile-${{ needs.resolve.outputs.version }}.apk" ;;
esac
mv "$src" "$name"
sha256sum "$name" > "$name.sha256"
{
echo "path=$PWD/$name"
echo "name=$name"
} >> "$GITHUB_OUTPUT"
ls -l "$name" "$name.sha256"
# An unsigned package is not a lesser APK, it is an unusable one: Android's
# package manager rejects it outright rather than reporting a signature
# mismatch. Say so where the run summary can be read, instead of letting a
# green build imply a shippable artifact.
- name: Say what an unsigned build means
if: steps.key.outputs.signed != 'true'
run: |
echo "::warning::no signing key configured, so this APK is unsigned. Android refuses to install an unsigned package, so the release will carry no APK; the file stays a workflow artifact for whoever signs it."
{
echo "### No APK on this release"
echo ""
echo "The build produced an unsigned APK, which Android refuses to install."
echo "It is attached to this run as a workflow artifact only. Add the four"
echo "signing secrets to publish one."
} >> "$GITHUB_STEP_SUMMARY"
- uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v8
with:
name: android-apk
path: |
android/app/build/outputs/apk/release/*.apk
android/app/build/outputs/apk/release/*.sha256
if-no-files-found: error
publish:
name: attach the APK
needs: [resolve, build]
# Nothing to attach unless the build was signed. Skipping is the honest
# outcome: a release page with an APK nobody can install is worse than a
# release page without one.
if: needs.build.outputs.signed == 'true' && needs.resolve.outputs.publish == 'true'
runs-on: ubuntu-latest
permissions:
contents: write
steps:
- uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8.0.2
with:
name: android-apk
path: dist
# The release itself belongs to release-studio.yml, which creates it and
# owns its notes. This waits for it rather than creating a second one: two
# jobs racing to define a release is how a page ends up with the wrong body.
- name: Attach to the release
env:
GH_TOKEN: ${{ github.token }}
TAG: ${{ needs.resolve.outputs.tag }}
REPO: ${{ github.repository }}
run: |
# 90 minutes: on the last Studio release the page appeared about 40
# minutes after the tag run started, and the APK build spends part of
# that before this wait begins.
for _ in $(seq 1 360); do
if gh release view "$TAG" --repo "$REPO" >/dev/null 2>&1; then
gh release upload "$TAG" dist/*.apk dist/*.sha256 --repo "$REPO" --clobber
echo "attached to $TAG"
exit 0
fi
sleep 15
done
echo "::error::release $TAG did not appear within 90 minutes; the APK was built but not published"
exit 1
+49
View File
@@ -0,0 +1,49 @@
name: Cache impact guard
on:
pull_request_target:
branches: [main-v2]
types: [opened, synchronize, reopened, edited, ready_for_review]
# The guard only cares about the prompt/tool surfaces listed in
# scripts/check-cache-impact.sh; PRs confined to clearly unrelated paths
# can never trip it, so skip the run entirely. Keep the ignore list
# conservative: when in doubt, run.
paths-ignore:
- 'docs/**'
- 'site/**'
- 'release-notes/**'
- 'workers/**'
- 'benchmarks/**'
- 'npm/**'
- '*.md'
permissions:
contents: read
pull-requests: read
concurrency:
group: cache-impact-${{ github.event.pull_request.number }}
cancel-in-progress: true
jobs:
cache-impact:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
with:
ref: ${{ github.event.pull_request.base.sha }}
- name: Collect PR changed files
env:
GH_TOKEN: ${{ github.token }}
PR_NUMBER: ${{ github.event.pull_request.number }}
REPOSITORY: ${{ github.repository }}
run: |
gh api --paginate "repos/${REPOSITORY}/pulls/${PR_NUMBER}/files" \
--jq '.[].filename' > "$RUNNER_TEMP/cache-impact-files.txt"
- name: Check cache-sensitive PR metadata
env:
PR_BODY: ${{ github.event.pull_request.body }}
CACHE_IMPACT_CHANGED_FILES_FILE: ${{ runner.temp }}/cache-impact-files.txt
run: bash scripts/check-cache-impact.sh
+294 -804
View File
File diff suppressed because it is too large. Load diff
-35
View File
@@ -1,35 +0,0 @@
name: Close linked issues
# GitHub closes an issue named by "Fixes #N" only when the pull request merges
# into the default branch. This does the same for the 1.x line. It reads the
# merged pull request's description and never checks out its code.
on:
pull_request_target:
types: [closed]
branches: [main-v2]
permissions:
issues: write
jobs:
close:
if: github.event.pull_request.merged == true
runs-on: ubuntu-latest
steps:
- uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9
with:
script: |
const pr = context.payload.pull_request;
const { owner, repo } = context.repo;
// GitHub's own closing-keyword grammar, applied to this repository's issues only.
const keyword = /\b(?:close[sd]?|fix(?:e[sd])?|resolve[sd]?)\s*:?\s+#(\d+)\b/gi;
const numbers = [...new Set([...(pr.body ?? '').matchAll(keyword)].map((m) => Number(m[1])))];
const sha = (pr.merge_commit_sha ?? '').slice(0, 9);
for (const issue_number of numbers) {
const { data } = await github.rest.issues.get({ owner, repo, issue_number });
if (data.pull_request || data.state !== 'open') continue;
await github.rest.issues.createComment({ owner, repo, issue_number,
body: `Fixed on the 1.x line by #${pr.number} (${sha}). It ships in the next 1.x release.` });
await github.rest.issues.update({ owner, repo, issue_number, state: 'closed', state_reason: 'completed' });
core.info(`closed #${issue_number}`);
}
+14 -27
View File
@@ -4,25 +4,23 @@ on:
# Docs/release-notes/root-Markdown changes carry no analyzable code; the
# weekly cron keeps the full scan cadence regardless.
push:
branches: ["main-v2", "studio"]
branches: ["main-v2"]
paths-ignore:
- 'docs/**'
- 'release-notes/**'
- '*.md'
pull_request:
branches: ["main-v2", "studio"]
branches: ["main-v2"]
paths-ignore:
- 'docs/**'
- 'release-notes/**'
- '*.md'
# A schedule runs only from the default branch, so the weekly scan checks out
# both lines itself: 1.x is in maintenance, and security is part of it.
schedule:
- cron: "27 3 * * 1"
jobs:
analyze:
name: Analyze (${{ matrix.language }}${{ matrix.branch && format(', {0}', matrix.branch) || '' }})
name: Analyze (${{ matrix.language }})
runs-on: ubuntu-latest
permissions:
security-events: write
@@ -32,32 +30,21 @@ jobs:
strategy:
fail-fast: false
matrix:
language: [go, javascript-typescript, actions]
# Empty means "the ref this event is about"; a scheduled run names each line.
branch: ${{ fromJSON(github.event_name == 'schedule' && '["main-v2", "studio"]' || '[""]') }}
include:
- language: go
build-mode: autobuild
- language: javascript-typescript
build-mode: none
- language: actions
build-mode: none
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
ref: ${{ matrix.branch }}
- id: head
if: matrix.branch != ''
run: echo "sha=$(git rev-parse HEAD)" >> "$GITHUB_OUTPUT"
- uses: actions/checkout@v7
- name: Initialize CodeQL
uses: github/codeql-action/init@9b3254085494b55a0f7c4c2b3d2d52cc712ce370 # v4.39.1
uses: github/codeql-action/init@v4.37.4
with:
languages: ${{ matrix.language }}
build-mode: ${{ matrix.language == 'go' && 'autobuild' || 'none' }}
build-mode: ${{ matrix.build-mode }}
- name: Perform CodeQL Analysis
if: matrix.branch == ''
uses: github/codeql-action/analyze@9b3254085494b55a0f7c4c2b3d2d52cc712ce370 # v4.39.1
uses: github/codeql-action/[email protected]
with:
category: "/language:${{ matrix.language }}"
# Results from a checked-out line are filed under that line's ref, not the
# default branch the schedule happened to run from.
- name: Perform CodeQL Analysis (${{ matrix.branch }})
if: matrix.branch != ''
uses: github/codeql-action/analyze@9b3254085494b55a0f7c4c2b3d2d52cc712ce370 # v4.39.1
with:
category: "/language:${{ matrix.language }}"
ref: refs/heads/${{ matrix.branch }}
sha: ${{ steps.head.outputs.sha }}
+3 -3
View File
@@ -19,12 +19,12 @@ jobs:
deploy:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- uses: pnpm/action-setup@ea17c68df8912ef543352723c149a84f56e3d413 # v6.1.0
- uses: actions/checkout@v7
- uses: pnpm/action-setup@v6.0.9
with:
version: 10
run_install: false
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
- uses: actions/setup-node@v7
with:
node-version: "24"
cache: pnpm
+2 -2
View File
@@ -19,8 +19,8 @@ jobs:
deploy:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
- uses: actions/checkout@v7
- uses: actions/setup-node@v7
with:
node-version: '22'
cache: npm
+3 -3
View File
@@ -19,12 +19,12 @@ jobs:
deploy:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- uses: pnpm/action-setup@ea17c68df8912ef543352723c149a84f56e3d413 # v6.1.0
- uses: actions/checkout@v7
- uses: pnpm/action-setup@v6.0.9
with:
version: 10
run_install: false
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
- uses: actions/setup-node@v7
with:
node-version: "24"
cache: pnpm
-44
View File
@@ -1,44 +0,0 @@
name: Deploy ops events relay
on:
workflow_dispatch:
permissions:
contents: read
concurrency:
group: deploy-ops-events
cancel-in-progress: false
jobs:
deploy:
if: github.ref == 'refs/heads/platform'
runs-on: ubuntu-latest
environment: production
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
with:
node-version: '22'
cache: npm
cache-dependency-path: workers/ops-events/package-lock.json
- name: Validate
working-directory: workers/ops-events
run: |
npm ci
npm run typecheck
npm test
- name: Deploy
working-directory: workers/ops-events
env:
CLOUDFLARE_API_TOKEN: ${{ secrets.CLOUDFLARE_API_TOKEN }}
run: npx wrangler deploy
- name: Sync secrets
working-directory: workers/ops-events
env:
CLOUDFLARE_API_TOKEN: ${{ secrets.CLOUDFLARE_API_TOKEN }}
GITHUB_WEBHOOK_SECRET: ${{ secrets.OPS_GITHUB_WEBHOOK_SECRET }}
OPS_LISTEN_TOKEN: ${{ secrets.OPS_LISTEN_TOKEN }}
OPS_EMIT_TOKEN: ${{ secrets.OPS_EMIT_TOKEN }}
run: |
node -e 'process.stdout.write(JSON.stringify({GITHUB_WEBHOOK_SECRET:process.env.GITHUB_WEBHOOK_SECRET||null,OPS_LISTEN_TOKEN:process.env.OPS_LISTEN_TOKEN||null,OPS_EMIT_TOKEN:process.env.OPS_EMIT_TOKEN||null}))' | npx wrangler secret bulk
-49
View File
@@ -1,49 +0,0 @@
name: Deploy Web Studio
on:
push:
branches: [studio]
paths:
- "desktop/frontend-next/**"
- ".github/workflows/deploy-web-studio.yml"
permissions:
contents: read
concurrency:
group: web-studio-production
cancel-in-progress: true
jobs:
deploy:
runs-on: ubuntu-latest
defaults:
run:
working-directory: desktop/frontend-next
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- uses: pnpm/action-setup@ea17c68df8912ef543352723c149a84f56e3d413 # v6.1.0
with:
version: 10
run_install: false
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
with:
node-version: "24"
cache: pnpm
cache-dependency-path: desktop/frontend-next/pnpm-lock.yaml
- name: Install
run: pnpm install --frozen-lockfile
- name: Build
env:
VITE_REMOTE_HOME: https://reasonix.io/remote/
run: pnpm build
- name: Deploy
env:
CLOUDFLARE_API_TOKEN: ${{ secrets.CLOUDFLARE_API_TOKEN }}
CLOUDFLARE_ACCOUNT_ID: ${{ secrets.R2_ACCOUNT_ID }}
run: npx --yes [email protected] deploy --config wrangler.jsonc
+39
View File
@@ -0,0 +1,39 @@
name: Documentation impact guard
on:
pull_request_target:
branches: [main-v2]
types: [opened, synchronize, reopened, edited, ready_for_review]
permissions:
contents: read
pull-requests: read
concurrency:
group: docs-impact-${{ github.event.pull_request.number }}
cancel-in-progress: true
jobs:
docs-impact:
runs-on: ubuntu-latest
steps:
# Execute only the trusted base-branch checker; never run contributor code
# from a pull_request_target workflow.
- uses: actions/checkout@v7
with:
ref: ${{ github.event.pull_request.base.sha }}
- name: Collect PR changed files
env:
GH_TOKEN: ${{ github.token }}
PR_NUMBER: ${{ github.event.pull_request.number }}
REPOSITORY: ${{ github.repository }}
run: |
gh api --paginate "repos/${REPOSITORY}/pulls/${PR_NUMBER}/files" \
--jq '.[].filename' > "$RUNNER_TEMP/docs-impact-files.txt"
- name: Check documentation impact declaration
env:
PR_BODY: ${{ github.event.pull_request.body }}
DOCS_IMPACT_CHANGED_FILES_FILE: ${{ runner.temp }}/docs-impact-files.txt
run: bash scripts/check-docs-impact.sh
+4 -4
View File
@@ -32,7 +32,7 @@ jobs:
environment: e2e-bot
steps:
- name: Acknowledge
uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9
uses: actions/github-script@v9
with:
script: |
await github.rest.reactions.createForIssueComment({
@@ -42,16 +42,16 @@ jobs:
# Default-branch checkout: this is where the harness (cmd/e2ebench), the
# suite, and a run --metrics-capable agent live.
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- uses: actions/checkout@v7
with:
fetch-depth: 0
- uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
- uses: actions/setup-go@v7
with:
go-version-file: go.mod
cache: true
- uses: actions/setup-python@6d4fd22b20465ccc0f7354476588e8df7c9ca0e0 # v7
- uses: actions/setup-python@v7
with:
python-version: '3.12'
-77
View File
@@ -1,77 +0,0 @@
name: Sync in-app feedback
# Turns feedback submitted from Reasonix Studio into issues and reports issue
# state (in progress, fixed in a tag, won't fix, duplicate) back to the worker,
# and mirrors the reporter's replies onto their issue as bot comments.
# Text from the public reaches the script only through the worker's JSON and
# the GitHub API, never through a shell command line or an expression.
on:
schedule:
- cron: '*/10 * * * *'
- cron: '5 3 * * *'
workflow_dispatch:
inputs:
dry_run:
description: Print what would happen without creating issues or writing back
required: false
default: false
type: boolean
full:
description: Re-read every open entry instead of only recently changed issues
required: false
default: false
type: boolean
permissions:
contents: read
issues: write
concurrency:
group: feedback-sync
cancel-in-progress: false
jobs:
sync:
if: github.repository == 'esengine/DeepSeek-Reasonix'
runs-on: ubuntu-latest
timeout-minutes: 8
steps:
- name: Check secret
id: gate
env:
FEEDBACK_ADMIN_TOKEN: ${{ secrets.FEEDBACK_ADMIN_TOKEN }}
run: |
if [ -z "$FEEDBACK_ADMIN_TOKEN" ]; then
echo "::notice::FEEDBACK_ADMIN_TOKEN is not set; feedback sync skipped"
echo "run=false" >> "$GITHUB_OUTPUT"
else
echo "run=true" >> "$GITHUB_OUTPUT"
fi
- if: steps.gate.outputs.run == 'true'
uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
# git tag --contains needs full history; the tag API would cost a request per entry.
fetch-depth: 0
persist-credentials: false
- if: steps.gate.outputs.run == 'true'
uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
with:
node-version: '22'
- name: Sync
if: steps.gate.outputs.run == 'true'
env:
FEEDBACK_ADMIN_TOKEN: ${{ secrets.FEEDBACK_ADMIN_TOKEN }}
# crash.reasonix.io answers runner IPs with a managed challenge; workers.dev serves only the converter's admin paths.
FEEDBACK_BASE_URL: https://reasonix-crash-report.reasonix.workers.dev
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
DRY_RUN: ${{ github.event.inputs.dry_run || 'false' }}
FULL: ${{ github.event.inputs.full || 'false' }}
SCHEDULE: ${{ github.event.schedule }}
run: |
set -euo pipefail
args=()
if [ "$DRY_RUN" = "true" ]; then args+=(--dry-run); fi
# The daily schedule sweeps everything the change window could have missed.
if [ "$FULL" = "true" ] || [ "$SCHEDULE" = "5 3 * * *" ]; then args+=(--full); fi
node scripts/feedback-sync.mjs "${args[@]}"
+3 -3
View File
@@ -20,7 +20,7 @@ jobs:
classify:
runs-on: ubuntu-latest
steps:
- uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9
- uses: actions/github-script@v9
env:
DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}
with:
@@ -35,7 +35,7 @@ jobs:
const body = (issue.body || '').slice(0, 4000);
const system = [
'You categorize GitHub issues for Reasonix, a Go-based AI coding agent with an Electron desktop app and a terminal UI.',
'You categorize GitHub issues for Reasonix, a Go-based AI coding agent with a Wails desktop app and a terminal UI.',
'Pick labels ONLY from these fixed sets. Never invent labels.',
'area (0-2, the affected subsystem):',
' agent: core agent loop / tool-calling / reasoning',
@@ -43,7 +43,7 @@ jobs:
' config: configuration, setup wizard, .toml/.env',
' updater: auto-update, installer, release packaging',
' provider: model providers, model selection/switching',
' desktop: Electron desktop GUI',
' desktop: Wails desktop GUI',
' tui: terminal UI / CLI',
' skills: skills system',
' rendering: terminal rendering / flicker / repaint',
-116
View File
@@ -1,116 +0,0 @@
name: Issue replies
# Posts maintainer replies as github-actions[bot], so a batch of triage answers
# spends the repository's quota instead of a person's content-creation limit.
on:
workflow_dispatch:
inputs:
replies:
description: 'JSON array of {"issue": 123, "body": "...", "close": true | "not_planned"} or {"discussion": 123, "body": "...", "close": "resolved" | "outdated" | "duplicate"}'
required: true
type: string
interval:
description: Seconds between entries
required: false
default: '20'
type: string
permissions:
issues: write
pull-requests: write
discussions: write
concurrency:
group: issue-replies
cancel-in-progress: false
jobs:
reply:
runs-on: win-gpu
timeout-minutes: 180
steps:
- uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9
env:
REPLIES: ${{ inputs.replies }}
INTERVAL: ${{ inputs.interval }}
with:
script: |
const replies = JSON.parse(process.env.REPLIES);
if (!Array.isArray(replies)) throw new Error('replies must be a JSON array');
const { owner, repo } = context.repo;
const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
const failed = [];
// Only a rate limit is worth waiting out; any other refusal is final.
const rateLimited = (error) => {
const headers = error.response?.headers ?? {};
return error.status === 429 ||
(error.status === 403 && (headers['retry-after'] !== undefined || headers['x-ratelimit-remaining'] === '0'));
};
// Discussions have no REST comment API; they are addressed by node id.
const replyToDiscussion = async (number, body) => {
const q = await github.graphql(
`query($owner:String!,$repo:String!,$n:Int!){repository(owner:$owner,name:$repo){discussion(number:$n){id comments(last:100){nodes{body author{login}}}}}}`,
{ owner, repo, n: number });
const d = q.repository.discussion;
if (!d) throw new Error('no such discussion');
if (d.comments.nodes.some((c) => c.author?.login === 'github-actions' && c.body === body)) return false;
await github.graphql(`mutation($id:ID!,$body:String!){addDiscussionComment(input:{discussionId:$id,body:$body}){comment{id}}}`, { id: d.id, body });
return true;
};
const closeReasons = { resolved: 'RESOLVED', outdated: 'OUTDATED', duplicate: 'DUPLICATE' };
const closeDiscussion = async (number, reason) => {
const q = await github.graphql(
`query($owner:String!,$repo:String!,$n:Int!){repository(owner:$owner,name:$repo){discussion(number:$n){id closed}}}`,
{ owner, repo, n: number });
const d = q.repository.discussion;
if (d.closed) return;
await github.graphql(`mutation($id:ID!,$r:DiscussionCloseReason!){closeDiscussion(input:{discussionId:$id,reason:$r}){discussion{id}}}`, { id: d.id, r: reason });
};
for (const { issue, discussion, body, close } of replies) {
if (Number.isInteger(discussion) && typeof body === 'string' && body.trim() !== '') {
try {
const posted = await replyToDiscussion(discussion, body);
if (close !== undefined && close !== false) {
const reason = closeReasons[close];
if (!reason) throw new Error(`unknown close reason ${close}`);
await closeDiscussion(discussion, reason);
}
core.info(`discussion #${discussion}: ${posted ? 'replied' : 'already replied'}${close ? ', closed' : ''}`);
} catch (error) {
failed.push(`discussion ${discussion}: ${error.message}`);
}
await sleep(Math.max(1, Number(process.env.INTERVAL) || 20) * 1000);
continue;
}
if (!Number.isInteger(issue) || typeof body !== 'string' || body.trim() === '') {
failed.push(`${issue}: malformed entry`);
continue;
}
try {
const comments = await github.paginate(github.rest.issues.listComments, { owner, repo, issue_number: issue, per_page: 100 });
const posted = comments.some((c) => c.user?.type === 'Bot' && c.body === body);
if (!posted) {
for (let attempt = 0; ; attempt++) {
try {
await github.rest.issues.createComment({ owner, repo, issue_number: issue, body });
break;
} catch (error) {
if (attempt >= 5 || !rateLimited(error)) throw error;
core.warning(`#${issue}: ${error.message}; retrying in 2 minutes`);
await sleep(120_000);
}
}
}
if (close === true || close === 'not_planned') {
const { data } = await github.rest.issues.get({ owner, repo, issue_number: issue });
if (data.state === 'open') {
await github.rest.issues.update({ owner, repo, issue_number: issue, state: 'closed', state_reason: close === 'not_planned' ? 'not_planned' : 'completed' });
}
}
core.info(`#${issue}: ${posted ? 'already replied' : 'replied'}${close ? ', closed' : ''}`);
} catch (error) {
failed.push(`${issue}: ${error.message}`);
}
await sleep(Math.max(1, Number(process.env.INTERVAL) || 20) * 1000);
}
if (failed.length) core.setFailed(`not done:\n${failed.join('\n')}`);
+25 -26
View File
@@ -2,8 +2,7 @@ name: Label issue version
# Issue forms can't let a reporter set a label directly (that needs triage rights),
# so the bug/feature forms carry a "Version line" dropdown and this workflow reads
# the submitted value and applies the matching label on their behalf:
# 0.x -> v1 (TypeScript), 1.x -> v2 (main-v2), 2.x -> v3 (studio).
# the submitted value and applies the matching v1/v2 label on their behalf.
on:
issues:
types: [opened, edited]
@@ -19,46 +18,46 @@ jobs:
label:
runs-on: ubuntu-latest
steps:
- uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9
- uses: actions/github-script@v9
with:
script: |
const body = context.payload.issue.body || '';
const sectionOf = (name) =>
body.split(/^###\s+/m).find((s) => new RegExp(`^${name}`, 'i').test(s)) || '';
const labelOf = (major) => ({ 0: 'v1', 1: 'v2', 2: 'v3' })[major];
const m = sectionOf('Version line').match(/\b([012])\.x\b/);
const m = sectionOf('Version line').match(/\bv([12])\b/i);
if (!m) {
core.info('No version line found; nothing to label.');
return;
}
let choice = labelOf(Number(m[1]));
let choice = 'v' + m[1];
// The dropdown is a choice, and its first entry gets picked by
// reporters who never looked. `Exact version` is pasted from
// `reasonix --version` or Studio, so when it parses it is the better
// evidence. Anything that does not parse (a commit sha, a branch
// name) or names no known line leaves the dropdown to decide.
// The dropdown is a choice, and v2 sits first, so it gets picked by
// reporters who are actually on 0.x. `Exact version` is pasted from
// `reasonix --version`, so when it parses it is the better evidence:
// 0.x is the TypeScript line, 1.x the Go rewrite. Anything that does
// not parse (a commit sha, "main-v2") leaves the dropdown to decide.
const exact = sectionOf('Exact version').match(/\b(\d+)\.(\d+)\.(\d+)/);
const fromVersion = exact && labelOf(Number(exact[1]));
if (fromVersion && fromVersion !== choice) {
core.info(`Version line says ${choice} but ${exact[0]} says ${fromVersion}; trusting the version.`);
choice = fromVersion;
if (exact) {
const fromVersion = Number(exact[1]) === 0 ? 'v1' : 'v2';
if (fromVersion !== choice) {
core.info(`Version line says ${choice} but ${exact[0]} says ${fromVersion}; trusting the version.`);
choice = fromVersion;
}
}
const other = choice === 'v2' ? 'v1' : 'v2';
await github.rest.issues.addLabels({
...context.repo,
issue_number: context.issue.number,
labels: [choice],
});
// Keep the line labels mutually exclusive in case an edit flipped the choice.
for (const other of ['v1', 'v2', 'v3'].filter((l) => l !== choice)) {
try {
await github.rest.issues.removeLabel({
...context.repo,
issue_number: context.issue.number,
name: other,
});
} catch (e) {
core.info(`No ${other} label to remove.`);
}
// Keep v1/v2 mutually exclusive in case an edit flipped the choice.
try {
await github.rest.issues.removeLabel({
...context.repo,
issue_number: context.issue.number,
name: other,
});
} catch (e) {
core.info(`No ${other} label to remove.`);
}
+39 -24
View File
@@ -1,32 +1,47 @@
name: Deploy website
name: Deploy site
# GitHub requires workflow_dispatch to be registered on the default branch.
# The production implementation is the same workflow path on website.
on:
push:
branches: [main-v2]
paths: ['site/**', 'release-notes/**', '.github/workflows/pages.yml']
workflow_dispatch:
inputs:
release_version:
description: "Optional product release version associated with this publication"
required: false
default: ""
type: string
release_request:
description: "Optional release workflow correlation ID"
required: false
default: ""
type: string
studio_version:
description: "Studio version that must be present in the generated changelog"
required: false
default: ""
type: string
permissions: {}
permissions:
contents: read
pages: write
id-token: write
concurrency:
group: pages
cancel-in-progress: false
jobs:
require-website:
build:
runs-on: ubuntu-latest
steps:
- run: |
echo "::error::Deploy the website with --ref website; production is restricted to that branch."
exit 1
- uses: actions/checkout@v7
- uses: actions/setup-node@v7
with:
node-version: '22'
cache: npm
cache-dependency-path: site/package-lock.json
- name: Build
working-directory: site
env:
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
run: |
npm ci
npm run build
- uses: actions/upload-pages-artifact@v5
with:
path: site/dist
deploy:
needs: build
runs-on: ubuntu-latest
environment:
name: github-pages
url: ${{ steps.deploy.outputs.page_url }}
steps:
- id: deploy
uses: actions/deploy-pages@v5
-27
View File
@@ -1,27 +0,0 @@
name: PR author gate
# Pull requests are limited to maintainers and the logins in .github/pr-authors.txt.
# This runs in the base repository's context, checks out the base branch only, and
# never executes pull-request code; the allowlist it reads is the base branch's.
on:
pull_request_target:
types: [opened]
permissions:
contents: read
pull-requests: write
concurrency:
group: pr-author-gate-${{ github.event.pull_request.number }}
cancel-in-progress: false
jobs:
gate:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
persist-credentials: false
- run: node scripts/pr-author-gate.mjs
env:
GH_TOKEN: ${{ github.token }}
+1 -1
View File
@@ -23,6 +23,6 @@ jobs:
label:
runs-on: ubuntu-latest
steps:
- uses: actions/labeler@bf12e9b00b37c5c0ca2b87b79b2daf7891dbda13 # v7
- uses: actions/labeler@v7
with:
sync-labels: false
+1 -1
View File
@@ -20,7 +20,7 @@ jobs:
label:
runs-on: ubuntu-latest
steps:
- uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9
- uses: actions/github-script@v9
with:
script: |
const base = context.payload.pull_request.base.ref;
+101
View File
@@ -0,0 +1,101 @@
name: Prepare release
on:
workflow_dispatch:
inputs:
version:
description: "Official version without a tag prefix (for example 1.20.0)"
required: true
type: string
permissions:
contents: write
pull-requests: write
jobs:
draft:
name: generate reviewable bilingual draft
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
with:
fetch-depth: 0
- uses: actions/setup-node@v7
with:
node-version: "22"
- name: Select the review branch
env:
GH_TOKEN: ${{ github.token }}
VERSION: ${{ inputs.version }}
run: |
set -euo pipefail
git config user.name "github-actions[bot]"
git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
git fetch origin main-v2
if [[ ! "$VERSION" =~ ^(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)$ ]]; then
echo "::error::version must be MAJOR.MINOR.PATCH"
exit 1
fi
branch="release-notes/v${VERSION}"
if git ls-remote --exit-code --heads origin "$branch" >/dev/null 2>&1; then
git fetch origin "refs/heads/$branch:refs/remotes/origin/$branch"
git checkout -b "$branch" --track "origin/$branch"
git merge --no-edit origin/main-v2
else
git checkout -b "$branch"
fi
echo "RELEASE_NOTES_BRANCH=$branch" >> "$GITHUB_ENV"
- name: Generate and validate release notes
env:
DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}
DEEPSEEK_MODEL: deepseek-v4-pro
GH_TOKEN: ${{ github.token }}
VERSION: ${{ inputs.version }}
run: |
set -euo pipefail
node scripts/generate-release-notes.mjs --version "$VERSION"
node scripts/release-notes.mjs validate
node --test scripts/release-notes.test.mjs
node scripts/release-notes.mjs render --version "$VERSION" --output /tmp/release-notes.md
{
echo "## Review draft for v${VERSION#v}"
echo
cat /tmp/release-notes.md
} >> "$GITHUB_STEP_SUMMARY"
- name: Open or update the review PR
env:
GH_TOKEN: ${{ github.token }}
VERSION: ${{ inputs.version }}
run: |
set -euo pipefail
git add release-notes/releases.json
if ! git diff --cached --quiet; then
git commit -m "docs(release): prepare v${VERSION#v} notes" \
-m "Summary:
Generate a bilingual, product-focused draft from merged pull request metadata. Reuse the selected release-bound PR when one is available.
Verification:
Validate the catalog, citations, bilingual fields, and rendered GitHub release notes before committing."
fi
git push -u origin "$RELEASE_NOTES_BRANCH"
if ! gh pr list --head "$RELEASE_NOTES_BRANCH" --state open --json number --jq 'length > 0' | grep -q true; then
if ! gh pr create \
--base main-v2 \
--head "$RELEASE_NOTES_BRANCH" \
--title "docs(release): 准备 v${VERSION#v} 中英更新日志" \
--body-file /tmp/release-notes.md; then
echo "::warning::GitHub Actions could not open the PR; the reviewed branch is preserved."
{
echo "## Manual PR handoff"
echo
echo '```sh'
echo "gh pr create --repo ${{ github.repository }} --base main-v2 --head $RELEASE_NOTES_BRANCH --fill"
echo '```'
} >> "$GITHUB_STEP_SUMMARY"
fi
fi
File diff suppressed because it is too large. Load diff
+26
View File
@@ -0,0 +1,26 @@
name: Release notes
on:
pull_request:
paths:
- "release-notes/**"
- "scripts/release-notes*.mjs"
- "site/src/pages/changelog/**"
- "site/src/components/ChangelogPage.astro"
- "desktop/frontend/src/App.tsx"
- "desktop/frontend/src/components/SettingsPanel.tsx"
- "desktop/frontend/src/components/UpdateBanner.tsx"
permissions:
contents: read
jobs:
validate:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: actions/setup-node@v7
with:
node-version: "22"
- run: node scripts/release-notes.mjs validate
- run: node --test scripts/release-notes.test.mjs
+206
View File
@@ -0,0 +1,206 @@
name: Release npm
# npm line. Stable releases are called by release-stable.yml after its single
# GitHub environment approval. Direct prerelease tag publication is disabled;
# manual workflow_dispatch recovers only an approved Stable version from an
# existing tag. Historical canary and next package identities remain readable,
# but normal publication advances only the official release.
on:
workflow_dispatch:
inputs:
channel:
description: "Standalone npm recovery channel"
required: true
default: stable
type: choice
options:
- stable
base_version:
description: "Stable version to recover exactly"
required: true
type: string
tag:
description: "stable: existing npm tag to publish (for example npm-v1.18.0)"
required: false
type: string
workflow_call:
inputs:
channel:
description: "Publish channel selected by the approved release orchestrator"
required: true
type: string
base_version:
description: "Version to publish"
required: true
type: string
tag:
description: "Existing npm tag to check out for stable publication"
required: false
default: ""
type: string
approved_cli_tag:
description: "Stable CLI tag recorded by the approved orchestrator"
required: true
type: string
approved_sha:
description: "Immutable commit recorded by the approved orchestrator"
required: true
type: string
orchestrated:
description: "True only when called by an approved release orchestrator"
required: false
default: false
type: boolean
orchestrator:
description: "Trusted release orchestrator (legacy Preview calls remain readable)"
required: false
default: stable
type: string
preview_number:
description: "Legacy Preview ordinal for old workflow-call compatibility"
required: false
default: ""
type: string
permissions:
contents: read
concurrency:
# Serialize every publisher for one npm dist-tag. Historical canary calls
# remain ordered even though normal publication now uses Stable only.
group: release-npm-${{ inputs.channel || 'next' }}
cancel-in-progress: false
jobs:
orchestration-guard:
name: verify approved orchestrator
if: ${{ inputs.orchestrated }}
runs-on: ubuntu-latest
permissions:
contents: read
steps:
- uses: actions/checkout@v7
with:
fetch-depth: 0
ref: ${{ github.sha }}
- name: Verify caller and approved release ref
env:
ACTUAL_CALLER_WORKFLOW_REF: ${{ github.workflow_ref }}
EXPECTED_CALLER_WORKFLOW_REF: ${{ format('{0}/.github/workflows/release-{1}.yml@{2}', github.repository, inputs.orchestrator, github.ref) }}
CALLER_EVENT_NAME: ${{ github.event_name }}
CALLER_REF: ${{ github.ref }}
CALLER_REF_PROTECTED: ${{ github.ref_protected }}
CALLER_WORKFLOW_SHA: ${{ github.workflow_sha }}
CALLER_SHA: ${{ github.sha }}
APPROVED_CLI_TAG: ${{ inputs.approved_cli_tag }}
APPROVED_SHA: ${{ inputs.approved_sha }}
APPROVED_CHANNEL: ${{ inputs.orchestrator }}
RELEASE_TAG: ${{ inputs.approved_cli_tag }}
VERIFY_RELEASE_CHECKOUT: false
run: |
bash scripts/verify-release-authorization.sh
bash scripts/verify-release-tag.sh
release-gate:
name: approve standalone npm release
if: ${{ !inputs.orchestrated }}
runs-on: ubuntu-latest
environment: release
steps:
- env:
RELEASE_TAG: ${{ inputs.tag || github.ref_name }}
run: echo "Approved standalone npm release $RELEASE_TAG"
cache-guard:
name: cache hit guard
needs: [orchestration-guard, release-gate]
if: ${{ always() && !cancelled() && ((inputs.orchestrated && needs.orchestration-guard.result == 'success') || (!inputs.orchestrated && needs.release-gate.result == 'success')) }}
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
with:
ref: ${{ inputs.approved_sha || inputs.tag || github.ref }}
- uses: actions/setup-go@v7
with:
go-version-file: go.mod
cache: true
- run: ./scripts/cache-guard.sh
- name: Verify embedded documentation identity
env:
DOCS_BUILD_VERSION: v${{ inputs.base_version }}
DOCS_SOURCE_REVISION: ${{ inputs.approved_sha }}
run: |
if [ ! -f scripts/verify-embedded-docs.sh ]; then
echo "Legacy candidate predates the embedded docs contract; skipping."
exit 0
fi
revision="$DOCS_SOURCE_REVISION"
if [ -z "$revision" ]; then revision="$(git rev-parse HEAD)"; fi
bash scripts/verify-embedded-docs.sh "$DOCS_BUILD_VERSION" "$revision"
npm:
name: publish npm packages
needs: cache-guard
if: ${{ always() && !cancelled() && needs.cache-guard.result == 'success' }}
runs-on: ubuntu-latest
# Orchestrated releases have already passed their GitHub environment
# approval. Direct prereleases and manual Stable recovery pass release-gate.
steps:
- uses: actions/checkout@v7
with:
ref: ${{ inputs.approved_sha || inputs.tag || github.ref }}
- name: Load approved standalone recovery control plane
if: ${{ !inputs.orchestrated }}
env:
RECOVERY_CONTROL_SHA: ${{ github.workflow_sha }}
run: |
set -euo pipefail
git fetch --no-tags --depth=1 origin "$RECOVERY_CONTROL_SHA"
git restore --source="$RECOVERY_CONTROL_SHA" -- \
npm/publish.mjs \
scripts/finalize-npm-official-release.mjs
- uses: actions/setup-go@v7
with:
go-version-file: go.mod
cache: true
- uses: actions/setup-node@v7
with:
node-version: '22'
registry-url: 'https://registry.npmjs.org'
# Stable publication uses the exact npm-v* tag. build.mjs strips the
# leading `npm-`/`v`; historical Preview inputs remain only for old
# workflow-call compatibility and are not reachable from a public entry.
- name: Resolve version
id: ver
env:
EVENT_NAME: ${{ github.event_name }}
IN_ORCHESTRATED: ${{ inputs.orchestrated }}
IN_CHANNEL: ${{ inputs.channel }}
IN_BASE_VERSION: ${{ inputs.base_version }}
IN_TAG: ${{ inputs.tag }}
REF_NAME: ${{ github.ref_name }}
RUN_NUMBER: ${{ github.run_number }}
IN_PREVIEW_NUMBER: ${{ inputs.preview_number }}
run: bash scripts/resolve-npm-release.sh
- name: Revalidate approved release ref
if: ${{ inputs.orchestrated }}
env:
RELEASE_TAG: ${{ inputs.approved_cli_tag }}
APPROVED_SHA: ${{ inputs.approved_sha }}
run: bash scripts/verify-release-tag.sh
- name: Publish or recover immutable npm packages
env:
VERSION_ARG: ${{ steps.ver.outputs.arg }}
NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
run: node npm/build.mjs "$VERSION_ARG" --publish
- name: Align legacy aliases with the official release
if: ${{ inputs.channel == 'stable' }}
env:
VERSION: ${{ inputs.base_version }}
APPROVED_SHA: ${{ inputs.approved_sha }}
NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
run: |
set -euo pipefail
candidate="$APPROVED_SHA"
if [ -z "$candidate" ]; then candidate="$(git rev-parse HEAD)"; fi
EXPECTED_SHA="$candidate" node scripts/finalize-npm-official-release.mjs "$VERSION"
@@ -0,0 +1,42 @@
name: Relay stable release
run-name: Relay stable release ${{ github.ref_name }}
# The tag event itself has a tag-shaped origin. Relay the immutable tag name to
# release-stable.yml on protected main-v2 so the production SignPath policy can
# allow exactly one branch name without trusting wildcard tag-like branches.
on:
push:
tags:
- 'v*'
- '!v*-*'
permissions:
actions: write
contents: read
jobs:
relay:
name: dispatch protected stable control plane
runs-on: ubuntu-latest
steps:
- name: Dispatch release-stable.yml on main-v2
uses: actions/github-script@v9
env:
RELEASE_TAG: ${{ github.ref_name }}
CONTROL_PLANE_REF: ${{ github.event.repository.default_branch }}
with:
script: |
if (process.env.CONTROL_PLANE_REF !== 'main-v2') {
core.setFailed(`Unexpected default branch: ${process.env.CONTROL_PLANE_REF}`);
return;
}
await github.rest.actions.createWorkflowDispatch({
owner: context.repo.owner,
repo: context.repo.repo,
workflow_id: 'release-stable.yml',
ref: process.env.CONTROL_PLANE_REF,
inputs: {
tag: process.env.RELEASE_TAG,
allow_recovery: 'false',
},
});
+278
View File
@@ -0,0 +1,278 @@
name: Release stable
run-name: Release stable ${{ inputs.tag || github.ref_name }}
# Stable publication has exactly one GitHub environment gate. Push the CLI,
# npm, and desktop tags atomically; release-stable-trigger.yml relays the vX.Y.Z
# tag to this workflow on protected main-v2. Preflight verifies that all three
# tags point to the reviewed Notes candidate on main-v2 history, and one GitHub
# approval releases every surface. After approval, a zero-publication Windows
# preflight verifies both architectures and signing stages before publication.
# Keeping the control-plane ref on main-v2 lets SignPath restrict
# production signing to that one protected origin instead of trusting wildcard
# tag-like branch names. Manual recovery uses the same fixed control plane while
# preserving an older tagged candidate on main-v2 history.
on:
workflow_dispatch:
inputs:
tag:
description: "Existing stable CLI tag to recover (for example v1.18.0)"
required: true
type: string
publish_cli:
description: "Recover the CLI/Homebrew channel"
required: false
default: true
type: boolean
publish_npm:
description: "Recover the npm channel"
required: false
default: true
type: boolean
publish_desktop:
description: "Recover the Desktop/R2 channel"
required: false
default: true
type: boolean
allow_recovery:
description: "Allow an existing stable tag on main-v2 history (manual recovery only)"
required: false
default: true
type: boolean
concurrency:
group: stable-release-${{ inputs.tag || github.ref_name }}
cancel-in-progress: false
# Reusable release workflows can only reduce caller permissions, so the
# orchestrator grants the union needed by CLI/Desktop publication.
permissions:
actions: write
contents: write
jobs:
preflight:
name: validate stable release set
runs-on: ubuntu-latest
permissions:
actions: read
contents: read
outputs:
version: ${{ steps.release.outputs.version }}
cli_tag: ${{ steps.release.outputs.cli_tag }}
npm_tag: ${{ steps.release.outputs.npm_tag }}
desktop_tag: ${{ steps.release.outputs.desktop_tag }}
sha: ${{ steps.release.outputs.sha }}
steps:
- uses: actions/checkout@v7
with:
fetch-depth: 0
# A recovery dispatch uses the fixed workflow/scripts from protected
# main-v2. Publishers still check out the immutable approved tag SHA.
ref: ${{ github.sha }}
- uses: actions/setup-go@v7
with:
go-version-file: go.mod
cache: true
- uses: actions/setup-node@v7
with:
node-version: "22"
- name: Resolve stable release
id: release
env:
RELEASE_TAG: ${{ inputs.tag || github.ref_name }}
ALLOW_STABLE_RECOVERY: ${{ inputs.allow_recovery }}
run: bash scripts/resolve-stable-release.sh
- name: Revalidate normal release candidate and exact push CI
if: ${{ !inputs.allow_recovery }}
env:
GH_TOKEN: ${{ github.token }}
RELEASE_REPOSITORY: ${{ github.repository }}
RELEASE_CI_WAIT_SECONDS: 60
RELEASE_VERSION: ${{ steps.release.outputs.version }}
RELEASE_SHA: ${{ steps.release.outputs.sha }}
run: |
bash scripts/validate-stable-candidate.sh "$RELEASE_VERSION" "$RELEASE_SHA"
bash scripts/verify-release-push-ci.sh "$RELEASE_SHA"
- name: Validate reviewed release notes
run: node scripts/release-notes.mjs render --version "${{ steps.release.outputs.cli_tag }}" --output /tmp/release-notes.md
# Recovery builds deliberately check out the immutable tagged candidate,
# which can predate its reviewed release-note entry. Carry the exact file
# validated by this protected control-plane job into both publishers.
- name: Upload reviewed release notes
uses: actions/upload-artifact@v7
with:
name: orchestrator-reviewed-release-notes
path: /tmp/release-notes.md
if-no-files-found: error
retention-days: 1
- name: Cache hit guard
run: ./scripts/cache-guard.sh
authorize:
name: approve stable release
needs: preflight
runs-on: ubuntu-latest
environment: release
permissions:
contents: read
outputs:
version: ${{ steps.approved.outputs.version }}
cli_tag: ${{ steps.approved.outputs.cli_tag }}
npm_tag: ${{ steps.approved.outputs.npm_tag }}
desktop_tag: ${{ steps.approved.outputs.desktop_tag }}
sha: ${{ steps.approved.outputs.sha }}
steps:
- name: Record approved release
id: approved
env:
VERSION: ${{ needs.preflight.outputs.version }}
CLI_TAG: ${{ needs.preflight.outputs.cli_tag }}
NPM_TAG: ${{ needs.preflight.outputs.npm_tag }}
DESKTOP_TAG: ${{ needs.preflight.outputs.desktop_tag }}
RELEASE_SHA: ${{ needs.preflight.outputs.sha }}
run: |
{
echo "version=$VERSION"
echo "cli_tag=$CLI_TAG"
echo "npm_tag=$NPM_TAG"
echo "desktop_tag=$DESKTOP_TAG"
echo "sha=$RELEASE_SHA"
} >> "$GITHUB_OUTPUT"
echo "Approved stable release $VERSION at $RELEASE_SHA"
cli:
name: publish CLI and Homebrew
needs: [authorize, signpath-preflight]
if: ${{ always() && !cancelled() && needs.authorize.result == 'success' && (needs.signpath-preflight.result == 'success' || needs.signpath-preflight.result == 'skipped') && (github.event_name != 'workflow_dispatch' || inputs.publish_cli) }}
uses: ./.github/workflows/release.yml
with:
tag: ${{ needs.authorize.outputs.cli_tag }}
approved_cli_tag: ${{ needs.authorize.outputs.cli_tag }}
approved_sha: ${{ needs.authorize.outputs.sha }}
orchestrated: true
secrets: inherit
npm:
name: publish npm
needs: [authorize, signpath-preflight]
if: ${{ always() && !cancelled() && needs.authorize.result == 'success' && (needs.signpath-preflight.result == 'success' || needs.signpath-preflight.result == 'skipped') && (github.event_name != 'workflow_dispatch' || inputs.publish_npm) }}
uses: ./.github/workflows/release-npm.yml
with:
channel: stable
base_version: ${{ needs.authorize.outputs.version }}
tag: ${{ needs.authorize.outputs.npm_tag }}
approved_cli_tag: ${{ needs.authorize.outputs.cli_tag }}
approved_sha: ${{ needs.authorize.outputs.sha }}
orchestrated: true
secrets: inherit
signpath-preflight:
name: verify stable SignPath control plane
needs: authorize
if: ${{ github.event_name != 'workflow_dispatch' || inputs.publish_desktop }}
uses: ./.github/workflows/release-desktop.yml
with:
channel: stable
tag: ${{ needs.authorize.outputs.desktop_tag }}
approved_cli_tag: ${{ needs.authorize.outputs.cli_tag }}
approved_sha: ${{ needs.authorize.outputs.sha }}
orchestrated: true
signing_preflight: true
secrets: inherit
desktop:
name: publish desktop
needs: [authorize, signpath-preflight]
if: ${{ always() && !cancelled() && needs.authorize.result == 'success' && needs.signpath-preflight.result == 'success' && (github.event_name != 'workflow_dispatch' || inputs.publish_desktop) }}
uses: ./.github/workflows/release-desktop.yml
with:
channel: stable
tag: ${{ needs.authorize.outputs.desktop_tag }}
approved_cli_tag: ${{ needs.authorize.outputs.cli_tag }}
approved_sha: ${{ needs.authorize.outputs.sha }}
orchestrated: true
signing_preflight_verified: true
secrets: inherit
postflight:
name: verify stable release artifacts
needs: [authorize, cli, npm, desktop]
if: ${{ always() && !cancelled() }}
runs-on: ubuntu-latest
permissions:
actions: write
contents: write
steps:
- name: Require every publisher to succeed
env:
PUBLISH_CLI: ${{ github.event_name != 'workflow_dispatch' || inputs.publish_cli }}
PUBLISH_NPM: ${{ github.event_name != 'workflow_dispatch' || inputs.publish_npm }}
PUBLISH_DESKTOP: ${{ github.event_name != 'workflow_dispatch' || inputs.publish_desktop }}
CLI_RESULT: ${{ needs.cli.result }}
NPM_RESULT: ${{ needs.npm.result }}
DESKTOP_RESULT: ${{ needs.desktop.result }}
run: |
set -euo pipefail
for channel in cli npm desktop; do
selected_var="PUBLISH_${channel^^}"
result_var="${channel^^}_RESULT"
selected="${!selected_var}"
result="${!result_var}"
if [ "$selected" != "true" ]; then
echo "$channel recovery skipped; public postflight will still verify it"
continue
fi
if [ "$result" != "success" ]; then
echo "::error::$channel stable publisher result is $result, expected success"
exit 1
fi
done
- uses: actions/checkout@v7
with:
# Postflight belongs to the trusted control plane, not the old build
# candidate, which may predate this verifier.
ref: ${{ github.sha }}
- uses: actions/setup-node@v7
with:
node-version: "22"
- name: Verify public artifacts and npm latest
env:
GH_TOKEN: ${{ github.token }}
RELEASE_REPOSITORY: ${{ github.repository }}
RELEASE_VERSION: ${{ needs.authorize.outputs.version }}
CLI_TAG: ${{ needs.authorize.outputs.cli_tag }}
DESKTOP_TAG: ${{ needs.authorize.outputs.desktop_tag }}
run: bash scripts/verify-stable-release-artifacts.sh
- name: Publish exact Stable release record
env:
GH_TOKEN: ${{ github.token }}
VERSION: ${{ needs.authorize.outputs.version }}
CLI_TAG: ${{ needs.authorize.outputs.cli_tag }}
RELEASE_SHA: ${{ needs.authorize.outputs.sha }}
run: |
set -euo pipefail
if ! node -e '
const catalog = require("./release-notes/releases.json");
const release = catalog.releases.find((item) => item.version === process.env.VERSION);
process.exit(release?.status === "reviewed" ? 0 : 1);
'; then
echo "Legacy Stable notes do not require a publication marker"
exit 0
fi
node scripts/release-event.mjs generate \
--version "$VERSION" --sha "$RELEASE_SHA" \
--published-at "$(gh api "repos/${{ github.repository }}/releases/tags/$CLI_TAG" --jq .published_at)" \
--output /tmp/release-event.json
existing="$(mktemp -d)"
if gh release download "$CLI_TAG" --pattern release-event.json --dir "$existing" 2>/dev/null; then
cmp -s /tmp/release-event.json "$existing/release-event.json" || {
echo "::error::published release-event.json differs from the approved Stable event"
exit 1
}
else
gh release upload "$CLI_TAG" /tmp/release-event.json
fi
- name: Refresh public changelog
env:
GH_TOKEN: ${{ github.token }}
run: gh workflow run pages.yml --ref main-v2
File diff suppressed because it is too large. Load diff
@@ -0,0 +1,62 @@
name: Request verification on released issues
# A PR that writes "Fixes #N" auto-closes its issue on merge. One that writes
# "Refs"/"Related"/"Addresses" does not, so the report stays open waiting for a
# follow-up nobody remembers to send after the release. This sends it.
on:
workflow_dispatch:
inputs:
tag:
description: Release tag to announce (e.g. desktop-v1.18.0)
required: true
type: string
dry_run:
description: Print the targets without commenting
required: false
default: true
type: boolean
release:
types: [published]
push:
branches:
- main-v2
paths:
- '.github/workflows/release-verify-issues.yml'
- 'scripts/release-verify-issues.mjs'
- 'scripts/release-verify-issues.test.mjs'
permissions:
contents: read
issues: write
concurrency:
group: release-verify-issues-${{ github.event.inputs.tag || github.event.release.tag_name || github.ref }}
cancel-in-progress: false
jobs:
notify:
if: github.repository == 'esengine/DeepSeek-Reasonix'
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
with:
fetch-depth: 0
- uses: actions/setup-node@v7
with:
node-version: '22'
- name: Test notifier
run: node --test scripts/release-verify-issues.test.mjs
# A push that only touched the script has no release to announce; the test
# above is the whole point of that trigger.
- name: Request verification
if: github.event_name != 'push'
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
TAG: ${{ github.event.inputs.tag || github.event.release.tag_name }}
DRY_RUN: ${{ github.event.inputs.dry_run || 'false' }}
run: |
set -euo pipefail
args=(--tag "$TAG")
if [ "$DRY_RUN" = "true" ]; then args+=(--dry-run); fi
node scripts/release-verify-issues.mjs "${args[@]}"
+581
View File
@@ -0,0 +1,581 @@
name: Release
# Native CLI binary line. Official releases are called by the protected Stable
# orchestrator. Manual dispatch remains available only for official recovery;
# historical Preview inputs below are workflow-call compatibility, not a public
# publication entrypoint.
on:
workflow_dispatch:
inputs:
channel:
description: "Standalone CLI recovery channel"
required: true
default: stable
type: choice
options: [stable]
tag:
description: "Existing official CLI tag (for example v1.18.0)"
required: true
type: string
workflow_call:
inputs:
channel:
description: "Native CLI release channel selected by the approved orchestrator"
required: false
default: stable
type: string
tag:
description: "Existing CLI tag selected by the approved release orchestrator"
required: true
type: string
approved_cli_tag:
description: "Stable CLI tag recorded by the approved orchestrator"
required: true
type: string
approved_sha:
description: "Immutable commit recorded by the approved orchestrator"
required: true
type: string
orchestrated:
description: "True only when called by an approved release orchestrator"
required: false
default: false
type: boolean
orchestrator:
description: "Trusted release orchestrator (legacy Preview calls remain readable)"
required: false
default: stable
type: string
allow_preview_recovery:
description: "Legacy compatibility for already-created Preview runs"
required: false
default: false
type: boolean
permissions:
contents: write # create the release and upload archives
concurrency:
# A channel pointer is a monotonic public state machine. Serialize all
# publishers for the same channel so an older recovery run cannot pass its
# read-before-write window after a newer release has published.
group: release-cli-${{ inputs.channel || 'stable' }}
cancel-in-progress: false
jobs:
resolve:
name: resolve CLI release
runs-on: ubuntu-latest
outputs:
tag: ${{ steps.release.outputs.tag }}
version: ${{ steps.release.outputs.version }}
base_version: ${{ steps.release.outputs.base_version }}
notes_version: ${{ steps.release.outputs.notes_version }}
channel: ${{ steps.release.outputs.channel }}
prerelease: ${{ steps.release.outputs.prerelease }}
sha: ${{ steps.candidate.outputs.sha }}
steps:
- uses: actions/checkout@v7
with:
fetch-depth: 0
ref: ${{ github.sha }}
- name: Resolve channel and tag
id: release
env:
EVENT_NAME: ${{ github.event_name }}
IN_ORCHESTRATED: ${{ inputs.orchestrated }}
IN_CHANNEL: ${{ inputs.channel }}
IN_TAG: ${{ inputs.tag }}
REF_NAME: ${{ github.ref_name }}
CALLER_REF: ${{ github.ref }}
CALLER_REF_PROTECTED: ${{ github.ref_protected }}
run: bash scripts/resolve-cli-release.sh
- name: Record immutable candidate
id: candidate
env:
RELEASE_TAG: ${{ steps.release.outputs.tag }}
RELEASE_CHANNEL: ${{ steps.release.outputs.channel }}
IN_ORCHESTRATED: ${{ inputs.orchestrated }}
IN_ORCHESTRATOR: ${{ inputs.orchestrator }}
ALLOW_PREVIEW_RECOVERY: ${{ inputs.allow_preview_recovery }}
run: |
set -euo pipefail
git fetch origin main-v2
sha="$(git rev-parse "$RELEASE_TAG^{commit}")"
if ! git merge-base --is-ancestor "$sha" origin/main-v2; then
echo "::error::$RELEASE_TAG points to $sha, which is not on main-v2 history"
exit 1
fi
if [ "$ALLOW_PREVIEW_RECOVERY" = "true" ]; then
if [ "$IN_ORCHESTRATED" != "true" ] || [ "$IN_ORCHESTRATOR" != "preview" ] || [ "$RELEASE_CHANNEL" != "preview" ]; then
echo "::error::Preview recovery requires the approved Preview orchestrator"
exit 1
fi
elif [ "$RELEASE_CHANNEL" = "preview" ] && [ "$sha" != "$(git rev-parse origin/main-v2)" ]; then
echo "::error::CLI Preview must tag current main-v2; $RELEASE_TAG points to $sha"
exit 1
fi
echo "sha=$sha" >> "$GITHUB_OUTPUT"
- name: Verify existing protected tag
env:
RELEASE_TAG: ${{ steps.release.outputs.tag }}
APPROVED_SHA: ${{ steps.candidate.outputs.sha }}
VERIFY_RELEASE_CHECKOUT: false
run: bash scripts/verify-release-tag.sh
orchestration-guard:
name: verify approved orchestrator
needs: resolve
if: ${{ inputs.orchestrated }}
runs-on: ubuntu-latest
permissions:
contents: read
steps:
- uses: actions/checkout@v7
with:
fetch-depth: 0
ref: ${{ github.sha }}
- name: Verify caller and approved release ref
env:
ACTUAL_CALLER_WORKFLOW_REF: ${{ github.workflow_ref }}
EXPECTED_CALLER_WORKFLOW_REF: ${{ format('{0}/.github/workflows/release-{1}.yml@{2}', github.repository, inputs.orchestrator, github.ref) }}
CALLER_EVENT_NAME: ${{ github.event_name }}
CALLER_REF: ${{ github.ref }}
CALLER_REF_PROTECTED: ${{ github.ref_protected }}
CALLER_WORKFLOW_SHA: ${{ github.workflow_sha }}
CALLER_SHA: ${{ github.sha }}
APPROVED_CLI_TAG: ${{ inputs.approved_cli_tag }}
APPROVED_SHA: ${{ inputs.approved_sha }}
APPROVED_CHANNEL: ${{ inputs.orchestrator }}
RELEASE_TAG: ${{ inputs.tag }}
VERIFY_RELEASE_CHECKOUT: false
run: |
bash scripts/verify-release-authorization.sh
bash scripts/verify-release-tag.sh
release-gate:
name: approve standalone CLI release
needs: resolve
if: ${{ !inputs.orchestrated }}
runs-on: ubuntu-latest
environment: release
steps:
- env:
RELEASE_TAG: ${{ needs.resolve.outputs.tag }}
RELEASE_CHANNEL: ${{ needs.resolve.outputs.channel }}
run: echo "Approved standalone CLI $RELEASE_CHANNEL release $RELEASE_TAG"
cache-guard:
name: cache hit guard
needs: [resolve, orchestration-guard, release-gate]
if: ${{ always() && !cancelled() && needs.resolve.result == 'success' && ((inputs.orchestrated && needs.orchestration-guard.result == 'success') || (!inputs.orchestrated && needs.release-gate.result == 'success')) }}
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
with:
ref: ${{ needs.resolve.outputs.sha }}
- uses: actions/setup-go@v7
with:
go-version-file: go.mod
cache: true
- run: ./scripts/cache-guard.sh
- name: Verify embedded documentation identity
env:
DOCS_BUILD_VERSION: ${{ needs.resolve.outputs.tag }}
DOCS_SOURCE_REVISION: ${{ needs.resolve.outputs.sha }}
run: |
if [ ! -f scripts/verify-embedded-docs.sh ]; then
echo "Legacy candidate predates the embedded docs contract; skipping."
exit 0
fi
bash scripts/verify-embedded-docs.sh "$DOCS_BUILD_VERSION" "$DOCS_SOURCE_REVISION"
goreleaser:
name: archives + checksums + homebrew tap
needs: [resolve, cache-guard]
if: ${{ always() && !cancelled() && needs.cache-guard.result == 'success' }}
runs-on: ubuntu-latest
# The Stable caller has already passed the single GitHub release approval.
# Official standalone recovery passes release-gate above. This job therefore
# must not add a second GitHub environment approval.
steps:
- uses: actions/checkout@v7
with:
fetch-depth: 0
ref: ${{ needs.resolve.outputs.sha }}
# Recovery may build an immutable tag that predates the current recovery
# policy. Keep product sources pinned above, but execute publication
# decisions from the protected workflow commit.
- uses: actions/checkout@v7
with:
fetch-depth: 0
path: release-control
ref: ${{ github.workflow_sha }}
- name: Isolate release-control checkout from product git state
run: |
set -euo pipefail
git_common_dir="$(git rev-parse --path-format=absolute --git-common-dir)"
exclude_file="$git_common_dir/info/exclude"
if ! grep -qxF '/release-control/' "$exclude_file"; then
printf '%s\n' '/release-control/' >> "$exclude_file"
fi
git check-ignore -q release-control/
dirty="$(git status --porcelain --untracked-files=all)"
if [ -n "$dirty" ]; then
printf 'product checkout is dirty before release:\n%s\n' "$dirty" >&2
exit 1
fi
- uses: actions/setup-go@v7
with:
go-version-file: go.mod
cache: true
- uses: actions/setup-node@v7
with:
node-version: "22"
- name: Download orchestrator-reviewed release notes
if: ${{ inputs.orchestrated }}
uses: actions/download-artifact@v8
with:
name: orchestrator-reviewed-release-notes
path: /tmp/orchestrator-reviewed-release-notes
- name: Use orchestrator-reviewed release notes
if: ${{ inputs.orchestrated }}
run: |
test -s /tmp/orchestrator-reviewed-release-notes/release-notes.md
cp /tmp/orchestrator-reviewed-release-notes/release-notes.md /tmp/release-notes.md
- name: Render reviewed release notes
if: ${{ !inputs.orchestrated }}
env:
RELEASE_TAG: ${{ needs.resolve.outputs.notes_version }}
run: node scripts/release-notes.mjs render --version "$RELEASE_TAG" --output /tmp/release-notes.md
- name: Revalidate approved release ref
env:
RELEASE_TAG: ${{ needs.resolve.outputs.tag }}
APPROVED_SHA: ${{ needs.resolve.outputs.sha }}
run: bash scripts/verify-release-tag.sh
- name: Decide whether CLI artifacts need publication
id: publication
env:
GH_TOKEN: ${{ github.token }}
TAG: ${{ needs.resolve.outputs.tag }}
CHANNEL: ${{ needs.resolve.outputs.channel }}
PRERELEASE: ${{ needs.resolve.outputs.prerelease }}
run: |
set -euo pipefail
validation_channel="$CHANNEL"
if [ "$CHANNEL" = "stable" ] && [ "$PRERELEASE" = "true" ]; then
validation_channel=any
fi
release_json=/tmp/existing-cli-release.json
release_error=/tmp/existing-cli-release.error
checksums=/tmp/existing-cli-release-SHA256SUMS
if gh api "repos/${{ github.repository }}/releases/tags/$TAG" \
>"$release_json" 2>"$release_error"; then
gh release download "$TAG" -R "${{ github.repository }}" \
--pattern SHA256SUMS --output "$checksums"
decision="$(
bash release-control/scripts/decide-cli-release-publication.sh \
"$validation_channel" "$TAG" "${{ github.repository }}" \
"$release_json" "$checksums"
)"
echo "existing CLI release $TAG is complete and checksum-bound; reusing it"
elif grep -Eiq 'HTTP 404|Not Found' "$release_error"; then
decision="$(
bash release-control/scripts/decide-cli-release-publication.sh \
"$validation_channel" "$TAG" "${{ github.repository }}" - -
)"
echo "CLI release $TAG does not exist; GoReleaser will publish it"
else
cat "$release_error" >&2
exit 1
fi
test "$decision" = "publish" -o "$decision" = "reuse"
echo "decision=$decision" >> "$GITHUB_OUTPUT"
- uses: goreleaser/goreleaser-action@v7
if: ${{ steps.publication.outputs.decision == 'publish' }}
with:
version: '~> v2'
args: release --clean
env:
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
HOMEBREW_TAP_TOKEN: ${{ needs.resolve.outputs.channel == 'stable' && secrets.HOMEBREW_TAP_TOKEN || '' }}
# workflow_dispatch recovery runs have a branch-shaped GITHUB_REF even
# though checkout is on the release tag. Pin GoReleaser explicitly, and
# avoid ambiguity from the three release tags sharing one commit.
GORELEASER_CURRENT_TAG: ${{ needs.resolve.outputs.tag }}
- name: Publish product release notes
env:
GH_TOKEN: ${{ github.token }}
TAG: ${{ needs.resolve.outputs.tag }}
run: gh release edit "$TAG" --notes-file /tmp/release-notes.md
- name: Publish CLI release metadata to R2
env:
GH_TOKEN: ${{ github.token }}
TAG: ${{ needs.resolve.outputs.tag }}
NOTES_TAG: ${{ needs.resolve.outputs.notes_version }}
HAS_R2: ${{ secrets.R2_ACCESS_KEY_ID != '' && secrets.R2_SECRET_ACCESS_KEY != '' && secrets.R2_ACCOUNT_ID != '' && secrets.R2_BUCKET != '' }}
AWS_ACCESS_KEY_ID: ${{ secrets.R2_ACCESS_KEY_ID }}
AWS_SECRET_ACCESS_KEY: ${{ secrets.R2_SECRET_ACCESS_KEY }}
AWS_DEFAULT_REGION: auto
R2_ACCOUNT_ID: ${{ secrets.R2_ACCOUNT_ID }}
R2_BUCKET: ${{ secrets.R2_BUCKET }}
run: |
set -euo pipefail
if [ "$HAS_R2" != "true" ]; then
echo "R2 secrets not configured; skipping CLI release metadata"
exit 0
fi
channel=""
if [[ "$TAG" =~ ^v(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)$ ]]; then
channel="stable"
elif [[ "$TAG" =~ ^v(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)-preview\.(0|[1-9][0-9]*)$ ]]; then
channel="preview"
else
echo "internal CLI release $TAG; publishing only immutable metadata"
fi
required_assets='[
"reasonix-darwin-amd64.tar.gz",
"reasonix-darwin-arm64.tar.gz",
"reasonix-linux-amd64.tar.gz",
"reasonix-linux-arm64.tar.gz",
"reasonix-windows-amd64.zip",
"reasonix-windows-arm64.zip",
"SHA256SUMS"
]'
gh api "repos/${{ github.repository }}/releases/tags/$TAG" > /tmp/cli-release.raw.json
jq --arg tag "$TAG" --arg notes_tag "$NOTES_TAG" --argjson required "$required_assets" '
if .tag_name != $tag then error("release tag mismatch") else . end |
if .draft then error("draft release cannot be published") else . end |
. as $release |
($release.assets | map({key: .name, value: .}) | from_entries) as $assets |
if ($required | all(. as $name | $assets[$name] != null))
then {
tag_name: $release.tag_name,
prerelease: $release.prerelease,
html_url: $release.html_url,
release_notes_url: ("https://reasonix.io/changelog/" + $notes_tag + "/"),
assets: [
$required[] as $name |
$assets[$name] |
{
name: .name,
browser_download_url: .browser_download_url,
size: .size
}
]
}
else error("release is missing one or more required CLI assets")
end
' /tmp/cli-release.raw.json > /tmp/cli-release.json
if [ -n "$channel" ]; then
bash scripts/validate-cli-release-manifest.sh \
"$channel" "$TAG" "${{ github.repository }}" /tmp/cli-release.json "$NOTES_TAG"
else
bash scripts/validate-cli-release-manifest.sh \
any "$TAG" "${{ github.repository }}" /tmp/cli-release.json "$NOTES_TAG"
fi
endpoint="https://${R2_ACCOUNT_ID}.r2.cloudflarestorage.com"
validation_channel="${channel:-any}"
immutable_key="cli/releases/${TAG}/latest.json"
immutable_error="$(mktemp)"
if aws s3 cp "s3://${R2_BUCKET}/${immutable_key}" /tmp/cli-release.immutable.json \
--endpoint-url "$endpoint" 2>"$immutable_error"; then
bash scripts/validate-cli-release-manifest.sh \
"legacy-${validation_channel}" "$TAG" "${{ github.repository }}" \
/tmp/cli-release.immutable.json "$NOTES_TAG"
if ! bash scripts/compare-cli-release-manifests.sh \
/tmp/cli-release.json /tmp/cli-release.immutable.json; then
echo "::error::immutable CLI release metadata for $TAG already exists with different content"
exit 1
fi
echo "immutable CLI release metadata for $TAG already exists; preserving it"
elif grep -Eiq '404|NoSuchKey|Not Found' "$immutable_error"; then
aws s3 cp /tmp/cli-release.json "s3://${R2_BUCKET}/${immutable_key}" \
--endpoint-url "$endpoint" \
--content-type "application/json; charset=utf-8" \
--cache-control "public, max-age=31536000, immutable"
else
cat "$immutable_error" >&2
exit 1
fi
rm -f "$immutable_error"
aws s3 cp "s3://${R2_BUCKET}/${immutable_key}" /tmp/cli-release.immutable.json \
--endpoint-url "$endpoint"
bash scripts/validate-cli-release-manifest.sh \
"legacy-${validation_channel}" "$TAG" "${{ github.repository }}" \
/tmp/cli-release.immutable.json "$NOTES_TAG"
bash scripts/compare-cli-release-manifests.sh \
/tmp/cli-release.json /tmp/cli-release.immutable.json
if [ -z "$channel" ]; then
echo "internal CLI release $TAG; Stable and Preview pointers remain unchanged"
exit 0
fi
current_tag=""
pointer_error="$(mktemp)"
if aws s3 cp "s3://${R2_BUCKET}/cli/${channel}/latest.json" /tmp/cli-release.pointer.json \
--endpoint-url "$endpoint" 2>"$pointer_error"; then
current_tag="$(jq -er '.tag_name | strings' /tmp/cli-release.pointer.json)"
bash scripts/validate-cli-release-manifest.sh \
"legacy-${channel}" "$current_tag" "${{ github.repository }}" \
/tmp/cli-release.pointer.json "$current_tag"
elif grep -Eiq '404|NoSuchKey|Not Found' "$pointer_error"; then
echo "CLI $channel pointer does not exist yet"
else
cat "$pointer_error" >&2
exit 1
fi
rm -f "$pointer_error"
pointer_manifest=-
if [ -n "$current_tag" ]; then
pointer_manifest=/tmp/cli-release.pointer.json
fi
pointer_decision="$(
bash scripts/decide-cli-pointer-update.sh \
"$channel" /tmp/cli-release.json "$pointer_manifest"
)"
if [ "$pointer_decision" = "skip" ]; then
echo "CLI $channel pointer remains ${current_tag:-unset}; candidate $TAG is not newer and needs no repair"
exit 0
fi
aws s3 cp /tmp/cli-release.json "s3://${R2_BUCKET}/cli/${channel}/latest.json" \
--endpoint-url "$endpoint" \
--content-type "application/json; charset=utf-8" \
--cache-control "public, max-age=300, stale-if-error=86400"
aws s3 cp "s3://${R2_BUCKET}/cli/${channel}/latest.json" /tmp/cli-release.pointer.json \
--endpoint-url "$endpoint"
bash scripts/validate-cli-release-manifest.sh \
"$channel" "$TAG" "${{ github.repository }}" \
/tmp/cli-release.pointer.json "$NOTES_TAG"
cmp -s /tmp/cli-release.json /tmp/cli-release.pointer.json
echo "CLI $channel pointer -> $TAG"
- name: Attach desktop manifest compatibility asset
env:
GH_TOKEN: ${{ github.token }}
TAG: ${{ needs.resolve.outputs.tag }}
HAS_R2: ${{ secrets.R2_ACCESS_KEY_ID != '' && secrets.R2_SECRET_ACCESS_KEY != '' && secrets.R2_ACCOUNT_ID != '' && secrets.R2_BUCKET != '' }}
R2_ACCOUNT_ID: ${{ secrets.R2_ACCOUNT_ID }}
R2_BUCKET: ${{ secrets.R2_BUCKET }}
run: |
set -euo pipefail
case "$TAG" in
*-*)
echo "prerelease $TAG — GitHub latest does not move here; skipping desktop manifest compatibility asset"
exit 0
;;
esac
if [ "$HAS_R2" != "true" ]; then
echo "R2 secrets not configured; skipping desktop manifest compatibility asset"
exit 0
fi
# dl.reasonix.io serves 403 to GitHub Actions egress IPs (Cloudflare bot
# protection), so read the manifest over the authenticated S3 API instead
# of the public edge.
aws configure set aws_access_key_id "${{ secrets.R2_ACCESS_KEY_ID }}"
aws configure set aws_secret_access_key "${{ secrets.R2_SECRET_ACCESS_KEY }}"
aws configure set region auto
aws s3 cp "s3://${R2_BUCKET}/latest/latest.json" latest.raw.json \
--endpoint-url "https://${R2_ACCOUNT_ID}.r2.cloudflarestorage.com"
jq '.download_page = "https://reasonix.io/?download=desktop#start"' latest.raw.json > latest.json
jq -e '
([.platforms[] | (.url, .sig)] |
all(type == "string" and startswith("https://dl.reasonix.io/") and (contains("/releases/latest/") | not)))
' latest.json >/dev/null
gh release upload "$TAG" latest.json --clobber
# The compatibility asset exists for pre-v1.16 desktop updaters that
# still poll GitHub's repository-wide latest URL. Desktop releases now
# own that Latest badge, but this check still exercises the public fallback
# path exactly the way those clients fetch it: anonymously, over the public
# edge, with a Go client UA. Unlike dl.reasonix.io (whose bot protection
# 403s Actions egress — see the R2 note above), GitHub serves its own
# runners, so this can hard-fail. #5826/#5858 shipped a broken update check
# for weeks precisely because nothing exercised the public path. Retries
# cover the release CDN propagating the freshly uploaded asset.
- name: Smoke public compatibility manifest
env:
TAG: ${{ needs.resolve.outputs.tag }}
HAS_R2: ${{ secrets.R2_ACCESS_KEY_ID != '' && secrets.R2_SECRET_ACCESS_KEY != '' && secrets.R2_ACCOUNT_ID != '' && secrets.R2_BUCKET != '' }}
run: |
set -euo pipefail
case "$TAG" in
*-*)
echo "prerelease $TAG — no compatibility asset uploaded; skipping"
exit 0
;;
esac
if [ "$HAS_R2" != "true" ]; then
echo "R2 secrets not configured; no compatibility asset uploaded; skipping"
exit 0
fi
url="https://github.com/${{ github.repository }}/releases/latest/download/latest.json"
for attempt in 1 2 3 4 5 6; do
if curl -fsSL -A "Go-http-client/2.0" -o /tmp/compat-latest.json "$url"; then
jq -e '(.version | type == "string") and (.platforms | type == "object")' /tmp/compat-latest.json >/dev/null
echo "public compatibility manifest OK (desktop version $(jq -r .version /tmp/compat-latest.json))"
exit 0
fi
echo "attempt $attempt failed; retrying in 10s"
sleep 10
done
echo "::error::public compatibility manifest unreachable at $url"
exit 1
# A stable CLI release must never leave the npm line behind: v1.17.5
# shipped as binaries/Homebrew while npm `latest` still pointed at 0.53.2
# (#5822) — every `npm update -g` user was silently downgraded to a
# months-old version, and nothing noticed because the npm line
# (release-npm.yml, `npm-vX.Y.Z` tags) is triggered independently and the
# stable npm tag was simply never pushed. release-npm.yml's own verify
# step only guards runs that happen; this guard catches the run that
# DIDN'T.
#
# Two distinct states, two responses (the approved orchestrator starts the
# CLI and npm reusable workflows concurrently, and npm dist-tags propagate
# asynchronously, so "tag pushed but latest not moved yet" is a NORMAL
# mid-release state, not a failure):
# - npm-v<version> tag missing -> hard fail. This is the #5822 gap:
# nobody pushed the npm release at all.
# - tag pushed, latest lagging -> poll briefly, then WARN and pass.
# The npm job may still be publishing; release-npm.yml's verify step
# owns asserting the dist-tag lands.
- name: Check npm latest dist-tag freshness
env:
TAG: ${{ needs.resolve.outputs.tag }}
run: |
set -euo pipefail
case "$TAG" in
*-*)
echo "prerelease $TAG — npm latest does not move on prereleases; skipping"
exit 0
;;
esac
version="${TAG#v}"
if ! git ls-remote --exit-code origin "refs/tags/npm-v$version" >/dev/null; then
echo "::error::the npm-v$version tag was never pushed — the npm channel is being left behind and 'npm update -g' users will be downgraded to the old 'latest'. Push it: git tag npm-v$version ${TAG} && git push origin npm-v$version (or 'npm dist-tag add reasonix@$version latest' for an already-published version)."
exit 1
fi
for attempt in 1 2 3 4 5 6; do
got="$(npm view reasonix dist-tags.latest 2>/dev/null || true)"
if [ -n "$got" ]; then
newest="$(printf '%s\n%s\n' "$got" "$version" | sort -V | tail -1)"
if [ "$newest" = "$got" ]; then
echo "npm latest -> $got (>= $version) OK"
exit 0
fi
fi
echo "npm latest -> ${got:-<unreadable>}, want >= $version (attempt $attempt)"
sleep 10
done
echo "::warning::npm-v$version is pushed but npm 'latest' is still ${got:-<unreadable>} — the concurrent npm publish is likely still running or propagating. Monitor the npm job; its verify step asserts the dist-tag lands."
exit 0
+1 -1
View File
@@ -24,7 +24,7 @@ jobs:
if: github.event.comment.user.type != 'Bot' && github.event.comment.author_association == 'NONE'
runs-on: ubuntu-latest
steps:
- uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9
- uses: actions/github-script@v9
with:
script: |
const EMOJI = /[\u{1F300}-\u{1FAFF}\u{2600}-\u{27BF}\u{2B00}-\u{2BFF}\u{FE0F}]/gu;
+2 -2
View File
@@ -40,10 +40,10 @@ jobs:
if: github.repository == 'esengine/DeepSeek-Reasonix'
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- uses: actions/checkout@v7
with:
fetch-depth: 0
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
- uses: actions/setup-node@v7
with:
node-version: '22'
- name: Test sweeper
@@ -1,117 +0,0 @@
name: Studio Certum signing smoke test
# Signs two throwaway probes with the certificate release-studio.yml would use,
# and publishes nothing. It reports the signer's subject, which is the value
# STUDIO_SIGNING_SUBJECT must hold.
on:
workflow_dispatch:
permissions:
contents: read
concurrency:
group: certum-signing
cancel-in-progress: false
jobs:
sign:
if: github.repository == 'esengine/DeepSeek-Reasonix' && github.ref == 'refs/heads/studio'
runs-on: windows-2022
environment: studio-release
timeout-minutes: 20
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
persist-credentials: false
- uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
with:
go-version-file: go.mod
cache: false
- name: Check signing configuration
shell: pwsh
env:
CERTUM_USERNAME: ${{ secrets.CERTUM_USERNAME }}
CERTUM_OTP_URI: ${{ secrets.CERTUM_OTP_URI }}
CERTUM_KEY_ID: ${{ secrets.CERTUM_KEY_ID }}
run: |
$ErrorActionPreference = 'Stop'
if ([string]::IsNullOrWhiteSpace($env:CERTUM_USERNAME)) { throw 'Missing secret CERTUM_USERNAME' }
if ($env:CERTUM_OTP_URI -notmatch '^otpauth://totp/') { throw 'Secret CERTUM_OTP_URI must be a complete TOTP URI' }
if ($env:CERTUM_KEY_ID -notmatch '^[0-9a-fA-F]{40}$') { throw 'Secret CERTUM_KEY_ID must be the 40-character SHA-1 certificate thumbprint' }
- name: Build open-source signing probes
shell: pwsh
run: |
$ErrorActionPreference = 'Stop'
# This minimal probe is covered by the repository's MIT license.
New-Item -ItemType Directory payload | Out-Null
'package main; func main() { println("Reasonix Studio Certum signing probe") }' | Set-Content probe.go
$env:GOOS = 'windows'
$env:GOARCH = 'amd64'
$env:CGO_ENABLED = '0'
go build -trimpath -o payload/probe.exe probe.go
if ($LASTEXITCODE -ne 0) { throw 'Probe compilation failed' }
'probe.exe' | Set-Content payload/signing-files.txt
Copy-Item payload/probe.exe installer-probe.exe
- name: Connect to SimplySign cloud certificate
uses: ./.github/actions/setup-certum
with:
username: ${{ secrets.CERTUM_USERNAME }}
otp-uri: ${{ secrets.CERTUM_OTP_URI }}
thumbprint: ${{ secrets.CERTUM_KEY_ID }}
- name: Sign and verify payload and installer probes
shell: pwsh
env:
CERTUM_KEY_ID: ${{ secrets.CERTUM_KEY_ID }}
run: |
./scripts/sign-certum.ps1 -PayloadDirectory payload
./scripts/sign-certum.ps1 -FilePath installer-probe.exe
# The session signs for any process on the runner while it is up.
- name: Disconnect from Certum
if: always()
shell: pwsh
run: Get-Process -Name '*SimplySign*' -ErrorAction SilentlyContinue | Stop-Process -Force
# With the switch on, an unset subject fails here as it would in a
# release; with it off, the value is reported so it can be recorded.
- name: Report and check the signer
shell: pwsh
env:
SIGNING_ENABLED: ${{ vars.STUDIO_SIGNING_ENABLED }}
EXPECTED_SUBJECT: ${{ vars.STUDIO_SIGNING_SUBJECT }}
run: |
$ErrorActionPreference = 'Stop'
$failed = $false
foreach ($file in @('payload/probe.exe', 'installer-probe.exe')) {
$signature = Get-AuthenticodeSignature -LiteralPath $file
$certificate = $signature.SignerCertificate
$stamp = if ($signature.TimeStamperCertificate) { $signature.TimeStamperCertificate.Subject } else { '(none)' }
@("### $file", '', "- subject: ``$($certificate.Subject)``", "- issuer: ``$($certificate.Issuer)``",
"- thumbprint: ``$($certificate.Thumbprint)``", "- timestamp: ``$stamp``", '') >> $env:GITHUB_STEP_SUMMARY
if (-not $signature.TimeStamperCertificate) {
Write-Output "::error title=studio-signing.timestamp-missing::$file carries no timestamp"
$failed = $true
}
if ([string]::IsNullOrWhiteSpace($certificate.Issuer)) {
Write-Output "::error title=studio-signing.issuer-missing::$file has a signer certificate without an issuer"
$failed = $true
}
if ([string]::IsNullOrWhiteSpace($env:EXPECTED_SUBJECT)) {
if ($env:SIGNING_ENABLED -eq 'true') {
Write-Output "::error title=studio-signing.subject-unset::STUDIO_SIGNING_SUBJECT is empty while STUDIO_SIGNING_ENABLED=true"
$failed = $true
} else {
Write-Output "::warning::STUDIO_SIGNING_SUBJECT is not set; record '$($certificate.Subject)'"
}
} elseif ($certificate.Subject -cne $env:EXPECTED_SUBJECT) {
Write-Output "::error title=studio-signing.unexpected-signer::STUDIO_SIGNING_SUBJECT is '$($env:EXPECTED_SUBJECT)' but $file is signed by '$($certificate.Subject)'"
$failed = $true
}
}
if ($failed) { exit 1 }
'Certum payload and installer signing, signer, trust and timestamps passed.' >> $env:GITHUB_STEP_SUMMARY
@@ -1,107 +0,0 @@
name: Studio Windows self-test
# An unsigned Windows installer for trying a branch on a real machine before it
# is released. It runs on the intranet runner, so it is dispatch-only: the
# repository is public, and nothing a pull request writes may run there.
on:
workflow_dispatch:
inputs:
ref:
description: Branch, tag or commit to build
required: false
default: studio
type: string
permissions:
contents: read
concurrency:
group: studio-selftest-windows
cancel-in-progress: true
jobs:
build:
runs-on: win-gpu
timeout-minutes: 90
steps:
# The machine's PATH finds WSL's bash before Git's, and a bash step run in
# WSL sees neither this checkout's toolchain nor Windows paths. cmd, since
# the machine's execution policy refuses the script a powershell step is.
- name: Put Git Bash first on PATH
shell: cmd
run: '>>"%GITHUB_PATH%" echo C:\Program Files\Git\bin'
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
ref: ${{ inputs.ref }}
- uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
with:
go-version-file: go.mod
cache: true
cache-dependency-path: go.sum
- uses: pnpm/action-setup@ea17c68df8912ef543352723c149a84f56e3d413 # v6.1.0
with:
version: 10
run_install: false
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
with:
node-version: "24"
cache: pnpm
cache-dependency-path: |
desktop/frontend-next/pnpm-lock.yaml
desktop/electron/pnpm-lock.yaml
# A patch above the newest release, as a prerelease: it ranks ahead of
# what is published, so the build does not offer itself that release as
# an update, and behind the next one, which it does offer.
- name: Choose the version
id: version
shell: bash
run: |
latest="$(git ls-remote --tags origin 'refs/tags/studio-v*' | sed -n 's#.*refs/tags/studio-v\([0-9.]*\)$#\1#p' | sort -V | tail -1)"
test -n "$latest" || { echo "::error::no studio-v* release tag"; exit 1; }
IFS=. read -r major minor patch <<< "$latest"
version="$major.$minor.$((patch + 1))-selftest.${{ github.run_number }}"
echo "version=$version" >> "$GITHUB_OUTPUT"
echo "Version: \`$version\` from \`$(git rev-parse --short HEAD)\`" >> "$GITHUB_STEP_SUMMARY"
# The intranet's route to the module proxy drops connections now and then.
# What arrives stays in this machine's module cache, so a retry resumes.
- name: Fetch Go modules
shell: bash
run: for i in 1 2 3 4 5; do go mod download && exit 0; sleep 10; done; exit 1
- name: Build the SPA
working-directory: desktop/frontend-next
shell: bash
run: pnpm install --frozen-lockfile && pnpm build
- name: Build the kernel
working-directory: desktop/electron
shell: bash
env:
REASONIX_VERSION: v${{ steps.version.outputs.version }}
run: pnpm install --frozen-lockfile && pnpm run build:host
- name: Build the bundle
working-directory: desktop/electron
shell: bash
run: pnpm exec electron-builder --dir -c.extraMetadata.version="${{ steps.version.outputs.version }}"
- name: Package the installer
working-directory: desktop/electron
shell: bash
run: |
pnpm exec electron-builder --prepackaged ../../dist/win-unpacked --win nsis \
-c.extraMetadata.version="${{ steps.version.outputs.version }}"
sha256sum ../../dist/ReasonixStudio-windows-amd64-installer.exe | tee -a "$GITHUB_STEP_SUMMARY"
- uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
with:
name: studio-selftest-windows-${{ steps.version.outputs.version }}
path: dist/ReasonixStudio-windows-amd64-installer.exe
if-no-files-found: error
retention-days: 7
-688
View File
@@ -1,688 +0,0 @@
name: Studio
# Studio's own surface — desktop/frontend-next (the SPA) and the Electron shell
# that serves it — exists only on the studio branch, so it cannot live in ci.yml
# without failing on main-v2 where neither directory is present. ci.yml covers
# the kernel; this file covers the frontend build, the desktop module's Go side,
# and the packaging contract on real artifacts.
on:
push:
branches: [studio]
pull_request:
branches: [studio]
permissions:
contents: read
concurrency:
group: studio-${{ github.ref }}
cancel-in-progress: ${{ github.event_name == 'pull_request' || github.ref == 'refs/heads/studio' }}
jobs:
frontend:
name: frontend-next
runs-on: ubuntu-latest
defaults:
run:
working-directory: desktop/frontend-next
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- uses: pnpm/action-setup@ea17c68df8912ef543352723c149a84f56e3d413 # v6.1.0
with:
version: 10
run_install: false
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
with:
node-version: "24"
cache: pnpm
cache-dependency-path: desktop/frontend-next/pnpm-lock.yaml
- name: Install
run: pnpm install --frozen-lockfile
- name: typecheck
run: pnpm typecheck
# The reducer and the event port carry the two pieces of logic a type
# error cannot reach: what a card means when a turn is interrupted, and
# what the client does about a frame that never arrived.
- name: test
run: pnpm test
# What the UI census holds. The gate reads structure — which inputs reach
# a mutating endpoint, and which of those the product names an action for
# — so UNDECLARED_MUTATION == 0 is a fact about the tree, not a lint over
# how anything is spelled.
- name: census invariants
run: pnpm census > /dev/null
# And that the analyzer still judges the way it did. _fx is a corpus
# written to exercise the rules, so its reports move only when a judgement
# moves; the product's name every root by file and line and are stale the
# moment a button moves, which is why they are not frozen.
- name: census fixture reports have not drifted
run: pnpm census:verify
# `build` runs tsc -b before vite, so this also proves the project
# references resolve — typecheck alone uses --noEmit and does not.
- name: build
run: pnpm build
# The published bundle must not carry MockPort's scripted fixtures: a
# release that cannot reach the kernel has to say so, not act out a fake
# session. main.tsx guards the import behind import.meta.env.DEV, and this
# is what proves the guard still holds.
- name: no fixtures in the bundle
run: |
if grep -rl 网关瞬时态 dist/ 2>/dev/null; then
echo "::error::MockPort fixtures reached the production bundle - check the import.meta.env.DEV guard in main.tsx"
exit 1
fi
# Real-Chromium guards that run without a kernel. Only the ones that passed
# three consecutive local runs are wired; the rest stay manual (see
# desktop/frontend-next/perf/README.md). Skipped unless the SPA changed.
browser-guards:
name: browser guards
runs-on: ubuntu-latest
timeout-minutes: 25
defaults:
run:
working-directory: desktop/frontend-next
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
fetch-depth: 0
- name: SPA changed?
id: spa
working-directory: .
run: |
if [ "${{ github.event_name }}" = "pull_request" ] &&
git diff --quiet "origin/${{ github.base_ref }}...HEAD" -- desktop/frontend-next; then
echo "run=false" >> "$GITHUB_OUTPUT"
else
echo "run=true" >> "$GITHUB_OUTPUT"
fi
- if: steps.spa.outputs.run == 'true'
uses: pnpm/action-setup@ea17c68df8912ef543352723c149a84f56e3d413 # v6.1.0
with:
version: 10
run_install: false
- if: steps.spa.outputs.run == 'true'
uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
with:
node-version: "24"
cache: pnpm
cache-dependency-path: desktop/frontend-next/pnpm-lock.yaml
- if: steps.spa.outputs.run == 'true'
name: Install
run: pnpm install --frozen-lockfile
- if: steps.spa.outputs.run == 'true'
name: Build the perf harness and install Chromium
run: |
pnpm exec vite build --config vite.perf.config.ts
pnpm exec playwright install --with-deps chromium
- if: steps.spa.outputs.run == 'true'
name: Guards
run: |
python3 -m http.server 4399 --directory dist-perf >/dev/null 2>&1 &
for _ in $(seq 30); do curl -sf http://localhost:4399/perf.html >/dev/null && break; sleep 1; done
node perf/intrinsic.mjs
node perf/firstscreen-fonts.mjs
node perf/contrast.mjs
node perf/stickyfold.mjs
node perf/crumb.mjs
node perf/hoverradius.mjs
node perf/jump-ends.mjs
node perf/modelrows.mjs
node perf/explorersize.mjs
node perf/git-panel.mjs
node perf/session-move-to.mjs
node perf/shell-refused.mjs
node perf/external-origin.mjs
node perf/rail-search-hold.mjs
node perf/rail-toggle-panes.mjs
node perf/session-move.mjs
node perf/session-ref.mjs
node perf/browser-full.mjs
node perf/mod-ui.mjs
node perf/codeeditor-size.mjs
node perf/composer-reflow.mjs
node perf/composer-toolbar-row.mjs
node perf/action-icons.mjs
node perf/rowhit.mjs
node perf/prefs-wide-centered.mjs
node perf/tab-pip.mjs
node perf/syntax-tokens.mjs
node perf/tab-close-inside.mjs
node perf/context-card-hover.mjs
node perf/sessquick.mjs
node perf/sessionrows.mjs
node perf/session-row-marker.mjs
node perf/session-selected-row.mjs
node perf/wheel.mjs
node perf/branch-tooltip.mjs
node perf/branch-alignment.mjs
node perf/branch-menu.mjs
node perf/image-preview.mjs
node perf/user-images.mjs
node perf/toast-anchor.mjs
node perf/addfolder-rest.mjs
node perf/scroll-jump.mjs
node perf/add-project-selects.mjs
node perf/project-header.mjs
node perf/setup-defer.mjs
node perf/narrow.mjs
node perf/settings-rows.mjs
node perf/narrow-steps.mjs
node perf/job-names.mjs
node perf/mcp-actions.mjs
node perf/web-access.mjs
node perf/peers-settings.mjs
node perf/output-cap.mjs
node perf/tool-output-clip.mjs
node perf/agent-run-tab.mjs
node perf/changes-cap.mjs
node perf/session-menu-fit.mjs
node perf/rail-rows-typography.mjs
node perf/settings-nav-fit.mjs
node perf/rail-rows-align.mjs
node perf/progress-tail.mjs
node perf/trust-wait.mjs
node perf/nest-nodes.mjs
node perf/provider-ports.mjs
node perf/peer-threads.mjs
node perf/text-size.mjs
node perf/fs-scale.mjs
node perf/composer-feel.mjs
node perf/side-panel.mjs
node perf/card-shell.mjs
node perf/line-cards.mjs
node perf/detail-panel.mjs
node perf/brand-lockup.mjs
node perf/notice-line.mjs
node perf/brought-line.mjs
node perf/modal-focus.mjs
node perf/focus-ring.mjs
node perf/narrow-forms.mjs
node perf/i18n-rendered.mjs
node perf/narrow-rail.mjs
node perf/selfcheck.mjs branch-tooltip
# What a hosted page does over the relay, end to end: the production bundle,
# the real Studio host and kernel, a real Chromium; only the accounts API and
# the relay are stand-ins (tools/relayfake). Event-transport changes are
# judged here, because a unit test of one side cannot see the other.
relay-harness:
name: relay harness
runs-on: ubuntu-latest
timeout-minutes: 15
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
fetch-depth: 0
- name: Transport changed?
id: scope
env:
BASE_REF: ${{ github.base_ref }}
run: |
paths="desktop/frontend-next internal/frontend/serve internal/platform/remotecloud internal/platform/account tools/relayfake cmd/reasonix-studio-host"
if [ "$GITHUB_EVENT_NAME" = "pull_request" ] &&
git diff --quiet "origin/$BASE_REF...HEAD" -- $paths; then
echo "run=false" >> "$GITHUB_OUTPUT"
else
echo "run=true" >> "$GITHUB_OUTPUT"
fi
- if: steps.scope.outputs.run == 'true'
uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
with:
go-version-file: go.mod
cache: true
- if: steps.scope.outputs.run == 'true'
uses: pnpm/action-setup@ea17c68df8912ef543352723c149a84f56e3d413 # v6.1.0
with:
version: 10
run_install: false
- if: steps.scope.outputs.run == 'true'
uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
with:
node-version: "24"
cache: pnpm
cache-dependency-path: desktop/frontend-next/pnpm-lock.yaml
- if: steps.scope.outputs.run == 'true'
name: Build the host, the stand-ins and the page
run: |
go build -o desktop/frontend-next/relay/bin/studio-host ./cmd/reasonix-studio-host
go build -o desktop/frontend-next/relay/bin/relayfake ./tools/relayfake
cd desktop/frontend-next
pnpm install --frozen-lockfile
pnpm exec vite build --outDir dist-relay --emptyOutDir
pnpm exec playwright install --with-deps chromium
- if: steps.scope.outputs.run == 'true'
name: Scenarios (hosted page over the relay; local page on a flapping network)
working-directory: desktop/frontend-next
env:
RELAY_HOST_BIN: ${{ github.workspace }}/desktop/frontend-next/relay/bin/studio-host
RELAY_FAKE_BIN: ${{ github.workspace }}/desktop/frontend-next/relay/bin/relayfake
RELAY_DIST: ${{ github.workspace }}/desktop/frontend-next/dist-relay
run: |
node relay/live.mjs
node relay/storm.mjs
# A pull request with no platform-relevant change (scripts/ci-platform-scope.sh)
# moves the Windows and macOS legs onto an ubuntu runner with every step
# skipped, so the check names still report without a scarce runner. A push to
# studio, or a failed `scope`, runs all three.
desktop:
name: desktop (${{ matrix.os }})
needs: scope
if: always()
runs-on: ${{ github.event_name == 'pull_request' && needs.scope.outputs.platform == 'false' && matrix.os != 'ubuntu-22.04' && 'ubuntu-latest' || matrix.os }}
env:
RUN_STEPS: ${{ github.event_name != 'pull_request' || needs.scope.outputs.platform != 'false' || matrix.os == 'ubuntu-22.04' }}
strategy:
fail-fast: false
matrix:
os: [ubuntu-22.04, windows-latest, macos-latest]
defaults:
run:
working-directory: desktop
steps:
- if: env.RUN_STEPS == 'true'
uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- if: env.RUN_STEPS == 'true'
uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
with:
go-version-file: desktop/go.mod
cache: true
cache-dependency-path: desktop/go.sum
# The desktop module's Go tests: the elevated update helper, the
# release-side tools and the Windows uninstall planner. The code this
# gate covers is what replaces a running application on a user's disk.
# The module is CGO-free since the Wails shell retired, so no platform
# carries build dependencies of its own.
# Same command as `make studio-test`; this job is already in desktop/.
- name: go test
if: env.RUN_STEPS == 'true'
run: go test ./...
- if: env.RUN_STEPS == 'true'
uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
with:
node-version: "24"
- name: CUA process and packaging contracts
if: env.RUN_STEPS == 'true'
timeout-minutes: 2
run: node --test --test-name-pattern=CUA electron/test/unit.mjs
- name: Install Linux native CUA fixture dependencies
if: env.RUN_STEPS == 'true' && runner.os == 'Linux'
run: sudo apt-get update && sudo apt-get install -y xvfb openbox dbus-x11 libgtk-3-dev at-spi2-core
- name: Native CUA application acceptance
if: env.RUN_STEPS == 'true'
timeout-minutes: 6
shell: bash
run: |
if [ "$RUNNER_OS" = Linux ]; then
dbus-run-session -- xvfb-run -a sh -c 'openbox >/tmp/reasonix-openbox.log 2>&1 & export NO_AT_BRIDGE=0 GTK_MODULES=gail:atk-bridge; node electron/test/cua-native.mjs'
else
node electron/test/cua-native.mjs
fi
# The host sits in the main module and nothing else builds it, so its
# dependency closure is checked here: it must stay CGO-free and
# platform-neutral or the Electron bundle cannot ship it. Cross-compiling
# is host-independent, so once is enough.
- name: host stays pure Go and platform-neutral
if: env.RUN_STEPS == 'true' && runner.os == 'Linux'
working-directory: .
run: |
for os in windows linux darwin; do
for arch in amd64 arm64; do
CGO_ENABLED=0 GOOS=$os GOARCH=$arch go build -o /dev/null ./cmd/reasonix-studio-host
done
done
# The ARM64 leg queues on a scarcer runner pool, so a pull request runs it
# only when it touches what the leg verifies; a push to studio always does.
# The macOS and Windows package legs follow `platform`, the same rule as the
# desktop matrix above.
scope:
name: package matrix
runs-on: ubuntu-latest
outputs:
matrix: ${{ steps.pick.outputs.matrix }}
arm64: ${{ steps.pick.outputs.arm64 }}
platform: ${{ steps.pick.outputs.platform }}
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
fetch-depth: 0
- id: pick
env:
EVENT: ${{ github.event_name }}
BASE_SHA: ${{ github.event.pull_request.base.sha }}
run: |
arm64=true
platform=true
if [ "$EVENT" = pull_request ]; then
arm64=false
platform=$(git diff -z --name-only --no-renames "$BASE_SHA" HEAD | bash scripts/ci-platform-scope.sh)
if git diff --name-only --no-renames "$BASE_SHA" HEAD |
grep -qE '^(desktop/electron/|desktop/packaging/|desktop/cmd/update-helper/|desktop/cmd/studio-manifest/|desktop/go\.(mod|sum)$|cmd/reasonix-studio-host/|internal/platform/(update|appupdate)/|go\.(mod|sum)$|scripts/(linux-install-smoke|verify-linux-package|verify-linux-appimage|verify-studio-artifacts)\.sh$|scripts/(windows-signing-lib|verify-windows-package|verify-windows-authenticode|install-smoke)\.ps1$|\.github/workflows/(studio|release-studio)\.yml$)'; then
arm64=true
fi
fi
legs='[
{"os":"macos-latest","arch":"arm64","platform":"darwin"},
{"os":"ubuntu-22.04","arch":"amd64","platform":"linux"},
{"os":"windows-latest","arch":"amd64","platform":"windows","ebarch":"x64"}
]'
if [ "$platform" = false ]; then
legs="$(jq -c 'map(select(.platform == "linux"))' <<<"$legs")"
elif [ "$arm64" = true ]; then
legs="$(jq -c '. + [{"os":"windows-11-arm","arch":"arm64","platform":"windows","ebarch":"arm64"},{"os":"ubuntu-22.04-arm","arch":"arm64","platform":"linux"}]' <<<"$legs")"
else
legs="$(jq -c . <<<"$legs")"
fi
{
echo "arm64=$arm64"
echo "platform=$platform"
echo "matrix={\"include\":$legs}"
} >> "$GITHUB_OUTPUT"
# The packaging contract, on real artifacts rather than on the config that
# describes them. Unsigned on purpose: signing and notarization are a
# distribution concern and belong to the release workflow, and treating their
# absence as "packaging unverified" is what left macOS and Linux resting on a
# cross-compile while only Windows had ever produced a package.
package:
name: package (${{ matrix.os }})
needs: scope
runs-on: ${{ matrix.os }}
timeout-minutes: 45
strategy:
fail-fast: false
matrix: ${{ fromJSON(needs.scope.outputs.matrix) }}
defaults:
run:
shell: bash
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e # v7
with:
go-version-file: desktop/go.mod
cache: true
cache-dependency-path: desktop/go.sum
- uses: pnpm/action-setup@ea17c68df8912ef543352723c149a84f56e3d413 # v6.1.0
with:
version: 10
run_install: false
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
with:
node-version: "24"
cache: pnpm
cache-dependency-path: desktop/electron/pnpm-lock.yaml
# electron-builder shells out to dpkg-deb through fakeroot for the .deb and
# to rpmbuild for the .rpm.
- name: Install Linux packaging deps
if: runner.os == 'Linux'
run: sudo apt-get update && sudo apt-get install -y fakeroot rpm cpio xvfb xauth policykit-1
# The fpm electron-builder carries is an x86_64 program, so on an ARM64
# runner the .deb and .rpm are built with one installed from RubyGems.
- name: Use a native fpm on ARM64 Linux
if: runner.os == 'Linux' && runner.arch == 'ARM64'
run: |
sudo apt-get install -y ruby ruby-dev build-essential
sudo gem install --no-document fpm -v 1.16.0
echo "USE_SYSTEM_FPM=true" >> "$GITHUB_ENV"
- name: Build the SPA
working-directory: desktop/frontend-next
run: pnpm install --frozen-lockfile && pnpm build
# The Windows legs name their architecture instead of taking the
# toolchain's, so a runner whose Go or Node is the other architecture
# fails here rather than shipping a mislabelled package.
- name: Build the kernel the shell speaks to
working-directory: desktop/electron
env:
REASONIX_GOARCH: ${{ matrix.platform == 'windows' && matrix.arch || '' }}
REASONIX_VERSION: v0.0.0-ci
run: pnpm install --frozen-lockfile && pnpm test && pnpm run build:host
# The .deb's absolute-path payload. A no-op off Linux, so it is not
# guarded here: the one place it must not be forgotten is the one place
# it does anything.
- name: Stage the Linux package payload
working-directory: desktop/electron
run: pnpm run build:linux-payload
# No identity on a runner, and none wanted here: auto-discovery would make
# this job fail for the one reason it is not testing.
- name: Package
working-directory: desktop/electron
env:
CSC_IDENTITY_AUTO_DISCOVERY: "false"
ELECTRON_BUILDER_7Z_FILTER: ${{ matrix.platform == 'windows' && matrix.arch == 'arm64' && 'BCJ' || '' }}
run: pnpm run build:dist ${{ matrix.ebarch && format('--{0}', matrix.ebarch) || '' }} -c.extraMetadata.version=0.0.0-ci
# The release vocabulary is GOARCH. electron-builder's own is not, and the
# rename that reconciles them runs here rather than in the manifest.
- name: Artifacts carry the release architecture
run: |
ls -la dist
if ls dist | grep -q -- '-x64'; then
echo "::error::an artifact still names x64; studio-manifest reads names as GOARCH"
exit 1
fi
if ! ls dist | grep -q -- "-${{ matrix.platform }}-${{ matrix.arch }}"; then
echo "::error::no artifact named for ${{ matrix.platform }}-${{ matrix.arch }}"
exit 1
fi
# An emulated x64 program runs on ARM64 Windows, so a package that merely
# starts proves nothing about its architecture: every PE image in the
# bundle has to be built for the one the package is named for.
- name: Every PE image is built for the package's architecture
if: matrix.platform == 'windows'
run: |
dir=$(find dist -maxdepth 1 -type d -name 'win*-unpacked' | head -1)
test -n "$dir"
node desktop/electron/packaging/pe.js "$dir" "${{ matrix.arch }}"
# An update starts the installer with --updated, and that is what keeps the
# shortcuts (and the taskbar pin hanging off them) from being deleted and
# recreated. Installed twice over itself the way an update does it, the
# desktop shortcut has to still be the same file, pointing at a path with
# no version in it, or every update costs the user their pin.
- name: An update keeps the shortcuts and they name no version
if: matrix.platform == 'windows'
shell: pwsh
run: |
$installer = (Get-ChildItem dist -Filter 'ReasonixStudio-windows-*-installer.exe' | Select-Object -First 1).FullName
$dir = Join-Path $env:RUNNER_TEMP 'rx-update-guard'
Start-Process $installer -ArgumentList '/S', "/D=$dir" -Wait
$lnk = Join-Path ([Environment]::GetFolderPath('Desktop')) 'Reasonix Studio.lnk'
$target = (New-Object -ComObject WScript.Shell).CreateShortcut($lnk).TargetPath
if ($target -ne (Join-Path $dir 'Reasonix Studio.exe')) { throw "the desktop shortcut targets $target, not the stable executable" }
$mark = [datetime]'2020-01-01'
(Get-Item $lnk).LastWriteTime = $mark
Start-Process $installer -ArgumentList '/S', '--updated', "/D=$dir" -Wait
if ((Get-Item $lnk).LastWriteTime -ne $mark) { throw 'an update rewrote the desktop shortcut' }
Start-Process (Join-Path $dir 'Uninstall Reasonix Studio.exe') -ArgumentList '/S', '/currentuser' -Wait
# The release packages a signed copy of the bundle and then reads both
# packages back against it. The same recipe on an unsigned copy, so the
# packaging contract (the installer's one application archive, the pinned
# elevate.exe, every PE image's architecture) is exercised on pull
# requests and not first by a release.
- name: The packages read back against the bundle they were made from
if: matrix.platform == 'windows'
shell: bash
env:
ELECTRON_BUILDER_7Z_FILTER: ${{ matrix.arch == 'arm64' && 'BCJ' || '' }}
run: |
rm -rf payload nsis-src packed zip-src
cp -r "$(find dist -maxdepth 1 -type d -name 'win*-unpacked' | head -1)" payload
rm -f payload/resources/elevate.exe
# The NSIS run writes elevate.exe into the tree it packs, so each
# packager gets a copy and payload stays what was signed.
cp -r payload zip-src
cp -r payload nsis-src
cd desktop/electron
pnpm exec electron-builder --prepackaged ../../zip-src --win zip --${{ matrix.ebarch }} \
-c.extraMetadata.version=0.0.0-ci -c.directories.output=../../packed-zip
pnpm exec electron-builder --prepackaged ../../nsis-src --win nsis --${{ matrix.ebarch }} \
-c.extraMetadata.version=0.0.0-ci -c.directories.output=../../packed
mv ../../packed-zip/ReasonixStudio-windows-${{ matrix.arch }}.zip ../../packed/
ls -la ../../packed
- name: verify-windows-package.ps1 accepts them
if: matrix.platform == 'windows'
shell: pwsh
run: |
$ErrorActionPreference = 'Stop'
. ./scripts/windows-signing-lib.ps1
$digest = Get-ManifestDigest -Manifest @(Get-TreeManifest -Root payload)
./scripts/verify-windows-package.ps1 `
-PayloadDirectory payload `
-ExpectedPayloadDigest $digest `
-InstallerPath "packed/ReasonixStudio-windows-${{ matrix.arch }}-installer.exe" `
-PortableArchivePath "packed/ReasonixStudio-windows-${{ matrix.arch }}.zip" `
-OutputPath "$env:RUNNER_TEMP/verify-output.txt" `
-Architecture ${{ matrix.arch }}
Get-Content "$env:RUNNER_TEMP/verify-output.txt"
# The release's own install smoke, against the unsigned installer: the
# package is installed where it is meant to run, its tree checked, and
# removed.
- name: The package installs, runs and uninstalls
if: matrix.platform == 'windows'
shell: pwsh
run: |
./scripts/install-smoke.ps1 `
-Installer "dist/ReasonixStudio-windows-${{ matrix.arch }}-installer.exe" `
-Architecture ${{ matrix.arch }}
- name: Keep the arm64 installer for the x64 refusal check
if: matrix.platform == 'windows' && matrix.arch == 'arm64'
uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
with:
name: studio-arm64-installer-ci
path: dist/ReasonixStudio-windows-arm64-installer.exe
if-no-files-found: error
retention-days: 1
# The update channel is the signed manifest. A second answer to which
# version is current is not a convenience. The .blockmap is the same
# channel's differential index and carries the release prefix, so it also
# inflates the count the release uses to refuse a partial publish.
- name: No electron-updater metadata
run: |
if ls dist | grep -qiE '^(latest|beta|alpha).*\.yml$'; then
echo "::error::electron-updater metadata was published beside the manifest channel"
exit 1
fi
if ls dist | grep -q '\.blockmap$'; then
echo "::error::a .blockmap ships under the release prefix; set differentialPackage: false"
exit 1
fi
# A binary inside app.asar cannot be spawned and the kernel serves the SPA
# off disk, so both have to be resources. Found rather than asserted at a
# path, because the unpacked directory is named by the builder.
- name: The kernel and the page are resources, not archive members
run: |
host=$(find dist -type f -path '*/bin/reasonix-studio-host*' ! -path '*app.asar*' | head -1)
page=$(find dist -type f -path '*/frontend-next/dist/index.html' ! -path '*app.asar*' | head -1)
echo "kernel: ${host:-<missing>}"
echo "page: ${page:-<missing>}"
test -n "$host" -a -n "$page"
- name: Linux packages carry their declared architecture and update payload
if: runner.os == 'Linux'
run: bash scripts/verify-linux-package.sh "dist/ReasonixStudio-linux-${{ matrix.arch }}.deb" "dist/ReasonixStudio-linux-${{ matrix.arch }}.rpm" "${{ matrix.arch }}"
- name: The Linux package installs, runs, reinstalls and uninstalls
if: runner.os == 'Linux'
run: bash scripts/linux-install-smoke.sh "dist/ReasonixStudio-linux-${{ matrix.arch }}.deb" "${{ matrix.arch }}" 0.0.0-ci
- name: The AppImage carries its architecture, the kernel and the page
if: runner.os == 'Linux'
run: bash scripts/verify-linux-appimage.sh "dist/ReasonixStudio-linux-${{ matrix.arch }}.AppImage" "${{ matrix.arch }}"
# Recorded, not asserted: the real bundle layout, so later work argues
# from what electron-builder produced rather than from its documentation.
- name: Record the macOS bundle layout
if: runner.os == 'macOS'
run: |
app=$(find dist -maxdepth 2 -name '*.app' | head -1)
echo "app: $app"
ls "$app/Contents"
ls "$app/Contents/Resources" | head -20
ls "$app/Contents/Frameworks" | head -20
# The end of the contract: not that the name parses, but that the manifest
# keys the artifact under the platform this runner actually is. The tool
# writes latest.json into the directory and logs to stdout, so the check
# reads the file -- captured output would match its own progress line.
- name: studio-manifest resolves the platform this runner is
working-directory: desktop
env:
GITHUB_REPOSITORY: ${{ github.repository }}
run: |
go run ./cmd/studio-manifest ../dist v0.0.0-ci v0.0.0-ci
cat ../dist/latest.json
if ! grep -q '"${{ matrix.platform }}-${{ matrix.arch }}"' ../dist/latest.json; then
echo "::error::the manifest has no entry for ${{ matrix.platform }}-${{ matrix.arch }}"
exit 1
fi
# The arm64 installer carries one application and no other. Run anywhere but
# ARM64 Windows it has to say so and stop, instead of exiting 0 with nothing
# installed.
arm64-installer-refuses-x64:
name: arm64 installer refuses an x64 machine
needs: [scope, package]
if: needs.scope.outputs.arm64 == 'true'
runs-on: windows-latest
steps:
- uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8.0.2
with:
name: studio-arm64-installer-ci
path: installer
- name: Run it silently
shell: pwsh
run: |
$ErrorActionPreference = 'Stop'
$target = Join-Path $env:RUNNER_TEMP 'rx-refused'
$run = Start-Process -FilePath installer/ReasonixStudio-windows-arm64-installer.exe -ArgumentList '/S', "/D=$target" -Wait -PassThru
Write-Host "exit code $($run.ExitCode)"
if ($run.ExitCode -eq 0) { throw 'the arm64 installer exited 0 on an x64 machine' }
if (@(Get-ChildItem -Recurse -Force $target -ErrorAction SilentlyContinue).Count -ne 0) { throw 'the arm64 installer wrote into the target on an x64 machine' }
+6 -7
View File
@@ -13,8 +13,8 @@ jobs:
update:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
- uses: actions/checkout@v7
- uses: actions/setup-node@v7
with:
node-version: '22'
- name: Update README contributor tables
@@ -22,10 +22,9 @@ jobs:
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
run: node scripts/update-acknowledgments.mjs
- name: Open pull request
uses: peter-evans/create-pull-request@5f6978faf089d4d20b00c7766989d076bb2fc7f1 # v8
uses: peter-evans/create-pull-request@v8
with:
# GITHUB_TOKEN-authored pull requests never trigger CI, so required checks would never report.
token: ${{ secrets.ACKNOWLEDGMENTS_TOKEN || secrets.GITHUB_TOKEN }}
token: ${{ secrets.GITHUB_TOKEN }}
branch: chore/update-acknowledgments
delete-branch: true
title: 'Update acknowledgments contributors / 更新致谢贡献者'
@@ -33,6 +32,6 @@ jobs:
body: |
Refreshes the README acknowledgments tables from the GitHub contributors API.
- Source: `repos/esengine/DeepSeek-Reasonix/contributors?per_page=50&anon=1`
- The repository owner, declared aliases, tool identities, bots and entries without a GitHub login are left out (`scripts/credit-identities.json`).
- Source: `repos/esengine/DeepSeek-Reasonix/contributors?per_page=20&anon=1`
- Anonymous contributors are shown without email addresses.
labels: documentation
+2 -2
View File
@@ -24,10 +24,10 @@ jobs:
if: github.repository == 'esengine/DeepSeek-Reasonix'
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
- uses: actions/checkout@v7
with:
fetch-depth: 0
- uses: actions/setup-node@949feb2413d6458794dcd2491c4babbbce0c15c1 # v7
- uses: actions/setup-node@v7
with:
node-version: '22'
- name: Test chart generator
@@ -1,29 +0,0 @@
name: 1.x nightly monitoring
# A schedule runs only from the default branch. These monitors measure the 1.x
# desktop app, which lives on main-v2 alone, so their nightly runs are started
# from here on that branch. Their push and pull_request gates run on main-v2 as
# before.
on:
schedule:
- cron: "17 3 * * *"
- cron: "43 4 * * *"
workflow_dispatch:
permissions:
actions: write
jobs:
dispatch:
runs-on: ubuntu-latest
steps:
- name: App memory screening
if: github.event_name == 'workflow_dispatch' || github.event.schedule == '17 3 * * *'
env:
GH_TOKEN: ${{ github.token }}
run: gh workflow run app-memory.yml --repo "$GITHUB_REPOSITORY" --ref main-v2
- name: Diagnostic overhead monitoring
if: github.event_name == 'workflow_dispatch' || github.event.schedule == '43 4 * * *'
env:
GH_TOKEN: ${{ github.token }}
run: gh workflow run diagnostic-overhead.yml --repo "$GITHUB_REPOSITORY" --ref main-v2
@@ -1,84 +0,0 @@
name: Withdraw a Studio version
# Pulls a published Studio release out of the update catalog (or puts it back),
# so installs stop being offered it without waiting for a new release. Only the
# catalog changes: the release's files stay where they are, and a withdrawn
# release can be restored with the same run.
on:
workflow_dispatch:
inputs:
version:
description: "Version to withdraw or restore, for example 2.36.0"
required: true
type: string
reason:
description: "Why it is withdrawn (shown to users on that version); ignored when restoring"
required: false
type: string
restore:
description: "Put a withdrawn version back instead of withdrawing it"
required: false
type: boolean
default: false
concurrency:
group: studio-catalog
cancel-in-progress: false
jobs:
withdraw:
name: ${{ inputs.restore && 'restore' || 'withdraw' }} ${{ inputs.version }}
runs-on: ubuntu-latest
timeout-minutes: 10
environment: studio-release
permissions:
contents: write
steps:
- uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
with:
ref: studio
persist-credentials: false
- name: Update the Studio catalog
env:
VERSION: ${{ inputs.version }}
REASON: ${{ inputs.reason }}
RESTORE: ${{ inputs.restore }}
R2_BUCKET: ${{ secrets.R2_BUCKET }}
HAS_R2: ${{ secrets.R2_ACCESS_KEY_ID != '' && secrets.R2_SECRET_ACCESS_KEY != '' && secrets.R2_ACCOUNT_ID != '' && secrets.R2_BUCKET != '' }}
run: |
if [ "$HAS_R2" != "true" ]; then
echo "::error::R2 credentials are not configured, so the catalog cannot be changed"
exit 1
fi
if ! [[ "$VERSION" =~ ^[0-9]+\.[0-9]+\.[0-9]+(-[0-9A-Za-z.]+)?$ ]]; then
echo "::error::version must look like 2.36.0"
exit 1
fi
aws configure set aws_access_key_id "${{ secrets.R2_ACCESS_KEY_ID }}"
aws configure set aws_secret_access_key "${{ secrets.R2_SECRET_ACCESS_KEY }}"
aws configure set default.region auto
endpoint="https://${{ secrets.R2_ACCOUNT_ID }}.r2.cloudflarestorage.com"
at="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
current=/tmp/studio-versions-current.json
next=/tmp/studio-versions-next.json
aws s3 cp "s3://${R2_BUCKET}/studio/versions.json" "$current" --endpoint-url "$endpoint"
if [ "$RESTORE" = "true" ]; then
bash scripts/withdraw-version.sh "$current" --restore "$VERSION" "$at" > "$next"
jq -e --arg v "$VERSION" '.versions | map(select((.version | ltrimstr("v")) == $v)) | length == 1' "$next" >/dev/null
else
bash scripts/withdraw-version.sh "$current" "$VERSION" "$REASON" "$at" > "$next"
jq -e --arg v "$VERSION" '(.versions | map(select((.version | ltrimstr("v")) == $v)) | length == 0) and (.withdrawn | map(select((.version | ltrimstr("v")) == $v)) | length == 1)' "$next" >/dev/null
fi
aws s3 cp "$next" "s3://${R2_BUCKET}/studio/versions.json" \
--endpoint-url "$endpoint" \
--content-type "application/json; charset=utf-8" \
--cache-control "public, max-age=300, stale-if-error=86400"
echo "Studio catalog updated for $VERSION (restore=$RESTORE)"
- name: Keep the GitHub release off the latest slot
if: ${{ !inputs.restore }}
env:
GH_TOKEN: ${{ github.token }}
VERSION: ${{ inputs.version }}
run: gh release edit "studio-v${VERSION}" --prerelease --latest=false --repo "$GITHUB_REPOSITORY"
+8 -27
View File
@@ -1,6 +1,4 @@
# Build artifacts. A `go build` of one command writes its binary into the
# working directory, so every main package in the root module can drop its own
# name here; repolint's build-artifact rule holds the two lists together.
# Build artifacts
/bin/
/dist/
/stage/
@@ -9,23 +7,11 @@
/reasonix.exe
/e2ebench
/e2ebench.exe
/context-retrieval
/compaction
/context-maintenance-e2e
/runtime-resume
/extension-protocol-gen
/reasonix-plugin-example
/mcp-line-counter-kit
/reasonix-studio-host
/signpath-contract
/studio-delta
/covergate
/pricecheck
/repolint
/workgroupassay
/write_heartbeat
/catalog-detector
/tool-surface-cache
# Website (Astro) build output
/site/dist/
/site/.astro/
/site/_preview.png
# Go test/build scratch
*.test
@@ -77,6 +63,7 @@ desktop/**/.reasonix/
Thumbs.db
.idea/
.vscode/
/site/public/av/
benchmarks/context-maintenance-e2e/run/
@@ -86,13 +73,7 @@ temp/
# Official theme art intermediate outputs / caches
scripts/official-theme-art/out/
scripts/official-theme-art/__pycache__/
site/.generated/
# Committed pointer at REASONIX.md; overrides a global CLAUDE.md ignore rule.
!/CLAUDE.md
# Go build output left in the tree by `go build ./tools/...`. The ~ form is
# what a linker leaves behind when it cannot replace a binary that is still
# running, which on Windows is every rebuild while Studio is open.
*.exe
*.exe~
/relayfake
+1 -1
View File
@@ -16,7 +16,7 @@ builds:
ldflags:
# Inject version identity for `reasonix version --verbose/--json` (Integration D/E).
# buildTimeUTC is the goreleaser run time (UTC), not vcs.time.
- -s -w -X main.version={{ .Tag }} -X main.gitCommit={{ .ShortCommit }} -X main.buildTimeUTC={{ .Date }} -X reasonix/internal/tools/productdocs.linkedVersion={{ .Tag }} -X reasonix/internal/tools/productdocs.linkedRevision={{ .Commit }}
- -s -w -X main.version={{ .Tag }} -X main.gitCommit={{ .ShortCommit }} -X main.buildTimeUTC={{ .Date }} -X reasonix/internal/productdocs.linkedVersion={{ .Tag }} -X reasonix/internal/productdocs.linkedRevision={{ .Commit }}
goos: [darwin, linux, windows]
goarch: [amd64, arm64]
@@ -0,0 +1,13 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- Canary-only configuration. SignPath test certificates intentionally do not
chain to a Windows trusted root, so authenticode-verify cannot be used for
the already test-signed payload. The workflow verifies that every payload
executable is signed and hash-identical to the rebuilt portable package.
Stable and RC releases use windows-installer-v2 instead. -->
<artifact-configuration xmlns="http://signpath.io/artifact-configuration/v1">
<zip-file>
<pe-file path="*installer*.exe">
<authenticode-sign />
</pe-file>
</zip-file>
</artifact-configuration>
@@ -0,0 +1,30 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- Sign the finished NSIS container only when the exact payload used to build it
is present and already Authenticode-signed. GitHub Actions wraps this flat
bundle in a zip before SignPath receives it. Keep this versioned separately
so provisioning it cannot break releases still using windows-installer. -->
<artifact-configuration xmlns="http://signpath.io/artifact-configuration/v1">
<zip-file>
<pe-file path="*installer*.exe">
<authenticode-sign />
</pe-file>
<pe-file path="reasonix-desktop.exe">
<authenticode-verify />
</pe-file>
<pe-file path="reasonix-guard.exe">
<authenticode-verify />
</pe-file>
<pe-file path="reasonix-launcher.exe">
<authenticode-verify />
</pe-file>
<pe-file path="reasonix-update-helper.exe">
<authenticode-verify />
</pe-file>
<pe-file path="reasonix-cli.exe">
<authenticode-verify />
</pe-file>
<pe-file path="reasonix-uninstall.exe">
<authenticode-verify />
</pe-file>
</zip-file>
</artifact-configuration>
@@ -0,0 +1,13 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- GitHub Actions wraps every uploaded artifact in a zip, so the installer arrives
zipped even though it is a single file — match and sign the .exe inside. -->
<artifact-configuration xmlns="http://signpath.io/artifact-configuration/v1">
<zip-file>
<pe-file-set>
<include path="*installer*.exe" min-matches="1" max-matches="1" />
<for-each>
<authenticode-sign />
</for-each>
</pe-file-set>
</zip-file>
</artifact-configuration>
@@ -0,0 +1,25 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- Sign every executable that the NSIS installer writes to disk. GitHub Actions
wraps the flat payload directory in this zip before SignPath receives it. -->
<artifact-configuration xmlns="http://signpath.io/artifact-configuration/v1">
<zip-file>
<pe-file path="reasonix-desktop.exe">
<authenticode-sign />
</pe-file>
<pe-file path="reasonix-guard.exe">
<authenticode-sign />
</pe-file>
<pe-file path="reasonix-launcher.exe">
<authenticode-sign />
</pe-file>
<pe-file path="reasonix-update-helper.exe">
<authenticode-sign />
</pe-file>
<pe-file path="reasonix-cli.exe">
<authenticode-sign />
</pe-file>
<pe-file path="reasonix-uninstall.exe">
<authenticode-sign />
</pe-file>
</zip-file>
</artifact-configuration>
+10 -11
View File
@@ -3,18 +3,17 @@ project_slug: DeepSeek-Reasonix
signing_policy_slug: release-signing
repository_url: https://github.com/esengine/DeepSeek-Reasonix.git
allowed_branch_names:
- studio
- main-v2
allowed_build_definitions:
- .github/workflows/release-studio.yml
- .github/workflows/studio-certum-signing-smoke.yml
- .github/workflows/release-stable.yml
- .github/workflows/release-desktop.yml
fingerprint_files:
- .github/workflows/release-studio.yml
- .github/workflows/studio-certum-signing-smoke.yml
- .github/actions/setup-certum/action.yml
- scripts/install-smoke.ps1
- scripts/sign-certum.ps1
- .github/workflows/release-stable.yml
- .github/workflows/release-stable-trigger.yml
- .github/workflows/release-desktop.yml
- .signpath/artifact-configurations/windows-payload.xml
- .signpath/artifact-configurations/windows-installer-v2.xml
- scripts/complete-signpath-request.ps1
- scripts/package-windows-desktop.sh
- scripts/verify-windows-authenticode.ps1
- scripts/verify-windows-package.ps1
- scripts/windows-signing-lib.ps1
- desktop/electron/electron-builder.yml
- cmd/signpath-contract/main.go
+1
View File
@@ -0,0 +1 @@
v2.13.0
-6
View File
@@ -1,6 +0,0 @@
# Reasonix
The project's standing instructions live in `REASONIX.md` — one file, read by
both Reasonix and Codex. Add rules there, never here.
@REASONIX.md
+362
View File
@@ -0,0 +1,362 @@
# Changelog
All notable changes to the Go line (Reasonix 1.0+) are recorded here. The legacy
`0.x` TypeScript history lives on the [`v1`](https://github.com/esengine/DeepSeek-Reasonix/tree/v1)
branch.
## Unreleased
### Fixed
- **v1.24.2 session snapshot & recovery root fix:** Keep PR #7982's WAL/CAS/lease
safety foundation, but replace process-level "I hold a lease" ownership with a
generation-bound `SessionWriteAuthority`. Same-revision tool-preview/load
reshapes no longer false-diverge; recovery files are bounded to one path per
writer/lineage; empty checkpoints heal from their own WAL; projection lineage
rebinds across upgrade/model switch and inherits across recovery forks without
changing provider-visible prompt bytes. Catalog upgrades to disposable
`session-catalog/v3.sqlite` with recovery lineage roles
(`normal|covered_copy|adopted|diverged`); covered idle copies move to the
recoverable `.trash` using a 15-minute idle threshold applied on two early
sweeps (at startup and ~20 minutes later), then a 24-hour threshold on the
6-hour background ticker; independent diverged branches stay and are listed
for user choice. v1/v2 catalogs are
left byte-unchanged for coexistence/downgrade.
**v1.24.1** only hid/reclaimed already-created covered copies and fixed Windows
flash-window startup; **v1.24.2** stops the misclassification source and repairs
existing user directories without rewriting authoritative JSONL/WAL/sidecar data.
- Goal now runs continuously by default: the former 16-round per-Run boundary,
10/20/40 cross-Run quotas, default wall-clock budget, and numeric
no-progress/Todo-stall pauses no longer stop valid work. Progress guards still
detect repeated host outcomes and zero-evidence work, but redirect the model
to re-plan instead of producing `goal_run_budget` or `goal_stuck`. Explicit
`[agent].goal_token_budget`, `--max-steps`, positive time/cost budgets, manual
pause/stop, genuine user/external blockers, and evaluator fail-closed behavior
remain available. The Goal token budget defaults to `0` (off); resuming its
`budget_spend` pause grants a fresh slice without clearing cumulative usage.
Goal status reports turns, provider requests, tokens, the optional configured
token threshold, and cumulative active work time. Bot `max_steps` also
defaults to `0` (continuous), while positive user configuration is enforced.
- Removed numeric Goal pauses in existing sidecars automatically normalize to
`running` without sending a model request. Active Goal sidecars write
`turnsLimit: -1` as a downgrade-safe unlimited sentinel while public runtime
APIs retain deprecated limit fields as `0`. The migration preserves unknown
fields, todos, checkpoints, usage, evidence, and historical metadata.
- Goal is now the sole long-task runtime. Historical AutoResearch sidecars
migrate transactionally into Goals with research compatibility metadata. Invalid archives block
fail closed and remain read-only, retaining the task id and compatibility mode
for a restart or `/goal resume` retry; successful Goal-only sidecars omit the
old task id and write an explicit downgrade fence so previous readers cannot
reactivate the removed runtime.
- Context-dependent workflow tools now share one host-side execution boundary.
Goal, Plan sign-off, and background-job calls cannot reach permissions,
hooks, leases, or Execute outside their owning context; mixed batches execute
valid calls once and stop safely after one repair. Child agents also isolate
inherited Goal, Jobs, and live memory queues, while persisted tool identity
records the effective child schema projection.
- **Issue #7575:** Linux Bash under bubblewrap no longer mounts a fresh empty
`--tmpfs /tmp` on every call. Consecutive commands in the same logical session
now share a private temporary directory (bound at `/tmp` on Linux, exported via
`TMPDIR`/`TMP`/`TEMP` on all platforms) without exposing the host public
temporary root. `/new`, `/clear`, resume of another session, and branch
switches rotate the directory; model/settings hot rebuilds keep it. Sub-agent
runs get independent directories. Temporary files are not durable across process
restarts.
### Added
- Added `[ui].show_turn_usage` so CLI/TUI users can hide per-request token and
cost receipts from transcript scrollback without disabling usage accounting.
## [1.20.0] — 2026-08-05
Extension kernel, Task Monitor, and safer Goal completion.
Compact decision surfaces, local decision receipts, unified extension kernel,
native Task Monitor, bounded sub-agent progress, Goal fail-closed completion,
MiMo and DashScope Responses fixes, SSH remote access simplification, and
multiple Desktop stability improvements.
### Highlights
- **Unified Extension Kernel and Extension Protocol v1**: Immutable runtime
snapshots, fail-atomic reload, Plugin Manifest v1 (prompts, themes, full-trust
code runtimes), stable JSON-RPC sidecar protocol, interceptor dispatch,
streaming provider adapter, structured UI, and Go SDK.
- **Native Task Monitor**: Monitor agent tasks natively in CLI and Desktop with
lifecycle semantics and session-scoped summary view.
- **Bounded Sub-agent Progress Forwarding**: Forward structured progress for
`task`, `parallel_tasks`, and `fleet` without flooding the parent stream.
Renders nested lifecycle cards in Desktop and stable per-child transcript
slots in CLI.
- **Goal Completion Fail-Closed**: Replace free-form Goal footer markers with a
stable `update_goal` tool and epoch-scoped per-turn reports. Centralized
completion logic with bounded evaluator, progress-aware budgets, and
pause/resume controls.
- **Ablation Subsystem Switches**: Switch subsystems off behind one shared
vocabulary for controlled experiments. Includes planner, subagent, retrieval,
evidence, and compaction.
- **Benchmark Cost per Solved Task**: Report cost per solved task, tokens per
solved, median wall time, and failure-class breakdown in e2e reports.
- **Compact Decision Surfaces and Local Receipts**: Compact footer decision-card
layout with bounded scroll, dense action rows, and overflow disclosure.
Record bounded Ask, approval, and recovery decisions as local transcript
receipts.
- **Simplified SSH Remote Access**: Remove Remote Workbench protocol and
stacks; reuse CLI/Serve remote model. Desktop opens per-host native web child
windows via SSH. Keyless remote Serve setup with loopback-only page.
- **Model Usage Charts with Primer Palette**: Replace monochrome accent ramp
with GitHub Primer data-viz two-set categorical palette. Fix donut overflow
on hover and keyboard accessibility.
- **Cross-platform Extension and Task Monitor Reliability**: Make
content-reference eviction deterministic, reject Unix and Windows absolute
plugin paths consistently, stabilize parallel-task cancellation, and restore
reliable Windows validation for Task Monitor and remote provider setup.
- **MiMo and DashScope Responses Wire Alignment**: Fix multi-turn tool loops,
reasoning round-trip, JSON output for MiMo; fix DashScope second-turn 400
error, all-zero usage suppression, and vendor-aware cache TTL.
- **Desktop Stability Fixes**: Recover stuck updates and legacy WebKit, contain
macOS alias repair startup crashes, keep composer overflow stacks readable,
and harden account verification and community flows.
- **Remote Web Recovery After SSH Drops**: Add integration regression test for
SSH drop, forward recovery, and window reload. Document transient outage
behavior.
- **CI: Auto-minimize Activity-Farming Spam Comments**: Detect and minimize
template spam comments from non-contributor accounts based on structural
signals.
### Added
- Added Extension Protocol v1 and the unified extension kernel: installed or
linked sidecars can contribute tools, skills, commands, hooks, MCP servers,
providers, interceptors, and structured UI surfaces through a versioned
NDJSON contract and the public Go SDK. CLI, Desktop, ACP, and Serve support
fail-atomic runtime reloads; Serve also renders extension surfaces and lists
extension-hosted providers without exposing credentials.
- Added the structured Goal completion protocol: the always-registered
`update_goal` tool (continue/complete/blocked with reason and next_action)
replaces the `[goal:*]` footer markers. The Goal FSM is now the exclusive
cross-turn decision point and validates every complete claim against Delivery
readiness; when the model submits no report, an independent bounded evaluator
(recovery_model → guardian_model → main model, no tools/history, usage
attributed to `goal-evaluator`) judges the turn once, and any evaluator
failure pauses the goal instead of continuing silently.
- Added Goal budget classes with safe pauses: simple 10 turns / 200k tokens,
write 20 turns / 400k tokens, AutoResearch 40 turns / 800k tokens, and a
4-turn no-host-verifiable-progress gate. Pauses keep all Goal state; `/goal
resume` continues and adds one slice of the current class when the pause was
budget-related. `/goal status` shows the full turn/token/no-progress runtime,
and `/goal pause` manually suspends a running Goal.
- Added the `goalRuntime` nested view to the desktop Meta, the remote protocol
(`session/goal/pause` operation, `goalRuntime` DTO on session meta), and the
ACP status payload; the desktop Composer goal menu shows the runtime summary
with distinct pause/end/resume actions.
### Changed
- Delivery no longer retries final-answer readiness with hidden model messages:
a plain Delivery run ends on the first unsatisfied final answer and surfaces
the recovery card, while a Goal + Delivery run has the Goal FSM absorb the
failure and continue under budget with the missing requirements as the next
turn's prompt. Historical `[goal:*]` footers are stripped from old transcripts
for display only and never participate in state decisions.
- Added a **Remote SSH** module (VS Code Remote-SSH style): a user-global
`[remote]` host config, `reasonix remote` CLI (add/list/remove/import/test/
connect/status/forward/serve/fs) and `/remote` slash command, an SSH transport
with trust-on-first-use host-key verification, keepalive + exponential-backoff
reconnect, `-L`/`-R` port forwarding, and SFTP file access. `connect`
bootstraps a persistent `reasonix serve` on the remote host and tunnels its
loopback port so the full agent runs remotely. The desktop app adds a
**Settings -> Remote SSH** host manager, a remote file browser/editor, a
port-forwarding panel, and a status-bar connection chip. Linux/macOS remotes.
- Added `reasonix serve --port-file/--token-file/--pid-file` so a supervised
headless serve can bind an ephemeral port and read its auth token from a file
(keeping it out of `ps`).
- Added an authenticated, loopback-only Provider setup page for `reasonix
serve`. A Serve whose selected Provider is missing its API key now remains
reachable, stores the submitted key in that host's Reasonix credential file,
and rebuilds the active controller in place without restarting Serve.
- Added Claude Code-style searchable CLI pickers for models, providers, and
sessions, with arrow, Vim, and `Ctrl+P` / `Ctrl+N` navigation.
- Added `-p` / `--print`, `text`, `json`, and `stream-json` output modes for
one-shot use and automation.
- Added session-scoped `--allowed-tools`, repeatable `--add-dir`, Claude-compatible
permission modes, flexible `--resume [QUERY]`, and the `--copy` resume escape
hatch.
- Added `/status` details for the active model, effort, cache, Git state,
background jobs, work profile, and provider balance where available.
- Remote SSH workspaces now open as a standalone remote web window again.
Opening a workspace from the status bar or the Remote Server tab starts or
reuses the remote `reasonix serve`, tunnels its loopback port, and opens the
Serve web client in a dedicated per-host window. The remote web page uses
the provider configuration and API keys on the **remote** host; the desktop
no longer exposes its local providers to remote hosts. If the selected remote
Provider is missing its API key, the window opens a setup page that saves the
key only on that host and then opens the normal Serve UI. The Remote Workbench
protocol, its Provider Broker, and the same-window remote projection were
removed. Legacy mirror and provider-trust files are not deleted
automatically; Settings -> Remote SSH shows a cleanup card when they exist.
The hidden `remote attach-workspace`, `remote runtime-workbench`, and
`remote workbench-build-id` commands now fail with a pointer to
`reasonix remote connect <host> --open`.
- Automatic Plan Mode has been retired. Plan Mode is now always entered through
an explicit user choice, and the one-time config v5 upgrade removes legacy
`agent.auto_plan` and `agent.auto_plan_classifier` values so upgraded users
receive the same behavior as new users.
- `Shift+Tab` now cycles CLI safe modes from Ask to Auto to Plan, while YOLO
remains an independent `Ctrl+Y` toggle.
- Model, provider, resume, and approval menus now use consistent row selection;
slash completion, help, aliases, and dispatch share one command registry.
- The full-screen CLI composer now uses theme-accented borders and a slim bar
cursor by default, grows within the available terminal height, scrolls long
drafts independently, and preserves selections across explicit image paste.
- The persistent CLI footer now uses a responsive, theme-aware layout for
interaction state, model, effort, localized work mode, Git identity, cache,
context, compaction headroom, jobs, and balance. Narrow terminals move or
compact complete groups instead of clipping labels.
- CLI clipboard actions now separate terminal-native text paste from explicit
image paste: `Ctrl+V` on macOS/Linux, `Alt+V` on Windows, or `/paste-image`.
Local transcript copy verifies the native clipboard write, while SSH uses a
clearly labelled OSC 52 fallback.
- Runtime rebuilds after model, effort, or work-mode changes now preserve the
conversation, session permission overrides, additional directories, and
session lease ownership.
- Agent execution now monitors host-observed Todo progress automatically. A
stalled current item receives a recovery nudge after 8 tool-call rounds with
no new completion, unique read, command, or mutation, and pauses with saved
work after 16. Exact repeats do not renew the progress lease; real work does.
Two-level task lists keep the single in_progress contract: the active
sub-step is the only current item while its phase stays pending, and the
phase becomes in_progress to sign off only after all of its sub-steps are
completed. A level-1 sub-step with no phase header above it is rejected.
Executor and planner rounds now use automatic progress management. Retired
`[agent].max_steps` and `planner_max_steps` keys remain parseable for upgrades,
but are ignored and removed by a one-time migration so stale hidden limits
cannot truncate new behavior. One-off CLI and unattended bot limits remain.
### Fixed
- Fixed long parallel sub-agent research being silently lost when combined
`parallel_tasks` or `fleet` answers exceeded the single-tool output limit.
Persisted sessions now keep each child transcript independently, return a
bounded fair preview plus stable reference for every result, and page full
answers through the conversation-scoped `read_subagent_result` tool.
- Fixed Remote Workbench failing with only `initialize: workbench-desktop:
connection closed` on fresh or cross-platform SSH hosts. Desktop now proves
the exact Host CLI Build ID, provisions the matching verified release without
requiring remote npm, runs the managed binary explicitly, and preserves a
safe structured bootstrap error when the remote command exits early.
- Hardened Bash permission reuse for dynamic and indirect execution. Parameter/arithmetic expansions,
assignments, redirects, heredocs, and globs can only be remembered as exact
`Bash=<literal>` rules, while still using Auto's normal fallback. Nested or
indirect execution now requires a human in interactive Ask/Auto and fails
closed in headless Ask/Auto/DontAsk. Broad Bash rules, Guardian/hook allows,
and the approved-plan window can no longer silently authorize that stricter
class; YOLO remains the explicit full-access bypass and sandbox enforcement
is unchanged.
- Fixed Desktop sessions incorrectly locking themselves during Goal + Delivery
mode changes, controller rebuilds, duplicate-tab restore, and background
reattachment. Desktop now keeps one process-local runtime owner per canonical
session, fences stale controller events by runtime epoch, blocks sends until
that runtime is ready, and scopes single-instance ownership to
`REASONIX_HOME` instead of the executable path. Switching saved sessions is
now transactional: a target build, restore, or lease failure leaves the
current controller, lease, path, mode profile, and runtime epoch untouched.
- Stabilized the desktop rich composer caret after skill and plugin invocation
tags. DOM→model and model→DOM selection mapping now treat invocation chips as
zero-length atoms while still counting user text that lands inside the NBSP
caret anchor (common on Windows WebView2), restore both selection ends, and
recover the insertion point from a `beforeinput` snapshot when the browser
temporarily loses selection — so mid-text edits no longer jump to the end.
- Isolated the Windows desktop WebView2 shell from stale system proxies, so an
exited proxy client cannot leave the embedded UI hidden during startup. If
WebView2 still does not reach DOM-ready within 15 seconds, Reasonix now shows
the native window with a recovery prompt instead of appearing not to launch.
Remote Markdown images are fetched by the backend with Reasonix's proxy
configuration instead of bypassing that proxy through the isolated WebView.
- Restored captured-mouse right-click text paste, made composer drag selection
copy through the verified native clipboard path, and kept non-Git footer
telemetry left-aligned without reserving an empty data band.
- Restored stateful MCP behavior after the v1.17.13 regression: user-added
servers work without extra trust settings (including delivery-mode on-demand
calls), repository-provided servers use one exact launch confirmation, and
stdio tools reuse one persistent process so browser sessions survive across
calls without repeated startup latency. The former trust/reverify/catalog
management UI and CLI are removed.
- Localized persistent-footer labels and displayed work-mode values in English,
Simplified Chinese, and Traditional Chinese, while keeping command arguments
stable.
- Restored the `0.53` content boundary: model output, tool output, session
transcripts, recovery branches, and background-job artifacts retain their
original text instead of being rewritten by heuristic secret redaction.
Credential masking remains in key-entry summaries and explicit diagnostic or
session-cleanup paths. Transcript-bearing session/job sidecars are kept
private (`0600`, with private job directories), and the retired
`redact_tool_output` setting is removed with a one-time upgrade notice.
### Notes
- Full bilingual release notes:
<https://reasonix.io/changelog/v1.20.0/> ·
[GitHub release](https://github.com/esengine/DeepSeek-Reasonix/releases/tag/desktop-v1.20.0).
- The detailed entries below accumulated on `main-v2` after 1.0.0 and shipped
across 1.1.0–1.20.0; per-version attribution lives in the per-version release
notes linked above.
## 1.1.0 – 1.19.7
Per-version entries for the intermediate releases are published in the
[bilingual release notes](https://reasonix.io/changelog/) and on the
[GitHub releases page](https://github.com/esengine/DeepSeek-Reasonix/releases).
## [1.0.0] — 2026-06-03
First stable release — a **ground-up rewrite in Go**. Not an upgrade of the `0.x`
TypeScript line; a new codebase that becomes the default (`main-v2`).
### Highlights
- **Go kernel**: a single static binary (CGO-free), cross-compiled for
darwin/linux/windows on amd64 + arm64. Distributed via npm (the package wraps
the native binary), Homebrew (`esengine/reasonix` tap), and release archives;
no Node runtime needed to run it.
- **Agent core**: the loop, built-in tools (read/write/edit/multi_edit/glob/grep/
ls/bash/web_fetch/todo_write), permission gate, sandboxed bash, and the
DeepSeek prefix-cache–oriented design.
- **Subagents**: `task` plus explore/research/review/security_review skill agents.
- **Skills & hooks**: Claude-Code-style skills (`internal/skill`) and hooks
(`internal/hook`), symlink-aware and slash-integrated.
- **MCP client**: connect external servers over stdio / Streamable HTTP; reads
`[[plugins]]` and a Claude-Code `.mcp.json`.
- **Code intelligence via CodeGraph**: a tree-sitter symbol/call graph
(`codegraph_*` tools) replaces embedding semantic search — no embedding service
or API cost. Fetched into a local cache on first use (or `reasonix codegraph
install`) and indexed in the background, so installs and startup stay fast.
- **Plan mode** with evidence-backed step sign-off (`complete_step`).
- **Memory**: `REASONIX.md` hierarchy + auto-memory, folded into the cache-stable
prefix.
- **ACP** (`reasonix acp`) and an HTTP/SSE server frontend; desktop app (Wails).
### Fixed
- **File encoding support restored** — GBK/GB18030 (and other non-UTF-8) files
can now be read, edited, and grepped correctly. The v2 rewrite had dropped
v1's encoding detection; files in CJK Windows charsets were silently misread
or rejected as binary. The read/edit/write round-trip now preserves the
original file encoding. (#2637)
### Notes
- Versions: the legacy TypeScript line stays in `0.x`; the Go line starts at
`1.0.0`. See [docs/MIGRATING.md](docs/MIGRATING.md).
- Release archives ship a bare binary; CodeGraph is fetched on first use. Windows
support for the fetched runtime is unverified — install `codegraph` on PATH if
the auto-fetch doesn't resolve there.
[1.20.0]: https://github.com/esengine/DeepSeek-Reasonix/releases/tag/desktop-v1.20.0
[1.0.0]: https://github.com/esengine/DeepSeek-Reasonix/releases/tag/v1.0.0
+42 -91
View File
@@ -5,10 +5,11 @@ everything you need to get started.
## Prerequisites
- **Go 1.26+** — the version `go.mod` requires
- **Go 1.25+** — the project targets the latest stable Go release
- **Git** — for version control
- **Node.js 24+ and pnpm 10** (optional) — only if you work on Studio
(`desktop/`)
- **Node.js** (optional) — only if you work on the desktop app (`desktop/`)
- **Wails CLI** (desktop only) — run `make wails-install` so the CLI matches
`desktop/go.mod`
## Getting started
@@ -24,22 +25,22 @@ go test ./... # runs the full test suite
| Directory | Purpose |
|-----------|---------|
| `cmd/reasonix` | CLI entry point |
| `internal/runtime/agent` | Agent loop, session, coordinator |
| `internal/frontend/cli` | TUI, subcommands, setup wizard |
| `internal/session/control` | Transport-agnostic controller |
| `internal/contract/config` | TOML configuration loading |
| `internal/tools/builtin` | Built-in tools (bash, read_file, …) |
| `internal/contract/provider` | Model-backend abstraction |
| `internal/model/openai` | OpenAI-compatible provider |
| `internal/ext/plugin` | MCP client (stdio + HTTP) |
| `internal/contract/event` | Typed event stream |
| `internal/ext/hook` | Shell hooks (PreToolUse, …) |
| `internal/state/memory` | REASONIX.md hierarchy + auto-memory |
| `internal/ext/skill` | Skill discovery from Markdown |
| `internal/safety/sandbox` | OS-level sandboxing |
| `internal/frontend/serve` | HTTP/SSE server frontend |
| `internal/state/checkpoint` | Snapshot-based rewind |
| `desktop/` | Studio: the Electron shell, the SPA it serves (separate Go module) |
| `internal/agent` | Agent loop, session, coordinator |
| `internal/cli` | TUI, subcommands, setup wizard |
| `internal/control` | Transport-agnostic controller |
| `internal/config` | TOML configuration loading |
| `internal/tool/builtin` | Built-in tools (bash, read_file, …) |
| `internal/provider` | Model-backend abstraction |
| `internal/provider/openai` | OpenAI-compatible provider |
| `internal/plugin` | MCP client (stdio + HTTP) |
| `internal/event` | Typed event stream |
| `internal/hook` | Shell hooks (PreToolUse, …) |
| `internal/memory` | REASONIX.md hierarchy + auto-memory |
| `internal/skill` | Skill discovery from Markdown |
| `internal/sandbox` | OS-level sandboxing |
| `internal/serve` | HTTP/SSE server frontend |
| `internal/checkpoint` | Snapshot-based rewind |
| `desktop/` | Wails-based desktop app (separate Go module) |
| `docs/` | Engineering spec, migration guide |
### Dependency direction
@@ -79,10 +80,11 @@ REASONIX_HOME=/tmp/reasonix-dev go run ./cmd/reasonix
# REASONIX_HOME=/tmp/reasonix-dev ./bin/reasonix
```
**Studio**
**Desktop**
```bash
REASONIX_HOME=/tmp/reasonix-dev-isolated make studio
cd desktop && wails build
REASONIX_HOME=/tmp/reasonix-dev-isolated build/bin/reasonix-desktop
```
On Windows, use `$env:REASONIX_HOME` in PowerShell or `set REASONIX_HOME=` in
@@ -104,35 +106,26 @@ serialization, compaction, or MCP/tool registration need explicit cache review.
For these changes:
- Keep system prompt changes low-frequency and require explicit review.
- Say in the PR what the change does to the provider-visible prefix, and why
that cost is worth paying.
- Fill the PR body `Cache-impact:` line with `none`, `low`, `medium`, or `high`
plus the reason.
- Fill the PR body `Cache-guard:` line with the focused guard test/command added
or run, or explain why an existing guard covers the change.
- Fill `System-prompt-review:` when system prompt, memory prefix, output style,
or skill index behavior changes.
- Prefer focused guard tests near the changed surface; `scripts/cache-guard.sh`
remains the broader release-level cache-hit check.
CI enforces this metadata for cache-sensitive paths so prompt/tool prefix churn
is called out before review.
### Running tests
```bash
go test ./... # all tests
go test ./internal/runtime/agent/ -v # verbose, one package
go test ./internal/tools/builtin/ -run TestGrep # one test
go test ./internal/agent/ -v # verbose, one package
go test ./internal/tool/builtin/ -run TestGrep # one test
```
`testenv.IsolateUserState` installs a disposable user home and config-edit lock
scope. Default-path tests can clear `REASONIX_HOME` without losing the test
registry, while explicitly isolated Reasonix homes retain their own locks.
`REASONIX_TEST_HOME` carries the scope to helper subprocesses. A helper that
skips the isolation guard must install the inherited home with
`testscope.InstallHome` before running tests.
The environment variable alone cannot activate a test scope. Production
executables still use the OS account's home for shared config locks.
The guard removes its scratch home on normal completion. Scratch homes are
created under the original home, so an interrupted run can still leave
`.reasonix-test-home-*` directories; this is separate from persistent
config-edit locks in the real `.reasonix/locks` directory.
### Code style
- `gofmt` is enforced by CI — format before committing
@@ -155,81 +148,39 @@ ci: add golangci-lint and govulncheck
## Adding a new built-in tool
1. Create `internal/tools/builtin/mytool.go`
1. Create `internal/tool/builtin/mytool.go`
2. Implement the `tool.Tool` interface: `Name()`, `Description()`, `Schema()`, `ReadOnly()`, `Execute()`
3. Register via `func init() { tool.RegisterBuiltin(myTool{}) }`
4. Add tests in `internal/tools/builtin/builtin_test.go` or a separate `mytool_test.go`
4. Add tests in `internal/tool/builtin/builtin_test.go` or a separate `mytool_test.go`
5. The tool is automatically available — `main` blank-imports `builtin`
## Adding a new model provider
(For MCP tool servers see `internal/ext/plugin` instead — that's a different layer.)
(For MCP tool servers see `internal/plugin` instead — that's a different layer.)
1. Create `internal/contract/provider/myprovider/`
1. Create `internal/provider/myprovider/`
2. Implement `provider.Provider`: `Name()`, `Stream()`
3. Register via `func init() { provider.Register("mykind", New) }`
4. The provider is available from config with `kind = "mykind"`
## Adding i18n strings
1. Add the field to `internal/base/i18n/i18n.go` (`Messages` struct)
2. Add the value in `internal/base/i18n/messages_en.go` and `messages_zh.go`
1. Add the field to `internal/i18n/i18n.go` (`Messages` struct)
2. Add the value in `internal/i18n/messages_en.go` and `messages_zh.go`
3. The `TestCatalogsComplete` test will fail if you miss a locale
## Submitting changes
Reasonix ships on two lines (see the
[version roadmap](https://github.com/esengine/DeepSeek-Reasonix/discussions/10748)),
and each takes different changes:
| Line | Branch | Takes |
| --- | --- | --- |
| 2.x | `studio` | Features and fixes: active development |
| 1.x | `main-v2` | Bug fixes, provider/API compatibility, release/updater, security, platform stability |
A fix that matters on both lines lands on `main-v2` and is reviewed for
porting to `studio`; do not open the same PR against both.
1. Fork the repository
2. Create a feature branch from the branch your change targets
2. Create a feature branch from `main-v2`
3. Make your changes with tests
4. Ensure `go test ./...` passes
5. Ensure `gofmt -l .` shows no changes
6. Submit a pull request to that same branch
## Pull request policy
**Related changes.** If you have several related changes of one pattern, a
tracking issue plus one pull request is easiest to review.
**Issue first for features.** A change that adds behaviour rather than fixing a
defect should link an issue the maintainers have agreed to before the work
starts. A fix for a reproducible defect may open directly, with the failing test
in the PR. Documentation-only and test-only PRs are reviewed at lower priority.
**Review tiers.**
- Changes under the `sensitive:` paths declared in `REASONIX.md` get a full
review plus a security review, regardless of author.
- Every PR carries the four sections of the PR template: Cause, Blast radius,
Neighbouring behaviours tested, Why this layer.
- The new test must fail without the change.
- A UI change (`desktop/frontend-next/src` or the website pages) fills the
template's UI section: screenshots before and after, light and dark, wide
and phone width.
- The same section carries the line "Entries moved, hidden or removed".
- Hiding or removing a visible control needs a linked issue and a release-notes
line naming it, and updates `docs/STUDIO_PARITY.md` for a 1.x feature.
- Review holds these; no check reads them.
**Transparency.** The PR template has an optional checkbox for disclosing AI
assistance. It is for transparency only and does not change how a change is
reviewed.
6. Submit a pull request to `main-v2`
## Reporting issues
Open an issue on GitHub with:
- Which line (1.x or 2.x) and the exact version
- Steps to reproduce
- Expected vs actual behavior
- Go version and OS
+16 -46
View File
@@ -1,4 +1,4 @@
VERSION := $(shell git describe --tags --always 2>/dev/null | sed 's/^studio-//' || echo dev)
VERSION := $(shell git describe --tags --always 2>/dev/null || echo dev)
BUILD_TIME_UTC := $(shell date -u +%Y-%m-%dT%H:%M:%SZ)
GIT_COMMIT := $(shell git rev-parse --short=12 HEAD 2>/dev/null || echo unknown)
LDFLAGS := -s -w \
@@ -8,8 +8,9 @@ LDFLAGS := -s -w \
GOEXE := $(shell go env GOEXE)
# One pin for the Makefile and the CI lint job; see .github/workflows/ci.yml.
GOLANGCI_VERSION := $(shell cat .golangci-version)
WAILS_VERSION := $(shell tr -d '[:space:]' < .wails-version)
.PHONY: build vet fmt lint lint-go lint-install lint-cross lint-update coverage-gate coverage-gate-update pricecheck acp-conformance check test studio-test sdk-test sdk-test-race hooks cross clean studio
.PHONY: build vet fmt lint lint-go lint-install lint-cross lint-update wails-install test desktop-test desktop-test-short desktop-test-times sdk-test sdk-test-race hooks cross clean
build:
CGO_ENABLED=0 go build -ldflags "$(LDFLAGS)" -o bin/reasonix$(GOEXE) ./cmd/reasonix
@@ -18,13 +19,6 @@ build:
vet:
go vet ./...
# Build Studio and launch it. The Electron shell is the one this runs: it is a
# real GUI app on every platform without being bundled first, so there is one
# launch path rather than a macOS one and an everywhere-else one.
studio:
bash desktop/electron/run-studio.sh
fmt:
gofmt -w .
@@ -33,6 +27,8 @@ fmt:
# particular never surface in `go vet`.
lint: lint-go
go run ./tools/repolint
bash scripts/check-wails-pin.sh
bash scripts/check-wails-pin.test.sh
lint-go:
@command -v golangci-lint >/dev/null || { echo "golangci-lint not installed; run: make lint-install"; exit 1; }
@@ -45,33 +41,12 @@ lint-go:
lint-install:
CGO_ENABLED=0 go install github.com/golangci/golangci-lint/v2/cmd/golangci-lint@$(GOLANGCI_VERSION)
# repolint over what the working tree changed. It exists as a target rather than
# a command to paste because the host recognizes `make check` as a verification
# it can read the result of, while an arbitrary `go run ./tools/...` is an
# unknown program it must assume writes — so only this form can be cited as
# evidence that the standards gate passed.
check:
go run ./tools/repolint -changed
lint-update:
go run ./tools/repolint -update
# The paths the project declared sensitive, held to the coverage they already
# have. It runs their tests, so it is its own target rather than part of the
# fast `make lint`. -update records the current numbers; lowering one needs
# -allow-drop, the way widening a repolint budget does.
coverage-gate:
go run ./tools/covergate
coverage-gate-update:
go run ./tools/covergate -update
# The vendors publish their rates on pages this reads and compares the table
# against. It reaches the network, so it stays out of lint and out of CI: run it
# before touching a rate and when a vendor announces a change. A page it cannot
# read exits non-zero — unread never counts as agreement.
pricecheck:
go run ./tools/pricecheck
wails-install:
bash scripts/check-wails-pin.sh
go install "github.com/wailsapp/wails/v2/cmd/wails@$(WAILS_VERSION)"
# Linting one GOOS leaves every //go:build windows and darwin file unchecked.
lint-cross:
@@ -81,22 +56,17 @@ lint-cross:
(cd $$2 && GOOS=$$1 golangci-lint run --timeout=5m ./...) || exit 1; \
done
# Every module, because a nested one is where a red test hides longest: the
# shell's route allowlist sat broken through a release, and `go test ./...` from
# the root cannot see it.
test: studio-test sdk-test
test:
go test ./...
# Drives `reasonix acp` with the official ACP TypeScript SDK and holds every
# message it writes to the schema that SDK ships. Needs node and network for
# `npm ci`, which runs without install scripts against a lockfile.
acp-conformance: build
cd tools/acp-conformance && npm ci --ignore-scripts --no-audit --no-fund
cd tools/acp-conformance && REASONIX_BIN=$(CURDIR)/bin/reasonix$(GOEXE) npm test
desktop-test:
cd desktop && go test .
# The shell is a nested module, so the root `test` target never reaches it.
studio-test:
cd desktop && go test ./...
desktop-test-short:
cd desktop && go test -short .
desktop-test-times:
cd desktop && go test -count=1 -json . | python3 ../scripts/desktop-test-times.py
sdk-test:
cd sdk/go && go test ./...
+64 -154
View File
@@ -18,12 +18,9 @@
<a href="https://esengine.github.io/DeepSeek-Reasonix/">Website</a>
&nbsp;·&nbsp;
<strong><a href="https://discord.gg/XF78rEME2D">Discord</a></strong>
&nbsp;·&nbsp;
<strong><a href="https://qm.qq.com/q/i59b0z2R8s">QQ</a></strong>
</p>
<p align="center">
<a href="https://github.com/esengine/DeepSeek-Reasonix/releases?q=studio-v&expanded=true"><img src="https://img.shields.io/github/v/release/esengine/DeepSeek-Reasonix.svg?filter=studio-v*&include_prereleases&style=flat-square&color=8250df&labelColor=161b22&label=studio%202.x" alt="Reasonix Studio 2.x"/></a>
<a href="https://www.npmjs.com/package/reasonix"><img src="https://img.shields.io/npm/v/reasonix.svg?style=flat-square&color=cb3837&labelColor=161b22&logo=npm&logoColor=white" alt="npm version"/></a>
<a href="https://github.com/esengine/DeepSeek-Reasonix/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/esengine/DeepSeek-Reasonix/ci.yml?style=flat-square&label=ci&labelColor=161b22&logo=githubactions&logoColor=white" alt="CI"/></a>
<a href="./LICENSE"><img src="https://img.shields.io/npm/l/reasonix.svg?style=flat-square&color=8b949e&labelColor=161b22" alt="license"/></a>
@@ -33,7 +30,6 @@
<a href="https://github.com/esengine/DeepSeek-Reasonix/graphs/contributors"><img src="https://img.shields.io/github/contributors/esengine/DeepSeek-Reasonix.svg?style=flat-square&color=bc8cff&labelColor=161b22&logo=github&logoColor=white" alt="contributors"/></a>
<a href="https://github.com/esengine/DeepSeek-Reasonix/discussions"><img src="https://img.shields.io/github/discussions/esengine/DeepSeek-Reasonix.svg?style=flat-square&color=58a6ff&labelColor=161b22&logo=github&logoColor=white" alt="Discussions"/></a>
<a href="https://discord.gg/XF78rEME2D"><img src="https://img.shields.io/badge/discord-join-5865F2.svg?style=flat-square&labelColor=161b22&logo=discord&logoColor=white" alt="Discord"/></a>
<a href="https://qm.qq.com/q/i59b0z2R8s"><img src="https://img.shields.io/badge/QQ%20group-1093562660-12B7F5.svg?style=flat-square&labelColor=161b22&logo=tencentqq&logoColor=white" alt="QQ group 1093562660"/></a>
</p>
<p align="center">
@@ -43,93 +39,23 @@
<br/>
<p align="center"><strong>Open source · MIT · runs on your computer, model of your choice</strong></p>
<h3 align="center">An open-source coding agent for your terminal, desktop, browser and editor.</h3>
<p align="center">It reads your project, edits files and runs commands and tests. Have it ask before each step, or work on its own within limits you set. Files its edit tools changed can be rewound. New here? Start with the Studio desktop app.</p>
<p align="center">Maintained by <strong>Huahui Yu</strong>.</p>
<p align="center"><strong>Open source · MIT · a single Go binary</strong></p>
<h3 align="center">A coding agent you can leave running.</h3>
<p align="center">One local engine, four ways in — terminal, desktop app, browser, or your editor over ACP. Plan mode, permissions, a workspace sandbox and per-turn checkpoints keep a long autonomous run something you can still read and undo.</p>
> [!IMPORTANT]
> **Community · 加入社区** — bilingual Discord for setup help (`#help` / `#求助`), workflow showcases, and feature ideas. → **<https://discord.gg/XF78rEME2D>**
>
> QQ group **DeepSeek-Reasonix官方群** (`1093562660`) → **<https://qm.qq.com/q/i59b0z2R8s>** ([QR code](./docs/assets/qq-group.svg))
> Douyin: **做游戏的小鱼** (Douyin ID `22703872788`) · [QR code](./docs/assets/douyin.png)
<div align="center">
<video src="https://github.com/user-attachments/assets/ab2f3878-e224-4931-8254-060e7695cfb9" controls preload="metadata" width="560"></video>
</div>
<br/>
## ❤️Sponsor
> [!IMPORTANT]
> **Community · 加入社区** — bilingual Discord for setup help (`#help` / `#求助`), workflow showcases, and feature ideas. → **<https://discord.gg/XF78rEME2D>**
<table><tr><td width="180" align="center" valign="middle"><a href="https://www.rightapi.ai/register?aff=e083bf73" target="_blank" rel="sponsored noopener noreferrer"><img src="./.github/sponsor/rightcode.png" alt="Right Code" width="64"></a></td><td valign="middle">Thanks to Right Code for sponsoring this project! Right Code reliably provides routing services for models such as Claude Code, Codex, and Gemini, with both pay-as-you-go and monthly subscription billing options available. Invoices are available upon top-up, and enterprise and team users can receive dedicated one-on-one support. Register via <a href="https://www.rightapi.ai/register?aff=e083bf73" target="_blank" rel="sponsored noopener noreferrer">this link</a>.</td></tr><tr><td width="180" align="center" valign="middle"><a href="https://www.packyapi.ai/register?aff=rIbQ" target="_blank" rel="sponsored noopener noreferrer"><picture><source media="(prefers-color-scheme: dark)" srcset="./.github/sponsor/packycode-dark.svg"><img src="./.github/sponsor/packycode-light.svg" alt="PackyCode" width="80"></picture></a></td><td valign="middle">Thanks to PackyCode for sponsoring this project! Access leading AI models through PackyCode with one API endpoint and one API key, with automatic failover and dedicated high-speed routes for Codex and Claude Code. Register via <a href="https://www.packyapi.ai/register?aff=rIbQ" target="_blank" rel="sponsored noopener noreferrer">this link</a>.</td></tr></table>
## What it is
Reasonix is an open-source AI coding assistant. The program runs on your computer; the AI model is a service you configure, which receives what it reads. You describe a task in plain language; it reads code, edits files and runs commands and tests inside the project folder you choose, and shows you each step.
You pick the model: a DeepSeek preset is built in, any OpenAI-compatible service is one config entry, and you pay the model provider directly.
## What you get
- **Read, edit, run.** It reads code, edits files and runs tests and commands, all inside your project folder.
- **Permission levels.** Have it ask every time, or let everyday actions run on their own; your forbid rules always hold. To see the approach first, use plan mode: it starts work only once you approve.
- **Rewind.** A file is saved before it changes, so you can rewind by conversation turn without touching your git history. Only files its edit tools changed are covered, not what a shell command did.
## 30-second start
Download, add a model key, open a project. New here? Start with the Studio desktop app.
1. Download and install the desktop app, Reasonix Studio, from the [releases page](https://github.com/esengine/DeepSeek-Reasonix/releases?q=studio-v&expanded=true).
2. Open it and enter your model provider and key when asked.
3. Open your project folder and write what you want, for example "add retry on failure to the http client".
4. At the level you chose, it stops and asks before anything that needs your say-so. Allow it to continue, or refuse.
Prefer a terminal? Download the `reasonix` command-line tool, run `reasonix setup` to pick a model, then run `reasonix`. See [Install](#install) and [Quick start](#quick-start).
## Four ways to use it
- **Desktop app (Studio).** A graphical app; the best place to start.
- **Terminal.** A full-screen text interface; `reasonix run "task"` also works in scripts.
- **Browser.** `reasonix web` opens a web interface on your own machine.
- **Editor.** Install the extension in VS Code; other editors that support ACP (Agent Client Protocol, a common way for an editor to connect to an AI assistant) can connect to `reasonix acp`.
All four are the same program underneath. What follows is for advanced users and contributors.
## Use it from DeepSeek Harness
DeepSeek Harness (`dsh`) can hand a task to Reasonix as a subagent over ACP. This repository publishes a plugin index for it:
1. Run `reasonix setup` on the machine that runs `dsh`. In the DSH Web UI turn on Custom plugin markets (Plugins, More) and add this index in the Discover tab: `https://raw.githubusercontent.com/esengine/DeepSeek-Reasonix/studio/integrations/dsh-plugin/market-index.json`
2. Install "Reasonix subagent" from it.
3. Open the plugin's page and turn its switch on. While it is off the model has no `reasonix` tool and may run `reasonix -p` in its own shell, which looks like a delegation and is not one.
Setup details, the permission policy and limits are in [Using Reasonix from DeepSeek Harness](./docs/DSH.md).
## Versions
Reasonix ships on two lines. See the [version roadmap announcement](https://github.com/esengine/DeepSeek-Reasonix/discussions/10748) for the reasoning behind the split.
| Line | Branch | Status | Get it |
| --- | --- | --- | --- |
| **Reasonix 2.x** | `studio` (this branch) | Active development | [Studio releases](https://github.com/esengine/DeepSeek-Reasonix/releases?q=studio-v&expanded=true) |
| **Reasonix 1.x** | [`main-v2`](https://github.com/esengine/DeepSeek-Reasonix/tree/main-v2) | Maintenance / stable | `npm i -g reasonix` · [desktop download](https://reasonix.io/?download=desktop#start) |
- **Want something stable?** Stay on 1.x. It keeps receiving bug fixes,
provider/API compatibility, updater and security fixes; its core architecture
is no longer being extended.
- **Want the new architecture?** Use 2.x Reasonix Studio. It is still moving
quickly; reports are welcome.
Valuable fixes, tests and behaviour from 1.x are reviewed one by one and ported
or reimplemented on 2.x where they fit the new architecture.
- [Moving from 1.x to 2.x](./docs/MIGRATING.md): what the two lines share, and
how to run both on one machine without losing sessions.
- [2.x roadmap](./docs/ROADMAP.md): what 2.x still has to deliver, how it ships,
and the decisions still open.
<br/>
## Features
Technical highlights for advanced users.
- **Config-driven.** Providers, the agent, enabled tools, and plugins are all
declared in `reasonix.toml`. No hardcoded models.
- **Multi-model & composable.** DeepSeek ships as a preset; any
@@ -147,73 +73,53 @@ Technical highlights for advanced users.
## Install
This branch is **Reasonix 2.x**. Its desktop app is Reasonix Studio; the CLI,
Studio, and editor integrations all run the same local Reasonix engine.
Choose the path that matches how you want to use Reasonix. The CLI/TUI,
desktop app, and VS Code extension all use the same local Reasonix engine.
### Reasonix Studio (2.x)
### Path A: CLI / TUI
Download the package for your platform from the latest
[Studio release](https://github.com/esengine/DeepSeek-Reasonix/releases?q=studio-v&expanded=true)
(tagged `studio-v2.*`; candidates such as `-rc.N` are marked as
pre-releases):
| Platform | Package | Architecture |
| --- | --- | --- |
| macOS | `.dmg` or `.zip` | Apple Silicon (`arm64`) / Intel (`amd64`) |
| Windows | Installer `-installer.exe` or portable `.zip` | x64 |
| Linux | `.deb`, `.rpm`, or `.AppImage` (download only) | x64 (`amd64`) / ARM64 (`arm64`, aarch64) |
Every package ships with a `.minisig` signature, and the release carries
`SHA256SUMS`. Once installed, Studio updates itself in place.
The same release also carries the 2.x `reasonix` CLI as archives for
`darwin|linux|windows × amd64|arm64`.
The Android build ships on that release as `ReasonixMobile-<version>.apk`
(`arm64`, Android 8.0 or newer). The kernel and the page run on the phone, and
the agent reaches your servers over SSH; pairing with a computer is done by
scanning the code it shows.
The APK is signed with the Android release key rather than minisign, so it
carries a `.sha256` and no `.minisig`.
When that key is absent the build is named
`ReasonixMobile-<version>-unsigned.apk`. Android refuses to install an unsigned
package, so those builds stay a workflow artifact for whoever wants to sign one,
and the release carries no APK at all.
### Reasonix 1.x (stable)
The 1.x CLI installs through npm on any supported platform, or Homebrew on
macOS:
Install the native binary through npm on any supported platform, or use
Homebrew on macOS:
```sh
npm i -g reasonix # any OS; pulls the prebuilt native binary
brew install esengine/reasonix/reasonix # macOS
```
The 1.x desktop app is on the
[official download page](https://reasonix.io/?download=desktop#start). Windows
installers are code-signed with a Certum code-signing certificate.
Prebuilt archives (`darwin|linux|windows × amd64|arm64`) and `SHA256SUMS` are on
every [GitHub release](https://github.com/esengine/DeepSeek-Reasonix/releases).
### VS Code extension
### Path B: Desktop app
The extension does not bundle the CLI; it starts your local `reasonix acp`
backend and adds native chat, editor context, tool-call approvals, model
selection, and workspace sessions. Install the 1.x CLI first.
Use the [official download page](https://reasonix.io/?download=desktop#start)
for the latest desktop build.
| Platform | Package | Architecture |
| --- | --- | --- |
| macOS | Universal `.dmg` or `.zip` | Apple Silicon / Intel |
| Windows | Installer `.exe` or portable `.zip` | x64 / ARM64 |
| Linux | `.deb` or `.tar.gz` | x64 |
Windows installers are code-signed through [SignPath.io](https://signpath.io/)
with a free certificate provided by the [SignPath Foundation](https://signpath.org/).
### Path C: VS Code extension
Complete Path A first. The extension does not bundle the CLI; it starts your
local `reasonix acp` backend and adds native chat, editor context, tool-call
approvals, model selection, and workspace sessions.
- **VS Code:** [install from Visual Studio Marketplace](https://marketplace.visualstudio.com/items?itemName=SivanLiu.reasonix-agent)
- **VSCodium / Eclipse Theia:** [install from Open VSX Registry](https://open-vsx.org/extension/SivanLiu/reasonix-agent)
- **Extension ID:** `SivanLiu.reasonix-agent` · [source and usage guide](https://github.com/SivanCola/reasonix-vscode)
### Build from source
### Path D: Build from source
Clone the repository; `studio` builds 2.x and `main-v2` builds 1.x:
Clone the repository first:
```sh
git clone https://github.com/esengine/DeepSeek-Reasonix.git
cd DeepSeek-Reasonix
git switch studio
```
#### CLI
@@ -226,27 +132,28 @@ make build # -> bin/reasonix(.exe)
make cross # -> dist/ (darwin|linux|windows × amd64|arm64)
```
#### Studio
#### Desktop
Studio additionally requires **Node 24+ and pnpm 10**
(`npm install -g pnpm@10`) for the frontend.
The desktop build additionally requires:
- **Node 24+ and pnpm 10** (`npm install -g pnpm@10`) for the frontend
- **Wails CLI** matching the shared `.wails-version` pin
```sh
make studio
make wails-install
cd desktop
wails build
```
See the [Studio build guide](desktop/README.md#prerequisites) for platform
See the [desktop build guide](desktop/README.md#prerequisites) for platform
webview dependencies and Linux build tags.
## Quick start
### Reasonix Studio
Install and launch Studio, then connect a provider and model in the app. No CLI
setup is needed.
### CLI / TUI
These commands are for the CLI/TUI installed through Path A:
```sh
reasonix setup # configure a provider and model
reasonix # start an interactive session
@@ -256,6 +163,13 @@ reasonix run "implement the TODOs in main.go"
In an interactive session, run `/init` when you want Reasonix to create project
instructions.
### Desktop app
Download the installer for your platform from the
[official download page](https://reasonix.io/?download=desktop#start), install
and launch Reasonix, then configure a provider and model in the app. The CLI
commands above are not required for the desktop app.
For advanced CLI usage and configuration, see the **[CLI reference](./docs/CLI.md)**,
**[Guide](./docs/GUIDE.md)**, and
**[configuration paths](./docs/CONFIG_PATHS.md)**.
@@ -263,19 +177,16 @@ For advanced CLI usage and configuration, see the **[CLI reference](./docs/CLI.m
## Documentation
- **Getting started:** [Guide](./docs/GUIDE.md) · [CLI reference](./docs/CLI.md) ·
[Configuration paths](./docs/CONFIG_PATHS.md) · [ACP editor integration](./docs/ACP.md) ·
[Use from DeepSeek Harness](./docs/DSH.md)
[Configuration paths](./docs/CONFIG_PATHS.md) · [ACP editor integration](./docs/ACP.md)
- **Features & troubleshooting:** [Subagent profiles](./docs/SUBAGENT_PROFILES.md) ·
[Context Engine v2](./docs/SESSION_MEMORY_RETRIEVAL.md) ·
[Capability diagnostics](./docs/CAPABILITY_DIAGNOSTICS.md) ·
[Recovery and updates](./docs/RECOVERY.md) ·
[Recovery and updates](./docs/RECOVERY.md) · [Bot guide](./docs/BOT_GUIDE.md) ·
[Checkpoints & rewind](./docs/CHECKPOINTS.md)
- **Agent presets:** [Session Agent compositions](./docs/AGENT_COMPOSITIONS.md)
- **Engineering & migration:** [Spec](./docs/SPEC.md) ·
[Task contracts & pause policy](./docs/TASK_CONTRACT.md) ·
[Tool contract](./docs/TOOL_CONTRACT.md) · [Moving from 1.x to 2.x](./docs/MIGRATING.md) ·
[2.x roadmap](./docs/ROADMAP.md)
- **Extension development:** [Skills](./docs/SKILLS.md) · [Extensions](./docs/EXTENSIONS.md) ·
[Tool contract](./docs/TOOL_CONTRACT.md) · [Migrating from 0.x](./docs/MIGRATING.md)
- **Extension development:** [Extensions](./docs/EXTENSIONS.md) ·
[Plugin packages and Manifest v1](./docs/PLUGIN_PACKAGES.md) ·
[Extension Protocol](./docs/EXTENSION_PROTOCOL.md) ·
[Go SDK and starter](./sdk/go/README.md)
@@ -301,11 +212,11 @@ A small list of folks whose work has shaped Reasonix the most — the current to
<!-- reasonix-top-contributors:start -->
| Contributor | Contributor | Contributor | Contributor |
| --- | --- | --- | --- |
| [**SivanCola**](https://github.com/SivanCola) | [**KHG420**](https://github.com/KHG420) | [**ttmouse**](https://github.com/ttmouse) | [**chuiwenwei**](https://github.com/chuiwenwei) |
| [**lifu963**](https://github.com/lifu963) | [**XTLine**](https://github.com/XTLine) | [**HUQIANTAO**](https://github.com/HUQIANTAO) | [**xxoingr**](https://github.com/xxoingr) |
| [**Harbor404**](https://github.com/Harbor404) | [**GTC2080**](https://github.com/GTC2080) | [**mchenziyi**](https://github.com/mchenziyi) | [**Li-Charles-One**](https://github.com/Li-Charles-One) |
| [**light-front-theory**](https://github.com/light-front-theory) | [**imcreazy**](https://github.com/imcreazy) | [**eghrhegpe**](https://github.com/eghrhegpe) | [**noreply**](https://github.com/noreply) |
| [**par73e**](https://github.com/par73e) | [**Bernardxu123**](https://github.com/Bernardxu123) | [**lancy69**](https://github.com/lancy69) | [**Guan0923**](https://github.com/Guan0923) |
| [**SivanCola**](https://github.com/SivanCola) | [**esengine**](https://github.com/esengine) | [**ttmouse**](https://github.com/ttmouse) | [**lifu963**](https://github.com/lifu963) |
| **reasonix** | [**HUQIANTAO**](https://github.com/HUQIANTAO) | [**GTC2080**](https://github.com/GTC2080) | [**light-front-theory**](https://github.com/light-front-theory) |
| **merge-order-check** | [**Li-Charles-One**](https://github.com/Li-Charles-One) | [**eghrhegpe**](https://github.com/eghrhegpe) | **wufengfan** |
| [**CVEngineer66**](https://github.com/CVEngineer66) | [**dependabot\[bot\]**](https://github.com/apps/dependabot) | [**lanshi17**](https://github.com/lanshi17) | [**SuMuxi66**](https://github.com/SuMuxi66) |
| [**CnsMaple**](https://github.com/CnsMaple) | [**cyq1017**](https://github.com/cyq1017) | [**JesonChou**](https://github.com/JesonChou) | [**XTLine**](https://github.com/XTLine) |
<!-- reasonix-top-contributors:end -->
Special thanks to [**Bernardxu123**](https://github.com/Bernardxu123) for designing the project logo and intro video.
@@ -335,9 +246,8 @@ coffee, not a contract — donations don't buy feature priority or change how
issues get triaged.
- **International** — PayPal: [paypal.me/yuhuahui](https://paypal.me/yuhuahui)
- **国内** — 微信支付 / 支付宝(扫码)
- **国内** — 微信支付(扫码)
<p align="center">
<img src=".github/sponsor/wechat-pay.jpg" alt="WeChat Pay QR code" width="180"/>
<img src=".github/sponsor/alipay.jpg" alt="Alipay QR code" width="180"/>
</p>
+65 -166
View File
@@ -9,21 +9,18 @@
&nbsp;·&nbsp;
<a href="./docs/GUIDE.zh-CN.md">指南</a>
&nbsp;·&nbsp;
<a href="./docs/ACP.md">ACP</a>
<a href="./docs/ACP.zh-CN.md">ACP</a>
&nbsp;·&nbsp;
<a href="./docs/EXTENSIONS.md">扩展开发</a>
<a href="./docs/EXTENSIONS.zh-CN.md">扩展开发</a>
&nbsp;·&nbsp;
<a href="./docs/SPEC.md">规格</a>
<a href="./docs/SPEC.zh-CN.md">规格</a>
&nbsp;·&nbsp;
<a href="https://esengine.github.io/DeepSeek-Reasonix/">官方网站</a>
&nbsp;·&nbsp;
<strong><a href="https://discord.gg/XF78rEME2D">Discord</a></strong>
&nbsp;·&nbsp;
<strong><a href="https://qm.qq.com/q/i59b0z2R8s">QQ</a></strong>
</p>
<p align="center">
<a href="https://github.com/esengine/DeepSeek-Reasonix/releases?q=studio-v&expanded=true"><img src="https://img.shields.io/github/v/release/esengine/DeepSeek-Reasonix.svg?filter=studio-v*&include_prereleases&style=flat-square&color=8250df&labelColor=161b22&label=studio%202.x" alt="Reasonix Studio 2.x"/></a>
<a href="https://www.npmjs.com/package/reasonix"><img src="https://img.shields.io/npm/v/reasonix.svg?style=flat-square&color=cb3837&labelColor=161b22&logo=npm&logoColor=white" alt="npm version"/></a>
<a href="https://github.com/esengine/DeepSeek-Reasonix/actions/workflows/ci.yml"><img src="https://img.shields.io/github/actions/workflow/status/esengine/DeepSeek-Reasonix/ci.yml?style=flat-square&label=ci&labelColor=161b22&logo=githubactions&logoColor=white" alt="CI"/></a>
<a href="./LICENSE"><img src="https://img.shields.io/npm/l/reasonix.svg?style=flat-square&color=8b949e&labelColor=161b22" alt="license"/></a>
@@ -33,7 +30,6 @@
<a href="https://github.com/esengine/DeepSeek-Reasonix/graphs/contributors"><img src="https://img.shields.io/github/contributors/esengine/DeepSeek-Reasonix.svg?style=flat-square&color=bc8cff&labelColor=161b22&logo=github&logoColor=white" alt="contributors"/></a>
<a href="https://github.com/esengine/DeepSeek-Reasonix/discussions"><img src="https://img.shields.io/github/discussions/esengine/DeepSeek-Reasonix.svg?style=flat-square&color=58a6ff&labelColor=161b22&logo=github&logoColor=white" alt="Discussions"/></a>
<a href="https://discord.gg/XF78rEME2D"><img src="https://img.shields.io/badge/discord-join-5865F2.svg?style=flat-square&labelColor=161b22&logo=discord&logoColor=white" alt="Discord"/></a>
<a href="https://qm.qq.com/q/i59b0z2R8s"><img src="https://img.shields.io/badge/QQ%20group-1093562660-12B7F5.svg?style=flat-square&labelColor=161b22&logo=tencentqq&logoColor=white" alt="QQ group 1093562660"/></a>
</p>
<p align="center">
@@ -43,87 +39,21 @@
<br/>
<p align="center"><strong>开源 · MIT · 程序跑在你电脑上,模型自选</strong></p>
<h3 align="center">终端、桌面、浏览器、编辑器里都能用的 AI 编程助手。</h3>
<p align="center">它会读你的项目、改文件、跑命令和测试。每一步先问你,或在你划定的范围内自己做,由你选;它用编辑工具改过的文件可以退回去。第一次用,从桌面应用 Studio 开始。</p>
<p align="center">维护者:<strong>Huahui Yu</strong>。</p>
<p align="center"><strong>开源 · MIT · 单个 Go 二进制</strong></p>
<h3 align="center">可以一直开着跑的编码 Agent。</h3>
<p align="center">一套本地引擎,四个入口——终端、桌面端、浏览器,或通过 ACP 接入你的编辑器。计划模式、权限、工作区沙箱与逐轮 checkpoint,让长时间自治运行始终可读、可撤销。</p>
<div align="center">
<video src="https://github.com/user-attachments/assets/ab2f3878-e224-4931-8254-060e7695cfb9" controls preload="metadata" width="560"></video>
</div>
<br/>
> [!IMPORTANT]
> **加入社区 · Community** — 双语 Discord,提供安装答疑(`#help` / `#求助`)、工作流展示与功能想法。→ **<https://discord.gg/XF78rEME2D>**
>
> QQ 群 **DeepSeek-Reasonix官方群**(群号 `1093562660`)→ **<https://qm.qq.com/q/i59b0z2R8s>**([二维码](./docs/assets/qq-group.svg))
> 抖音:**做游戏的小鱼**(抖音号 `22703872788`)· [二维码](./docs/assets/douyin.png)
## ❤️赞助商
<table><tr><td width="180" align="center" valign="middle"><a href="https://www.rightapi.ai/register?aff=e083bf73" target="_blank" rel="sponsored noopener noreferrer"><img src="./.github/sponsor/rightcode.png" alt="Right Code" width="64"></a></td><td valign="middle">感谢 Right Code 赞助了本项目!Right Code 稳定提供 Claude Code、Codex、Gemini 等模型的中转服务,并可选按量、包月两种计费模式。充值即可开票,企业、团队用户一对一对接。通过<a href="https://www.rightapi.ai/register?aff=e083bf73" target="_blank" rel="sponsored noopener noreferrer">此链接</a>注册。</td></tr><tr><td width="180" align="center" valign="middle"><a href="https://www.packyapi.ai/register?aff=rIbQ" target="_blank" rel="sponsored noopener noreferrer"><picture><source media="(prefers-color-scheme: dark)" srcset="./.github/sponsor/packycode-dark.svg"><img src="./.github/sponsor/packycode-light.svg" alt="PackyCode" width="80"></picture></a></td><td valign="middle">感谢 PackyCode 赞助了本项目!PackyCode 是一家稳定、高效的 API 中转服务商,统一域名、统一密钥、智能容灾切换。通过<a href="https://www.packyapi.ai/register?aff=rIbQ" target="_blank" rel="sponsored noopener noreferrer">此链接</a>注册。</td></tr></table>
## 它是什么
Reasonix 是一个开源的 AI 编程助手。程序跑在你的电脑上;AI 模型是你自己配置的服务,它读到的内容会发给这个服务。你用大白话描述任务,它在你选的项目文件夹里读代码、改文件、跑命令和测试,每一步都摆给你看。
模型你自己选:内置 DeepSeek 预设,其他兼容 OpenAI 接口的服务填一条配置就行,费用直接付给模型服务商。
## 你能得到什么
- **读、改、跑**:读你的项目、改文件、跑测试和命令,都在你选的项目文件夹里。
- **权限分档**:每次都问你,或让日常操作自动放行;你设的禁止规则始终有效。想先看方案,就开计划模式,批准后它才动手。
- **可回退**:改文件前先存快照,按对话轮次回退,不碰 git 历史。只覆盖编辑工具改的文件,命令行命令造成的改动不在内。
## 30 秒上手
下载、填模型密钥、打开项目。第一次用,从桌面应用 Studio 开始。
1. 从[发布页](https://github.com/esengine/DeepSeek-Reasonix/releases?q=studio-v&expanded=true)下载并安装桌面应用 Reasonix Studio。
2. 打开应用,按提示填入模型服务商和密钥。
3. 打开你的项目文件夹,写下想做的事,例如“给 http 客户端加上失败重试”。
4. 按你选的档位,该问你的操作它会先停下来问;同意就继续,不同意就拒绝。
更喜欢终端的话,下载 `reasonix` 命令行后运行 `reasonix setup` 配好模型,再运行 `reasonix` 即可。详见[安装](#安装)和[快速开始](#快速开始)。
## 四种用法
- **桌面应用(Studio)**:图形界面,适合第一次使用。
- **终端**:全屏文字界面,也能用 `reasonix run "任务"` 在脚本里运行。
- **浏览器**:`reasonix web` 在本机打开网页界面。
- **编辑器**:VS Code 安装扩展即可;其他支持 ACP(Agent Client Protocol,编辑器接入某个 AI 助手的通用约定)的编辑器,可连接 `reasonix acp`。
四种用法背后是同一套程序。下面的内容写给进阶用户和贡献者。
## 在 DeepSeek Harness 里使用
DeepSeek Harness(`dsh`)可以通过 ACP 把任务交给 Reasonix 作为子智能体去做。本仓库为此发布了一个插件索引:
1. 在运行 `dsh` 的机器上先运行 `reasonix setup`。在 DSH 网页界面打开“自定义插件市场”(插件,更多),再到 Discover 页添加这个索引:`https://raw.githubusercontent.com/esengine/DeepSeek-Reasonix/studio/integrations/dsh-plugin/market-index.json`
2. 从中安装“Reasonix subagent”。
3. 打开这个插件的页面,把它的开关打开。关着的时候模型没有 `reasonix` 工具,可能改在自己的 shell 里直接运行 `reasonix -p`,看起来像委派,其实不是。
配置细节、权限策略和限制见 [在 DeepSeek Harness 中使用 Reasonix](./docs/DSH.md)。
## 版本
Reasonix 分为两条版本线,调整的原因见[版本路线公告](https://github.com/esengine/DeepSeek-Reasonix/discussions/10748)。
| 版本线 | 分支 | 状态 | 获取方式 |
| --- | --- | --- | --- |
| **Reasonix 2.x** | `studio`(当前分支) | 活跃开发 | [Studio 发布页](https://github.com/esengine/DeepSeek-Reasonix/releases?q=studio-v&expanded=true) |
| **Reasonix 1.x** | [`main-v2`](https://github.com/esengine/DeepSeek-Reasonix/tree/main-v2) | 维护 / 稳定 | `npm i -g reasonix` · [桌面端下载](https://reasonix.io/?download=desktop#start) |
- **想要稳定**:继续使用 1.x。它会持续收到 Bug 修复、Provider / API 兼容、
更新器和安全修复,但不再扩展核心架构。
- **想体验新架构**:使用 2.x Reasonix Studio。它仍在快速迭代,欢迎反馈问题。
1.x 中有价值的修复、测试和行为会逐项 review,适合新架构的会迁移或重新实现到 2.x。
- [从 1.x 迁移到 2.x](./docs/MIGRATING.md)(英文):两条线共用哪些数据,以及如何在同一台
机器上同时使用而不丢会话。
- [2.x 路线图](./docs/ROADMAP.md)(英文):2.x 还要交付什么、如何发布、哪些决定尚未做出。
## 特性
以下是写给进阶用户的技术要点。
- **配置驱动**:provider、agent、启用的工具、插件全部在 `reasonix.toml` 中声明,
内核无硬编码模型。
- **多模型 · 可组合**:DeepSeek 作为预设内置;任何 OpenAI 兼容
@@ -137,93 +67,58 @@ Reasonix 分为两条版本线,调整的原因见[版本路线公告](https://
## 安装
当前分支是 **Reasonix 2.x**,桌面端是 Reasonix Studio;CLI、Studio 和编辑器集成
都运行同一套本地 Reasonix 引擎。
选择适合你的使用路径。CLI/TUI、桌面端和 VS Code 扩展都使用同一套本地
Reasonix 引擎。
### Reasonix Studio(2.x)
### 路径 A:CLI / TUI
从最新的 [Studio 发布](https://github.com/esengine/DeepSeek-Reasonix/releases?q=studio-v&expanded=true)
下载对应平台的安装包(标签为 `studio-v2.*`;`-rc.N` 等候选版本标为预发布):
| 平台 | 安装包 | 架构 |
| --- | --- | --- |
| macOS | `.dmg` 或 `.zip` | Apple Silicon(`arm64`)/ Intel(`amd64`) |
| Windows | 安装器 `-installer.exe` 或便携 `.zip` | x64 |
| Linux | `.deb`、`.rpm` 或 `.AppImage`(只提供下载) | x64(`amd64`)/ ARM64(`arm64`,aarch64) |
每个安装包都附带 `.minisig` 签名,发布页另有 `SHA256SUMS`。安装后 Studio 会在
应用内自行更新。
同一个发布页也提供 2.x 的 `reasonix` CLI 归档(`darwin|linux|windows × amd64|arm64`)。
安卓版在同一个发布页上,文件名是 `ReasonixMobile-<版本>.apk`(`arm64`,Android 8.0 及以上)。
这个 APK 用 Android 发布密钥签名,不走 minisign,因此只有一个 `.sha256`,没有 `.minisig`。
该密钥缺失时产物名为 `ReasonixMobile-<版本>-unsigned.apk`。Android 不接受未签名的包,
这类构建只留作 workflow 产物供自行签名,发布页上不会挂 APK。
内核与界面都跑在手机上,Agent 通过 SSH 连接你自己的服务器;与电脑配对时扫它显示的二维码。
### Reasonix 1.x(稳定版)
1.x CLI 在任意支持的平台上都可以通过 npm 安装,macOS 也可以使用 Homebrew:
任意支持的平台都可以通过 npm 安装原生二进制;macOS 也可以使用 Homebrew:
```sh
npm i -g reasonix # 任意系统;自动拉取对应平台的原生二进制
brew install esengine/reasonix/reasonix # macOS
```
1.x 桌面端请前往[官方下载页](https://reasonix.io/?download=desktop#start)。
Windows 安装器使用 Certum 代码签名证书签名。
预编译归档(`darwin|linux|windows × amd64|arm64`)和 `SHA256SUMS` 见每个
[GitHub release](https://github.com/esengine/DeepSeek-Reasonix/releases)。
### VS Code 扩展
### 路径 B:桌面端
扩展不内置 CLI,而是启动本机的 `reasonix acp` 后端,并提供原生聊天、编辑器
上下文、工具调用审批、模型选择和工作区会话。请先安装 1.x CLI。
前往[官方下载页](https://reasonix.io/?download=desktop#start)获取最新桌面版本。
| 平台 | 安装包 | 架构 |
| --- | --- | --- |
| macOS | 通用 `.dmg` 或 `.zip` | Apple Silicon / Intel |
| Windows | 安装器 `.exe` 或便携 `.zip` | x64 / ARM64 |
| Linux | `.deb` 或 `.tar.gz` | x64 |
Windows 安装器通过 [SignPath.io](https://signpath.io/) 完成代码签名,证书由
[SignPath 基金会](https://signpath.org/) 免费提供。
### 路径 C:VS Code 扩展
请先完成路径 A。扩展不内置 CLI,而是启动本机的 `reasonix acp` 后端,
并提供原生聊天、编辑器上下文、工具调用审批、模型选择和工作区会话。
- **VS Code:** [从 Visual Studio Marketplace 安装](https://marketplace.visualstudio.com/items?itemName=SivanLiu.reasonix-agent)
- **VSCodium / Eclipse Theia:** [从 Open VSX Registry 安装](https://open-vsx.org/extension/SivanLiu/reasonix-agent)
- **扩展 ID:** `SivanLiu.reasonix-agent` · [源码与使用说明](https://github.com/SivanCola/reasonix-vscode)
### 从源码构建
克隆仓库;`studio` 构建 2.x,`main-v2` 构建 1.x:
### 路径 D:从源码构建
```sh
git clone https://github.com/esengine/DeepSeek-Reasonix.git
cd DeepSeek-Reasonix
git switch studio
```
#### CLI
CLI 构建需要 **Go 1.25+**。模块固定了 `toolchain`,保持 `GOTOOLCHAIN=auto`
即可让 Go 自动下载对应工具链,也可以手动安装。
```sh
make build # -> bin/reasonix(.exe)
make cross # -> dist/(darwin|linux|windows × amd64|arm64)
```
#### Studio
Studio 的前端还需要 **Node 24+ 与 pnpm 10**(`npm install -g pnpm@10`)。
```sh
make studio
```
各平台的 webview 依赖和 Linux 构建标签见 [Studio 构建指南](desktop/README.md#prerequisites)。
## 快速开始
### Reasonix Studio
安装并启动 Studio,然后在应用内连接 provider 和模型即可使用,无需配置 CLI。
### CLI / TUI
以下命令仅适用于通过路径 A 安装的 CLI/TUI:
```sh
reasonix setup # 配置 provider 和模型
reasonix # 启动交互式会话
@@ -232,29 +127,34 @@ reasonix run "把 main.go 里的 TODO 实现掉"
需要项目指令时,可在交互式会话中运行 `/init`。
### 桌面端
从[官方下载页](https://reasonix.io/?download=desktop#start)下载对应系统的安装包,
安装并启动 Reasonix,然后在应用内配置 provider 和模型即可使用。桌面端无需执行
上面的 CLI 命令。
CLI 进阶用法和详细配置见 **[CLI 命令参考](./docs/CLI.zh-CN.md)**、
**[指南](./docs/GUIDE.zh-CN.md)** 和
**[配置路径](./docs/CONFIG_PATHS.md)**。
**[配置路径](./docs/CONFIG_PATHS.zh-CN.md)**。
## 文档
- **开始使用:** [指南](./docs/GUIDE.zh-CN.md) ·
[CLI 命令参考](./docs/CLI.zh-CN.md) · [配置路径](./docs/CONFIG_PATHS.md) ·
[ACP 编辑器接入](./docs/ACP.md) · [在 DeepSeek Harness 中使用](./docs/DSH.md)
- **功能与排障:** [子智能体 Profile](./docs/SUBAGENT_PROFILES.md) ·
[Context Engine v2](./docs/SESSION_MEMORY_RETRIEVAL.md) ·
[能力诊断](./docs/CAPABILITY_DIAGNOSTICS.md) ·
[恢复与安全模式](./docs/RECOVERY.md) ·
[Checkpoints 与 rewind](./docs/CHECKPOINTS.md)
- **Agent 预设:** [会话级组合](./docs/AGENT_COMPOSITIONS.zh.md)
- **工程与迁移:** [规格](./docs/SPEC.md) ·
[任务合约与暂停策略](./docs/TASK_CONTRACT.md) ·
[工具合约](./docs/TOOL_CONTRACT.md) ·
[从 1.x 迁移到 2.x](./docs/MIGRATING.md) ·
[2.x 路线图](./docs/ROADMAP.md)
- **扩展开发:** [扩展概览](./docs/EXTENSIONS.md) ·
[插件包与 Manifest v1](./docs/PLUGIN_PACKAGES.md) ·
[Extension Protocol](./docs/EXTENSION_PROTOCOL.md) ·
[CLI 命令参考](./docs/CLI.zh-CN.md) · [配置路径](./docs/CONFIG_PATHS.zh-CN.md) ·
[ACP 编辑器接入](./docs/ACP.zh-CN.md)
- **功能与排障:** [子智能体 Profile](./docs/SUBAGENT_PROFILES.zh-CN.md) ·
[Context Engine v2](./docs/SESSION_MEMORY_RETRIEVAL.zh-CN.md) ·
[能力诊断](./docs/CAPABILITY_DIAGNOSTICS.zh-CN.md) ·
[恢复与安全模式](./docs/RECOVERY.zh-CN.md) ·
[机器人使用指南](./docs/BOT_GUIDE.zh-CN.md) ·
[Checkpoints 与 rewind](./docs/CHECKPOINTS.zh-CN.md)
- **工程与迁移:** [规格](./docs/SPEC.zh-CN.md) ·
[任务合约与暂停策略](./docs/TASK_CONTRACT.zh-CN.md) ·
[工具合约](./docs/TOOL_CONTRACT.zh-CN.md) ·
[从 0.x 迁移](./docs/MIGRATING.zh-CN.md)
- **扩展开发:** [扩展概览](./docs/EXTENSIONS.zh-CN.md) ·
[插件包与 Manifest v1](./docs/PLUGIN_PACKAGES.zh-CN.md) ·
[Extension Protocol](./docs/EXTENSION_PROTOCOL.zh-CN.md) ·
[Go SDK 与 starter](./sdk/go/README.md)
## Star 趋势
@@ -278,11 +178,11 @@ CLI 进阶用法和详细配置见 **[CLI 命令参考](./docs/CLI.zh-CN.md)**
<!-- reasonix-top-contributors:start -->
| Contributor | Contributor | Contributor | Contributor |
| --- | --- | --- | --- |
| [**SivanCola**](https://github.com/SivanCola) | [**KHG420**](https://github.com/KHG420) | [**ttmouse**](https://github.com/ttmouse) | [**chuiwenwei**](https://github.com/chuiwenwei) |
| [**lifu963**](https://github.com/lifu963) | [**XTLine**](https://github.com/XTLine) | [**HUQIANTAO**](https://github.com/HUQIANTAO) | [**xxoingr**](https://github.com/xxoingr) |
| [**Harbor404**](https://github.com/Harbor404) | [**GTC2080**](https://github.com/GTC2080) | [**mchenziyi**](https://github.com/mchenziyi) | [**Li-Charles-One**](https://github.com/Li-Charles-One) |
| [**light-front-theory**](https://github.com/light-front-theory) | [**imcreazy**](https://github.com/imcreazy) | [**eghrhegpe**](https://github.com/eghrhegpe) | [**noreply**](https://github.com/noreply) |
| [**par73e**](https://github.com/par73e) | [**Bernardxu123**](https://github.com/Bernardxu123) | [**lancy69**](https://github.com/lancy69) | [**Guan0923**](https://github.com/Guan0923) |
| [**SivanCola**](https://github.com/SivanCola) | [**esengine**](https://github.com/esengine) | [**ttmouse**](https://github.com/ttmouse) | [**lifu963**](https://github.com/lifu963) |
| **reasonix** | [**HUQIANTAO**](https://github.com/HUQIANTAO) | [**GTC2080**](https://github.com/GTC2080) | [**light-front-theory**](https://github.com/light-front-theory) |
| **merge-order-check** | [**Li-Charles-One**](https://github.com/Li-Charles-One) | [**eghrhegpe**](https://github.com/eghrhegpe) | **wufengfan** |
| [**CVEngineer66**](https://github.com/CVEngineer66) | [**dependabot\[bot\]**](https://github.com/apps/dependabot) | [**lanshi17**](https://github.com/lanshi17) | [**SuMuxi66**](https://github.com/SuMuxi66) |
| [**CnsMaple**](https://github.com/CnsMaple) | [**cyq1017**](https://github.com/cyq1017) | [**JesonChou**](https://github.com/JesonChou) | [**XTLine**](https://github.com/XTLine) |
<!-- reasonix-top-contributors:end -->
特别感谢 [**Bernardxu123**](https://github.com/Bernardxu123) 设计的项目 logo和开场视频。
@@ -310,10 +210,9 @@ CLI 进阶用法和详细配置见 **[CLI 命令参考](./docs/CLI.zh-CN.md)**
如果 Reasonix 帮你省了时间或 token,欢迎请杯咖啡。捐助不会换来 feature
优先级,也不会影响 issue 的处理顺序——就是「谢谢」。
- **国内** — 微信支付 / 支付宝(扫下方二维码)
- **国内** — 微信支付(扫下方二维码)
- **海外** — PayPal: [paypal.me/yuhuahui](https://paypal.me/yuhuahui)
<p align="center">
<img src=".github/sponsor/wechat-pay.jpg" alt="微信支付收款码" width="180"/>
<img src=".github/sponsor/alipay.jpg" alt="支付宝收款码" width="180"/>
</p>
+42 -389
View File
@@ -9,25 +9,22 @@ agent. It is the Reasonix analog of Claude Code's CLAUDE.md.
- Go kernel under `internal/`; each package owns one concern. A package's long
explanation belongs in its `doc.go`, not spread across implementation files.
- One transport-agnostic `control.Controller` sits behind every frontend (chat
TUI, HTTP/SSE serve, Electron desktop). Add behavior to the controller, not a
TUI, HTTP/SSE serve, Wails desktop). Add behavior to the controller, not a
frontend, so all three inherit it.
- Layering (enforced): utility packages import nothing under `reasonix/` except
one another — that layer is closed, so a leaf reaching a leaf drags no graph
along, while refusing it leaves one judgement copied into several of them; only
the frontends `cli`, `serve`, `acp`, `boot` and the hosts `cmd/`, `desktop/`
may import `control`; nothing below a frontend may import one. The declared
sets live in `tools/repolint/layers.go`.
- Layering (enforced): utility packages import nothing under `reasonix/`; only
the frontends `cli`, `serve`, `acp`, `bot`, `botruntime`, `boot` and the hosts
`cmd/`, `desktop/` may import `control`; nothing below a frontend may import
one. The declared sets live in `tools/repolint/layers.go`.
- Subagent delegation keeps five concepts apart: a profile says how a worker
thinks, `TaskSpec` what this call wants, `CapabilityGrant` what it may touch,
`ContextRequest` what it starts from, `SchedulerPolicy` when it runs. Put a
field in whichever member decides its value — profiles carry ceilings, never
per-call values. `internal/runtime/delegation/profile_boundary_test.go` enforces it.
- Cache-first: the system-prompt prefix (base prompt, tools, declared prefix
configuration) must stay byte-stable across turns so DeepSeek's automatic
prefix cache stays warm. Never mutate it mid-session — ride the turn tail
instead (see `control.Compose`), under the rules in *Context projection*.
per-call values. `internal/agent/profile_boundary_test.go` enforces it.
- Cache-first: the system-prompt prefix (base prompt + tools + memory) must stay
byte-stable across turns so DeepSeek's automatic prefix cache stays warm. Never
mutate it mid-session — ride the turn tail instead (see `control.Compose`).
- Performance features land with an effect test at their final boundary
(`internal/assembly/boot/effect_test.go` pattern): assert what actually reaches the
(`internal/boot/effect_test.go` pattern): assert what actually reaches the
provider request, frontend sink, or trajectory through the real `boot.Build`
assembly. Component correctness is not system effectiveness.
- A mutex- or atomic-guarded struct is ratcheted on its **scalar** field count
@@ -36,100 +33,6 @@ agent. It is the Reasonix analog of Claude Code's CLAUDE.md.
the move this blocks — group by lifetime into a named sub-state instead
(`agent.perTurnState` is the pattern), which costs one field and removes the
whole product.
- The context window (projection, compaction, estimation) owns `windowState`;
the loop reaches it via `a.window()`, it reaches the loop via `windowHost`.
`window-state` fails any other read of `sess.win`.
- Judgements read structure, never wording: shell ASTs (`shellparse`), types,
contracts, tool schemas. Phrase tables and message sniffing have been retired
in batches (task-policy prose, the planner's approval phrases, the executor
handoff tables) because each one misfires on real input, and a misfiring
judgement inside a gate is worse than no gate. Security allow-lists are the
exception and stay. Something that can only be built by matching words does
not get built — say what is missing instead of guessing at it.
- The other half of that rule: an error a caller must tell apart **carries an
identity** — a sentinel, a typed code — never a sentence. Only the producer
knows a deadline expired rather than a socket closed, and a message is where
that knowledge goes to die: the reader is left matching words, or with
nothing. Across HTTP the identity is a dotted code through `refuse`;
`refusal-path` fails a new `http.Error` in `internal/frontend/serve`, and the parity
test fails a code the frontend cannot say. Inside Go it is `errors.Is` on a
sentinel the wrapping preserves; `error-text` fails a new match against an
error message, following it one hop into a local because storing the text
first is what the direct form turns into.
- **Whoever first knows why an operation failed owns its class.** A transport
status is not a domain identity, and a display string is not a classification;
past that boundary a layer may project the class, never re-derive it. Both
directions were live: `catalog.go` answered 400 for a name nobody declared and
for a disk that would not write, so a failed store told the user their request
was wrong; `ServerRow` decided a connection was an authorisation problem by
matching `/401|403|unauthorized/` against prose an external MCP server wrote,
which the host had sanitised and truncated on the way through. Domain,
internal and dependency are three answers with three different things to do
next, and a caller that cannot tell them apart reports the last two as the
first.
- Failure attribution is host-owned state. When the host can determine why an
operation failed or stopped — truncation, timeout, unavailable context,
dependency skipping, policy refusal, output spill, argument parsing — that
cause goes to the model. Surfacing only the downstream symptom leaves it to
invent one, and the invented rule outlives the incident: a response cut at the
output limit reaches the model as "invalid arguments", and the session learns
to stop batching calls that were never the problem.
- A host-owned identity has to cross the model-visible boundary. Stable step
ids, typed error codes, boundary causes — anything the host expects the model
to cite must appear wherever that state is rendered for it. Telling the model
in a schema to prefer X while every concrete rendering shows only Y trains it
to use Y; `complete_step` said `step_id` PREFERRED while all four todo
renderers printed bare ordinals. Canonical dynamic state still stays out of
the cache-stable prefix — project the identities needed to act on it at the
turn tail.
- A metric is its definition, not its name. Before a number becomes a finding,
state what it counts, from which anchor, over which population, and what it
excludes. Three readings in one study died in that gap: a command of unknown
extent read as a proven write, the latest report's adjudication read as every
report's, and a syntactically accepted call read as a closed one. Each time a
convenient proxy had been left standing in for the state it was named after,
and each time the mechanism story built on it was wrong. A number without
those four lines is an observation, not a result.
- A constructed state is not a reachable one. Before a destructive test or a
counterfactual is reported as a defect, say who in production produces that
state, through which transitions, across which persistence boundary, under
whose authority, and whether a real run has been seen reaching it. A test that
assigns a private field proves the mechanism *given* the state and must be
reported as exactly that: `a.projectChecks` is written once at boot and never
again, so reassigning it stood in for a resume rather than demonstrating one,
and the bypass built on it was never shown to be production-reachable.
Severity is reachability times frequency times authority, not "can be
bypassed".
- A wire type the desktop re-declares by hand is held to the kernel's by
`wire-parity`, both directions: a field one side sends and the other cannot
read fails, and so does one it reads that nothing sends. The two are compiled
by different toolchains and meet only over JSON, so nothing else in the tree
notices when one of them grows a field. The pairs are declared in
`tools/repolint/wireparity.go`, never inferred — no spelling tells a mirrored
contract from a struct that merely carries json tags.
- "Add behavior to the controller so every frontend inherits it" is held by
`frontend-parity`: a capability one frontend already drives, a frontend whose
declared port carries it, and no call, is a missing edge, and the ratchet
counts edges. Which port each frontend drives is declared in
`tools/repolint/frontendparity.go`; a deliberate exception goes in the scope
table there with an issue and a reason, and costs what the edge cost, so the
table cannot become where debt disappears. The matrix is read from
`go/types` at one pinned build target, never from selector names — `Close`
and `Status` are capabilities and also stdlib methods, and a name match
records `os.File.Close` as a frontend driving `Lifecycle.Close`, a false
"wired" that hides real debt. How a verb is spelled — `/branch`,
`POST /branches` — is presentation and stays out of it.
- The desktop's presentation layer answers to canonical state and never
redefines it. Six rules, each held by a gate rather than by memory:
presentation reorders and reweights what the kernel owns but never restates
it; **mount is not a motion cause** — a fact enters once, and replay or
virtualization bringing its element back is not that fact happening again;
a search alias buys discoverability and never becomes a thing's identity;
when a setting takes effect is declared per block, from what its endpoint
does, never inferred from wording or handler names; ordering panels by
lifecycle moves them and must not remount them; and a control's identity
survives a change of where it is drawn — the settings sheet and the composer
render `chrome.preset` because that is one intent, not two.
## Comments
@@ -137,75 +40,17 @@ Default is none — the code is the truth. Write one only when the **why** is
non-obvious: a hidden constraint, a workaround anchored to something verifiable,
an invariant the type system cannot express, or an external-protocol quirk.
A comment states the constraint that holds now. It does not narrate how the code
got here: a reader who has never seen an earlier version must lose nothing, and
past tense about this code's own behaviour is the tell — an external protocol's
history is not. "Ids alone do not answer it" is the rule; "ids alone were the
wrong question, so the list went on reading as the model wrote it" is the
incident behind the rule, and the incident belongs in the commit message. This
outranks the surrounding style: much of the tree reads as prose and none of it
is the standard. No gate reads register — telling narrative from a statement of
constraint needs a phrase table, which is the one thing this repo does not
build — so this rule is held by review.
- **Stylesheets carry none.** A rule's selector and its declarations are the
whole statement. What a `.css` comment would have said goes to whichever of
two places can act on it: a constraint a component has to honour belongs in
that component, and a decision about the medium itself — a measured colour, a
layout mechanic, why the obvious rule is the wrong one — belongs in the
stylesheet's companion document (`src/styles/TOKENS.md`, `LAYOUT.md`), which
is read end to end rather than met 600 times as an aside.
- **English, always** — every comment, in every other language the tree is
written in. A codebase whose commentary is half one language reads as two
codebases, and the half a given reader cannot read is the half that rots.
- Declaration doc: ≤5 lines. Package comment: ≤8 lines, or ≤40 in a `doc.go`.
- Every other comment: ≤3 lines. Struct-field and trailing `//`: 1 line.
- Never: restatements of the code, phase/stage narrative, incident or
conversation history, section banners, commented-out code, `@param` lists.
- `TODO(#nnn):` and `HACK(#nnn):` need the issue anchor. `FIXME` is banned.
- One responsibility per file; 800 lines is the ceiling. Test files are exempt:
their length tracks how many cases they cover, not how many concerns they
carry, so splitting one only scatters a subject's table across files.
- One responsibility per file; 800 lines is the ceiling.
`go run ./tools/repolint` enforces the budgets above against a ratchet baseline:
recorded debt is tolerated, anything new fails CI. Never widen the baseline to
land a change — fix the code. `-update` lowers budgets freely and refuses to
raise one without `-allow-widen`, so carrying debt through a rename or an
extraction is asked for in the command and justified in the PR. A clean run also
reports budget the tree stopped using: it is only reclaimed by an `-update`, and
until then it is room a file can grow back into.
## Context projection
Canonical dynamic state — the skill registry, the memory store — lives in its own
owner, and what the model sees is a projection of it onto each request. The
prefix was never where these belonged wrongly; freshness ownership was missing.
- **Prefix ownership.** Only declared prefix configuration may move the
cache-stable prefix; ordinary memory writes, skill writes and runtime facts
ride the turn. Pinned memory bodies are the one declared exception, because
pinning is a request to pay one cold start.
- **Projection freshness.** When canonical state changes, the next eligible turn
carries the new projection or can say why it is not owed. "Right after a fold,
or in the next session" is not an answer.
- **Projection ownership.** Whoever owns a dynamic fact owns the invalidation
that makes its projection owed again. A projection kept fresh because every
writer remembered to announce itself is one forgotten call from stale —
`skillSet.owedCatalog` asks the registry rather than waiting to be told.
- **Determinism.** The same canonical state and the same projection state
produce byte-identical model-visible bytes.
- **Detection is semantic.** Freshness may not rest on writer notification, on
mtime, or on which call path made the change. The registry answers from
canonical state, so a switch flipped and a file edited past every API are the
same question, and a change no projection renders is not a change.
`internal/assembly/boot/context_projection_test.go` holds all five at the provider
boundary, with the skills listing and the standing instructions as the two
worked examples. A block is settled only once the turn carrying it reaches the
runner: a turn blocked by a hook or refused by an extension after composing put
those bytes nowhere, and clearing the debt there would leave the model holding
state nobody sent it. A projection that has become empty is still owed once —
otherwise the block the model already has goes on reading as current.
`go run ./tools/repolint` enforces all of it against a ratchet baseline: recorded
debt is tolerated, anything new fails CI. Never widen the baseline to land a
change — fix the code. `-update` exists for carrying debt through a rename or an
extraction, and that diff must be justified in the PR.
## Memory
@@ -219,58 +64,12 @@ otherwise the block the model already has goes on reading as current.
- `#<note>` in chat quick-adds an always-on instruction. The `remember` tool
instead saves a fallible background fact (frontmatter file + `MEMORY.md`
index). Fact `type` classifies content; independent `scope` controls whether it
is project-only (the default) or explicitly global.
- The saved-fact index stays out of the stable prefix and is reached through the
`memory` tool; pinned bodies and global user/feedback bodies load into it, the
latter as lower-priority compatibility guidance. The current turn receives a
tail note.
is project-only (the default) or explicitly global. The index loads into the
stable prefix on the next session; global user/feedback bodies also load as
lower-priority compatibility guidance. The current turn receives a tail note.
## Notes
One tool call costs one model round trip, so calls that do not depend on each
other belong in the same round: several reads, edits to different files, a check
after the edit that enables it. Several edits to the *same* file are one
`multi_edit` — it applies them against each other in memory and rewrites the
file only if all of them land, so a failure midway leaves nothing half-edited.
## Vendor rates
Official list prices live in one place: `officialRates` in
`internal/contract/pricing/rates.go`, oldest generation first, the last being what a
model charges today. A price change is an **append, never an edit** — the
superseded rate is what lets an installed config be recognised as holding one
of ours and brought forward, instead of quoting a retired rate for the rest of
its life. `config` reads these through `pricing.CurrentRate` and
`SupersededRates` and restates none of them; a table written twice is one that
drifts, which is how a vendor's USD rate ended up published on one side only.
`make pricecheck` reads each vendor's own page and compares it against that
table, and reports how long ago a person last confirmed each one. It reaches
the network, so it stays out of `make lint` and out of CI: run it before
touching a rate and when a vendor announces a change. A page it cannot read
exits non-zero — unread is its own verdict and never counts as agreement.
## Curated presets
A preset changing shape reaches the installs that already have the old one
through `shippedPresets` in `internal/contract/config/preset_upgrade.go`: one declared
shape per generation, and one upgrade that brings a matching entry forward. The
guard is the whole shape — catalog, window, default — because an entry holding
what we shipped is ours to move and anything else its user curated.
A preset is also what settles which of its models read images. Probing an
address a preset covers answers from that declaration, because reading a name
misses every Kimi, Qwen, MiniMax and Claude model they declare and `deepseek-
flash` spells nothing at all; the spelling is the last resort for an address
nothing has declared, and stays a correctable suggestion rather than a claim.
Three distinctions are load-bearing, and each was a separate migration function
before they were written down. A window the shape did not name is the user's,
so only a shape identified by its window moves one. `vision_models` **narrows**,
so an upgrade marks only the models it adds and only where the preset says they
read images. And an explicitly empty list is a choice — nothing here reads
images — which is not the same as the nil that means nobody has said.
## Pre-push CI simulation
Run these **before every commit** to catch the fastest CI failures locally:
@@ -279,7 +78,7 @@ Run these **before every commit** to catch the fastest CI failures locally:
gofmt -w . # catches gofmt (saves ~13s CI)
go vet ./... # catches vet warnings (saves ~52s CI/lint)
make lint # golangci-lint at CI's pin + repolint
go test ./internal/tools/builtin/ ./internal/assembly/boot/ # catches tool/boot test breaks
go test ./internal/tool/builtin/ ./internal/boot/ # catches tool/boot test breaks
```
`make lint` runs both gates CI runs, at the version in `.golangci-version`;
@@ -287,25 +86,6 @@ go test ./internal/tools/builtin/ ./internal/assembly/boot/ # catches tool/boot
shows up in `go vet`, and the CI round trip that catches it instead costs ten
minutes.
A full repolint run reports every file over budget, including debt your change
never touched. To see only what your own edits owe:
```bash
make check
```
Repo-wide ceilings still report under a narrowed run, because a file can push
the tree past one without exceeding its own budget.
`make check` owns the evidence-bearing changed-file universe — everything the
worktree altered *and* everything it added that git has not been told about.
Never substitute a direct `go run ./tools/repolint` invocation for it when
recording verification evidence: the host recognizes a `make` target as a check
whose result it can read, while an arbitrary `go run ./tools/...` is a program
it has to assume writes, and the flags behind the target are its implementation
rather than the contract. Building that universe from `git diff` alone once let
a new file reach the gate unopened and be reported clean.
## Import cycle rule
Before importing a new internal package from a non-test file, verify the target package's **test files** aren't already importing back to you:
@@ -322,159 +102,32 @@ Use `go test ./path/to/target/` to detect cycles **before** pushing. A `[setup f
- **Keep the PR diff minimal.** Only the files relevant to the PR's purpose — no stray changes from other branches.
- **Amend, don't add commits, for review feedback** — keeps the commit history clean.
## Change discipline
## PR metadata gates
**Assess before editing.** For every change, humans and agents list callers,
consumers of the same state, sibling features, reload/replay/restart paths,
error and edge states, TUI/CLI and desktop equivalents, and wire mirrors.
Two CI guards read the PR body. The scripts are the source of truth and both
run locally: `scripts/check-cache-impact.sh`, `scripts/check-docs-impact.sh`.
Separators must be an ASCII `-` or `:` — an em dash fails the docs guard.
A fix that breaks a neighbouring behaviour is a failed fix.
Cache-sensitive diffs (`internal/tool/`, `internal/provider/`,
`internal/boot/`, `internal/agent/agent.go`, and the rest of the list in the
script) require:
**Test the neighbourhood.** UI changes add or extend interaction tests for
adjacent functions of the same component or flow (composer: submit, clear,
queue, withdraw, restore, steer, failure, replay after reload) and run the whole
affected suite.
```
Cache-impact: <none|low|medium|high> - <reason>
Cache-guard: <focused guard test/command or existing guard rationale>
```
Write tests first; they must fail without the change.
`none` is a legitimate impact when the provider-visible prefix stays
byte-identical; only an empty value, `todo`, or `tbd` is rejected. If the diff
also touches `internal/config/`, `internal/memory/`, `internal/outputstyle/`,
`internal/skill/`, or `internal/boot/`, add `System-prompt-review: <note>` —
that field additionally rejects `none` and `n/a`, so it must name a reviewer.
**Fix at the owning layer.** Use the existing unified pattern: no special-case
flags, duplicated paths, or workarounds at the symptom site.
User-visible diffs (`cmd/reasonix/`, `desktop/`, `npm/`, and most of
`internal/`; tests and lockfiles are exempt) require one of these, chosen by
whether the same PR edited `docs/*.md`:
If the design is duplicated or wrong, make the smallest unifying change:
one path instead of two, deleting what becomes dead. Use current language and
framework idioms; do not add a new pattern where one exists.
**PR evidence.** Every PR description carries: Cause, Blast radius,
Neighbouring behaviours tested, and Why this layer.
## Reasonix host checks
The paths below decide what the agent is allowed to do — a wrong change here
is not a bug in a feature, it is a hole in a boundary. Declaring them makes
every change under them demand `review` plus `security_review` before a turn
can finish. Sensitivity is declared, never inferred: the host does not read it
out of how a path is spelled, which cannot tell `internal/auth` from
`session_write_authority.go`, or a trace file from a data race.
- sensitive: internal/safety/permission/**
- sensitive: internal/safety/sandbox/**
- sensitive: internal/safety/shellsafe/**
- sensitive: internal/base/shellparse/**
- sensitive: internal/session/control/approval.go
- sensitive: internal/session/control/approval_bridge.go
- sensitive: internal/session/control/approval_orchestration.go
- sensitive: internal/model/providerbroker/**
- sensitive: internal/ext/installsource/**
- sensitive: internal/ext/plugin/**
- sensitive: internal/ext/pluginpkg/**
- sensitive: internal/base/netclient/**
- sensitive: internal/safety/redirectguard/**
- sensitive: internal/safety/endpointclient/**
- sensitive: internal/safety/egress/**
- sensitive: internal/platform/browser/**
- sensitive: internal/platform/computer/**
- sensitive: internal/platform/gitpush/**
- sensitive: internal/session/control/git_push.go
- sensitive: internal/frontend/serve/git_push.go
- sensitive: internal/frontend/serve/host_routes.go
One declaration, two effects: the same list is also the coverage gate's subject.
`make coverage-gate` holds each path to the coverage it already has, because
when this was written these paths sat *below* the tree's median rather than
above it — the code deciding what the agent may execute was its least-tested
part. Floors ratchet like repolint's budgets: raising one is free, lowering one
needs `-allow-drop` and a reason in the PR. Adding a `sensitive:` path gets both
effects; there is no second list to keep in sync.
## Trust domain
Three sets, kept apart because conflating them turns a performance number into a
security claim: **VerificationRelevant** (state that can change whether a
delivery claim holds), **Observable** (state the host can reconcile after an
action, so it can say what that action did), **HostProtected** (state a
capability/sandbox boundary makes unwritable). The invariant is
`Writable ∩ VerificationRelevant ⊆ Observable ∪ HostProtected`:
> Any state that can both be modified by an action and influence the delivery
> claim must either be observable by the host or protected from modification by
> host-enforced boundaries. An incomplete observation never establishes scope.
Effects and state answer different questions and never stand in for each other —
observation proves what an action did, verification proves what the resulting
state satisfies — which is why `unproven_mutation` is settled only by
observation, never by a check that ran afterwards.
What follows bounds Observable today. It is not a restatement of the invariant:
moving a number here must not move the promise.
- `scanWorkspace` walks the workspace and every granted extra root recursively,
skips VCS store dirs, and is `complete` only below 50k files in total. Over that, or past a walk error, it
establishes nothing and the mutation stays unproven.
- Symlinks are recorded by `Lstat`, so a write through one lands outside what
the walk compares. Confining the resolved target is the sandbox's job.
- `internal/safety/sandbox` confines writes to the workspace, configured extras, temp
and toolchain caches on macOS and Linux.
- On Windows a user-enabled write-restricted token bounds the *creation* of files
(create, append, overwrite, `mkdir`, HKCU writes, global git config, writes through links).
- There, deleting or renaming away an existing file is not bounded, and with the
setting off nothing the host enforces bounds `Writable`.
- The caches a command may write can be poisoned for the user's later unsandboxed
builds, on every platform.
- The governed host authorities are a *named* set — ssh-agent, docker, podman.
DBus, Wayland, X11, gpg-agent, keyrings and arbitrary sockets stay reachable,
so the claim is "these endpoints are governed", never "host IPC is isolated".
`stateEpoch` tracks host-observed mutations, not an exact filesystem snapshot;
unobserved external writers are a gap shared by the whole verification model. Go
baseline test criteria use the toolchain's build overlay and therefore share
ordinary verification's epoch semantics; the overlay is not a general filesystem
view.
A captured criterion may be superseded only by authority outside
execution-mutable state; the current workspace, its tests, and the agent itself
cannot authorize supersession.
Two gaps follow and are open, not decided: a build that reads `.git` makes VCS
state VerificationRelevant while the scan skips it, and on Windows a symlink out
of the workspace is neither observed nor protected. Closing either means
widening observation or narrowing writability — never widening what a check is
taken to prove.
## Sandbox dimensions
`HostProtected` above is the delivery-level abstraction. Sandbox policy refines
it into three independent protection dimensions. Their separation is empirical:
measured bypasses showed that protection on one dimension cannot be assumed to
imply protection on another.
- **Integrity** — what host state may be mutated. Expressed by write roots.
- **Confidentiality** — what host information may be observed. Expressed by
`ForbidReadRoots`.
- **Authority** — what host capabilities may be exercised *without* possessing
their secrets. Expressed by `HostAuthorities`.
`ReadOnly` is therefore an Integrity property only; it makes no Confidentiality
or Authority claim.
Network transport and host authority are independent even when a backend
primitive conflates them: on macOS, `(deny network*)` also cut ssh-agent, so one
flag could not express "no internet, still sign".
Filesystem visibility is not authority: denying `file-read*` on a socket did not
stop a connect. Possession is not authority: the key can be unreadable while an
agent still signs with it.
A protection claim exists only where a backend has a demonstrated enforcement
primitive for that dimension. Unsupported dimensions stay explicitly unsupported,
because a claim that outruns its enforcement is worse than an absent one: it is
believed.
Windows enforces Integrity on file creation only, when the user turns it on.
Deletion and rename, Confidentiality and Authority (the named pipes of ssh-agent
and docker, COM and RPC services) are not enforced there.
**Known unmeasured surface — macOS host-mediated delegation:** Apple Events and
Mach-service lookup are not denied by the current allow-default profile. Prior
`osascript` and `launchctl submit` probes were inconclusive; no delegation
effect has been demonstrated. This surface is outside the currently governed
host-authority endpoint set.
```
Documentation-impact: updated - <what changed> # docs/*.md edited
Documentation-impact: none - <why the docs stay correct> # not edited
```
+6 -57
View File
@@ -56,10 +56,9 @@ Supported boundaries include:
- Sandbox behavior for built-in shell execution where the platform supports it.
- Secret handling for provider keys, bot credentials, OAuth tokens, plugin
headers, and credential-store fallback files.
- HTTP `serve` protections, including the default per-launch token, the launch
token every state-changing request requires when authentication is opted
out, localhost binding assumptions, JSON-only state-changing requests, and
CORS restrictions.
- HTTP `serve` protections for the unauthenticated local server, including
localhost binding assumptions, JSON-only state-changing requests, and CORS
restrictions.
- Desktop and bot session isolation, including per-workspace session metadata
and configured bot allowlists.
- Updater, install, and release verification paths.
@@ -68,10 +67,10 @@ The following are normally treated as trusted local/operator-controlled inputs
unless another bug lets an untrusted actor supply them:
- CLI arguments and text typed directly by the local user.
- The user's own configuration under the Reasonix home directory.
- Project configuration files intentionally loaded from the current workspace.
- Explicit `@path` references supplied by the local user to attach local files.
- MCP servers, language servers, hooks, and slash commands installed, enabled,
or approved by the local user.
- MCP servers, language servers, hooks, and slash commands installed or enabled
by the local user.
- Provider base URLs and model names configured by the local user.
The following can be security issues when reachable by an untrusted actor or
@@ -90,56 +89,6 @@ when they bypass the intended boundary:
- Trusting unverified update artifacts, plugin definitions, or downloaded
binaries.
## Project Configuration
Files that arrive with a workspace are untrusted input: `reasonix.toml`,
`.reasonix/settings.json`, `.mcp.json` and the project `.env`. Cloning a
repository is not a decision to trust its author, so these files may narrow
what the user configured and never widen it.
| Setting | What a workspace file may do |
| --- | --- |
| `sandbox.bash` | Move `off` to `enforce`, never back |
| `sandbox.network` | Move `true` to `false`, never back |
| `sandbox.forbid_read`, `sandbox.denied_domains` | Add to the user's lists |
| `sandbox.allow_write`, `sandbox.workspace_root` | Apply only when they resolve inside the workspace, symlinks followed; `${VAR}` in any sandbox path expands from the process environment only |
| `sandbox.host_authorities`, a user's `sandbox.allowed_domains` | Nothing |
| `permissions.deny`, `permissions.ask` | Add to the user's rules |
| `permissions.mode`, `permissions.allow`, `permissions.allow_dynamic_bash` | Nothing |
| `desktop.default_tool_approval_mode` | Nothing |
| `[network]` proxy settings | Nothing |
| `[[providers]]`, `tools.system_one` endpoints | Declare them; they stay out of use until approved, and model settings naming them fall back to the user's |
An "always allow" answer, or extra write access, granted in a workspace is
stored under the Reasonix home, not in the workspace.
Programs a workspace names for the host to run stay off until the user
approves them for that workspace:
- Hooks, language server commands, `tools.search.rg_path`,
`tools.shell.path`, `[browser]` and `tools.system_one.laya.python` are
approved with `reasonix trust`, or for hooks by saving them from the hooks
editor.
- An approval is stored under the Reasonix home with a digest of the
declaration and of the workspace files it names; any change needs approval
again. Approvals and grants belong to one workspace folder.
- Hooks and language servers are checked against their approval again before
each start, and refused if a file they name changed, appeared or now
resolves elsewhere.
- A single program path (`tools.shell.path`, `tools.search.rg_path`,
`[browser]`, `laya.python`) may not point into the workspace, an
`allow_write` root, a temporary directory or a toolchain cache the bash jail
leaves writable, and may not climb with `..`; its approval covers the
program file's content.
- `tools.system_one.laya.local` is read from the user's configuration only.
- Without an interactive frontend they stay off.
- MCP servers a workspace declares stay off until the user enables them for
that workspace.
A project value that is not applied is reported as a load warning with the
reason. A way for workspace files to widen any of the above without the
user's approval is a security issue.
## `@` File References
Reasonix supports `@path` references so users can include local files and images
-22
View File
@@ -1,22 +0,0 @@
# Build output and local tooling state.
build/
.gradle/
.kotlin/
local.properties
*.iml
.idea/
# The signing key and the file that names it.
#
# A key is not part of the project: it is the identity one person ships under,
# and publishing it would let anyone else sign an APK that installs over
# everyone's. Neither the keystore nor its password belongs in version control.
# `app/build.gradle` reads both from here when present and falls back to the
# debug key when they are not, so a fresh clone builds without them.
keystore/
# Build inputs, produced by scripts/ before Gradle runs.
# jniLibs/ holds the cross-compiled kernel, assets/page/ the built
# frontend. assets/LICENSE is hand-placed and stays tracked.
app/src/main/jniLibs/
app/src/main/assets/page/
-216
View File
@@ -1,216 +0,0 @@
---
owner: @esengine
backup: @SivanCola
status: active
reviewed: 2026-10-20
---
# Reasonix Mobile — the Android shell
Runs Reasonix as an Android app: the kernel ships inside the APK and starts on the
device, the interface is the same `desktop/frontend-next` build the desktop Studio
serves, and the agent reaches servers over SSH.
This file is for people building it. What a person installing it needs — what it
does, how to pair with a computer, what it cannot do — belongs in the release
notes.
## One command
```bash
./scripts/build.sh release # or debug
```
Three steps, in this order:
1. `build-kernel.sh` → `app/src/main/jniLibs/arm64-v8a/libreasonix.so`
2. `sync-page.sh` → `app/src/main/assets/page/`
3. `./gradlew assembleRelease`
Output: `app/build/outputs/apk/release/app-release.apk`
The kernel and the page are **inputs to packaging, not Gradle build steps**.
Running `gradle assembleRelease` on its own fails on purpose and names the script
that produces whichever input is missing (`requireBuildInputs` in
`app/build.gradle`): an APK with no kernel in it should not be buildable by
accident.
## Requirements
| Component | Version |
| --- | --- |
| JDK | 21 (17 also works) |
| Android SDK | `platforms;android-35`, `build-tools;35.0.0`, `platform-tools` |
| Gradle | the committed wrapper, 8.14.3 (for AGP 8.11.1) |
| Go | 1.26+ |
| Node / pnpm | Node 24+ / pnpm 10 |
```bash
export ANDROID_HOME="$HOME/Android/Sdk"
export ANDROID_SDK_ROOT="$ANDROID_HOME"
export JAVA_HOME="/path/to/jdk-21"
export PATH="$PATH:$(go env GOPATH)/bin"
```
`ANDROID_HOME` is the one thing nothing sets for you.
## Scripts
| Script | What it does |
| --- | --- |
| `build.sh [debug\|release]` | The whole chain |
| `build-kernel.sh` | Cross-compiles the kernel only (`GOOS=android GOARCH=arm64 CGO_ENABLED=0`) |
| `sync-page.sh` | Builds the frontend, copies `dist/` into assets, copies `LICENSE` |
| `verify-page.sh` | **No phone needed**: starts the real kernel against the packaged page |
| `run.sh [debug\|release]` | `adb install`, launch, follow the log |
`verify-page.sh` is the one to run after a change: it starts the kernel with the
same arguments `Kernel.kt` uses (`--auth token`, `--token-file`, `--port-file`,
`-page`) and checks that it binds a port and answers 200. Without a device, that
is as deep as the chain can be verified.
## Where the code is
| File | Owns |
| --- | --- |
| `MainActivity.kt` | The WebView, the back gesture, downloads, external links, renderer restart |
| `Kernel.kt` | Unpacking the page, starting and stopping the kernel, waiting for the port, health probes, logs, the workspace directory |
| `KernelService.kt` | The foreground service that keeps a long turn alive |
| `NativeBridge.kt` | `@JavascriptInterface`: what the page may ask the phone for |
| `BrowserShellServer.kt` | Loopback HTTP server the kernel's browser tools drive the overlay WebView through |
| `Downloader.kt` | Writing a kernel-produced file through the system document picker |
The page reaches the shell through `window.reasonixHost` and `ReasonixNative`,
defined in `desktop/frontend-next/src/port/host.ts` (`AndroidHost`).
**Phone and desktop share one page build.** What differs only on a phone is
decided at runtime, not by a second set of components. `phoneShell()` in
`src/port/host.ts` and the `[data-shell="android"]` selector in `mobile.css` both
read `<html data-shell="android">`, which `index.html` sets before the bundle
loads.
One attribute feeding both is what keeps style and components from disagreeing
about what a phone is.
## What the page may ask the phone for
`NativeBridge` is reachable from whatever document the WebView holds, so every
method that touches the device asks `trusted()` first.
The activity answers by comparing the **main frame's** origin against the
kernel's and against the one paired computer the person scanned a code for. A
page that navigated elsewhere is not trusted, even though it may still hold the
object.
Five methods answer without the check — `version`, `systemTheme`, `safeArea`,
`setLayerOpen`, `setRunning` — because they report back what the page already
knows about its own window and grant nothing.
`openInApp(url)` is the pairing decision, not a navigation one: it loads a URL in
this window only for the origin the scanner just returned, and only once.
Without that, any page holding the kernel origin could hand itself the bridge.
External links leave through the browser. A non-web scheme is routed to the
system only for `mailto:` and `tel:`; everything else is dropped, because an
`intent://` URL can start an arbitrary exported activity with caller-supplied
extras.
## Permissions, and what they cost
`MANAGE_EXTERNAL_STORAGE` — opening a folder on the phone as a workspace. The
kernel is a Go binary: it opens real paths with `open(2)`, and SAF hands back a
`content://` URI it cannot open.
Requested from the folder-picker action, never at launch. Android shows no dialog
for it and opens system settings instead.
**There is no OS-level sandbox around this kernel.** The PR sets the sandbox to
off on Android, so confining the agent to the chosen folder is Reasonix's own
write-root check and nothing else. A shell tool with all-files access can reach
all shared storage.
`REQUEST_IGNORE_BATTERY_OPTIMIZATIONS` — Doze freezes the app group, which stops
a turn that is still running. Asked once when a turn actually runs, never at
launch, and the person can refuse. Refusing costs battery life during long turns
and nothing else.
**Google Play restricts this permission**: fine for a sideloaded build, a
store-policy risk if this is ever listed.
`FOREGROUND_SERVICE` / `FOREGROUND_SERVICE_DATA_SYNC` — keeping the kernel alive
while the screen is off. `dataSync` is capped per 24 hours from Android 15. It is
the closest honest description of "an agent is working".
`CAMERA` — reading the pairing code a computer shows. Requested when the person
asks to scan, never at launch. `ScanActivity` is not exported.
`READ_EXTERNAL_STORAGE` / `WRITE_EXTERNAL_STORAGE` — the same folder walk below
API 33. `maxSdkVersion`-gated; above 33 they are superseded by the permission
above.
`allowBackup=false` with explicit extraction rules, so a `keystore` or a token in
app storage never leaves the device in a cloud backup.
## Cleartext HTTP
The kernel serves its page on `127.0.0.1:<port>`, and a paired computer serves
the same page over its LAN address. Both are plain HTTP.
Neither address can be written into a `network-security-config`: the port is
chosen at start, the LAN address at scan time, and the format matches literals
and domains, not subnets.
So the config permits cleartext and the boundary is in the code instead.
`shouldOverrideUrlLoading` sends every other http(s) origin to the browser, and
the bridge answers only the kernel and the just-paired computer. A plaintext page
on the LAN cannot become the document holding the bridge.
## What goes into the APK
| Content | Size | Why it is shaped this way |
| --- | --- | --- |
| `lib/arm64-v8a/libreasonix.so` | 45 MB | Go static binary, stripped. It is a `.so` rather than a file in `assets/` because only `nativeLibraryDir` is guaranteed executable — a binary copied to app storage fails on a device whose data partition is mounted `noexec`. |
| `assets/page/` | 14 MB | The frontend build, including the Inter and Noto Sans SC fonts, `noCompress` |
| `assets/LICENSE` | 1 KB | The repository's MIT licence, verbatim. MIT requires the notice and the licence text to accompany a binary distribution. |
`assets/LICENSE` is generated by `sync-page.sh` from the repository's root
`LICENSE`, not committed a second time: a checked-in copy drifts when the root
file is updated.
It sits in `assets/` rather than `assets/page/` because that directory is removed
and rebuilt on every build.
A release APK is about 31 MB.
## Building
**The Gradle wrapper is committed**, jar and all. `build.sh` refuses to run
without `gradlew` rather than generating one from whatever `gradle` is on `PATH`:
the dependency versions are part of the build.
`gradle-wrapper.properties` pins `distributionSha256Sum`, so a mirror that serves
a different distribution zip fails verification instead of being unpacked and
run. Dependency repositories are the official ones only — a mirror belongs in a
user-level init script.
**Signing.** A release with no key configured comes out **unsigned**
(`app-release-unsigned.apk`) with a warning. It used to fall back to the debug
key, which produced an artifact that looked shippable and was signed by a
per-machine key nobody chose: it could not install over the previous build.
Point the build at a real key by writing `keystore/keystore.properties`
(`storeFile`, `storePassword`, `keyAlias`, and optionally `keyPassword`) or by
setting the matching `REASONIX_KEYSTORE*` environment variables. Key material
stays out of the repository and out of build logs.
Debug builds always use the debug key. Signing a debug build with the release key
would let anyone who can read the keystore produce a build that installs over the
shipped one under the same signature.
**Non-ASCII checkout paths.** AGP refuses to build under a path with non-ASCII
characters on Windows unless `android.overridePathCheck=true` is set. That is a
property of one machine's disk layout, not of this project, so it is not in the
repository's `gradle.properties`.
Set it in a user-level `~/.gradle/gradle.properties` or pass
`-Pandroid.overridePathCheck=true`.
-263
View File
@@ -1,263 +0,0 @@
// The app module.
//
// Two things ship inside the APK and neither is fetched at runtime: the kernel
// binary, and the built page it serves. Bundling the page is what makes the app
// work offline and start instantly; bundling the kernel is what lets it run
// `reasonix serve` on the phone itself, which is the whole point — the agent
// starts locally and reaches the server over SSH, so the interface, the model
// catalogue and the credentials all live on the phone.
//
// The kernel is placed as `jniLibs/arm64-v8a/libreasonix.so`, not under
// `assets/`. That is not a trick for its own sake: Android permits executing a
// file under `nativeLibraryDir` and nothing under `filesDir` has that
// guarantee, so a binary shipped the ordinary way would have to be copied out
// and made executable at runtime — which fails outright on a device mounted
// noexec, and fails confusingly on the ones that are merely unusual. `scripts/
// build-kernel.sh` writes it there; the name is fixed by the packaging rules.
plugins {
id "com.android.application"
id "org.jetbrains.kotlin.android"
}
// One version, read by the About screen and the update check.
//
// A published build takes its version from the tag it was cut for, so the About
// screen and the update check name that release rather than the number this file
// was last edited to. Without the property — a local build — the numbers below
// are the answer.
def publishedVersion = findProperty("reasonixVersion") as String
def versionMajor = 2
def versionMinor = 0
def versionPatch = 0
def versionBuild = 1
if (publishedVersion) {
def parsed = publishedVersion =~ /^(\d+)\.(\d+)\.(\d+)/
if (!parsed.find()) {
throw new GradleException("reasonixVersion must start with MAJOR.MINOR.PATCH, got: ${publishedVersion}")
}
versionMajor = parsed.group(1) as int
versionMinor = parsed.group(2) as int
versionPatch = parsed.group(3) as int
}
// Where the two build inputs are expected. Both are produced from the checkout
// beside this module, by the scripts named in each message.
def upstream = rootProject.file("..")
def pageDir = new File(projectDir, "src/main/assets/page")
def kernelSo = new File(projectDir, "src/main/jniLibs/arm64-v8a/libreasonix.so")
android {
namespace "io.reasonix.mobile.app"
compileSdk 35
defaultConfig {
applicationId "io.reasonix.mobile.app"
minSdk 26
// 35: the API level this app is actually built and tested against.
//
// It was briefly lowered to 26 while chasing an install failure that
// turned out to be a package-name collision with an unrelated app on the
// user's phone — the signature error named the wrong cause. Nothing about
// the target level ever caused it, so it goes back to the level the code
// is really written for.
//
// One consequence to keep in mind: from 35, Android 15 forces the window
// edge-to-edge and ignores a request to lay out inside the safe area. The
// window therefore *is* drawn under the status bar and the home indicator,
// and the only correct answer is for the page to pad itself — which is
// what `MainActivity` feeds real inset values for, since a WebView does
// not implement `env(safe-area-inset-*)`.
targetSdk 35
// versionCode must strictly increase, because Android refuses an update
// whose versionCode is not higher. The build digit is a single decimal
// place, so the scheme collides once a patch reaches 10: 2.55.10 and
// 2.56.0 both come out as 25601. Patch releases above 9 are not expected
// on Studio, and widening the digit now would renumber every build, so
// the constraint is recorded rather than fixed.
versionCode versionMajor * 10000 + versionMinor * 100 + versionPatch * 10 + versionBuild
versionName "${versionMajor}.${versionMinor}.${versionPatch}"
// The kernel is built for arm64 only, so every other ABI would ship an
// APK that installs and then cannot start. Naming the one real ABI keeps
// the store listing and the installer honest.
ndk {
abiFilters "arm64-v8a"
}
resourceConfigurations += ["zh", "en"]
}
// Android identifies an app by package name *and* signing key: an APK signed
// with a different key is not an update, it is an impostor, and the
// installer refuses it with "the signature does not match the installed
// app". Gradle's default release key is the debug key in the user's home
// directory, which is per-machine, so a build made on a second computer
// could never be installed over one made on the first.
//
// Whoever ships a build therefore pins a key — and that key is theirs, not
// the project's. It lives beside the checkout, in `keystore/`, which is
// ignored; so does the file naming it. A checkout without either signs with
// the debug key instead, which is what a contributor building locally wants
// and is what the README tells them to expect. Point the build at a real key
// by writing `keystore/keystore.properties` (storeFile, storePassword,
// keyAlias) or by setting REASONIX_KEYSTORE / REASONIX_KEYSTORE_PASSWORD /
// REASONIX_KEY_ALIAS.
def keyProperties = new Properties()
def keyPropertiesFile = rootProject.file("keystore/keystore.properties")
if (keyPropertiesFile.exists()) {
keyPropertiesFile.withInputStream { keyProperties.load(it) }
}
def keyStorePath = keyProperties.getProperty("storeFile") ?: System.getenv("REASONIX_KEYSTORE")
def keyStorePassword = keyProperties.getProperty("storePassword") ?: System.getenv("REASONIX_KEYSTORE_PASSWORD")
// The key's own password is a separate fact from the store's. PKCS#12 keystores
// almost always use one password for both, so falling back to the store's
// password keeps those working — but assuming they are equal is how a keystore
// that separates them gets a build failure at signing time, after the work is
// done, with an error that does not say which of the two was wrong.
def keyPasswordValue = keyProperties.getProperty("keyPassword") ?: System.getenv("REASONIX_KEY_PASSWORD") ?: keyStorePassword
def keyAliasName = keyProperties.getProperty("keyAlias") ?: System.getenv("REASONIX_KEY_ALIAS")
def haveReleaseKey = keyStorePath && keyStorePassword && keyAliasName &&
rootProject.file(keyStorePath).exists()
signingConfigs {
if (haveReleaseKey) {
releaseKey {
storeFile rootProject.file(keyStorePath)
storePassword keyStorePassword
keyAlias keyAliasName
keyPassword keyPasswordValue
// v1 is not optional here. Android 7+ reads v2 and ignores v1, so
// a v2-only APK installs on a modern phone — and then fails on the
// compatibility layers older devices and HarmonyOS' 卓易通 run,
// which know only the JAR signature and report its absence as a
// signature *mismatch*. Signing both is what makes one APK install
// everywhere.
enableV1Signing = true
enableV2Signing = true
enableV3Signing = true
}
}
}
// Nothing native is compiled here — the one .so is a prebuilt the kernel
// script writes — so the external native build is left alone entirely.
buildTypes {
debug {
applicationIdSuffix ".debug"
versionNameSuffix "-debug"
// The debug key, always. Signing a debug build with the release key
// would mean anyone who can read the release keystore can produce a
// build that installs over the shipped one under the same signature,
// and it buys nothing: the debug package is suffixed `.debug`, so the
// two already install side by side.
}
release {
// The kernel is a stripped Go binary and the page is minified, but
// the Kotlin shell still benefits and shipping the shrunk build is
// how a size regression gets noticed.
minifyEnabled true
shrinkResources true
proguardFiles getDefaultProguardFile("proguard-android-optimize.txt"), "proguard-rules.pro"
// A release with no key of its own comes out *unsigned*
// (`app-release-unsigned.apk`). It used to fall back to the debug key
// so the artifact would install, which produced a release that looked
// shippable and was signed by a per-machine key nobody chose: it could
// not be installed over the previous build, and every machine made a
// different one. An unsigned artifact says what it is.
if (haveReleaseKey) {
signingConfig signingConfigs.releaseKey
} else {
logger.warn(
"release build has no signing key: producing an unsigned APK. " +
"Write keystore/keystore.properties (storeFile, storePassword, keyAlias) " +
"or set REASONIX_KEYSTORE / REASONIX_KEYSTORE_PASSWORD / REASONIX_KEY_ALIAS."
)
}
}
}
compileOptions {
sourceCompatibility JavaVersion.VERSION_17
targetCompatibility JavaVersion.VERSION_17
}
kotlinOptions {
jvmTarget = "17"
}
buildFeatures {
// AGP 8 stops generating BuildConfig unless asked. `Kernel` reads the
// version name and code out of it to decide whether the bundled page
// needs unpacking, so it is asked for.
buildConfig = true
}
packaging {
jniLibs {
// `extractNativeLibs=true`: the kernel must exist as a real file on
// disk to be executed, and an uncompressed-and-mmap'd .so is not
// one. This is the only reason the flag is on.
useLegacyPackaging true
}
resources {
excludes += ["META-INF/*.version", "META-INF/*.kotlin_module", "kotlin/**", "DebugProbesKt.bin"]
}
}
// The page is read out of assets by `Kernel.publish`, not by the resource
// system, so it is not compressed — a page is already-compressed text and
// shrinking it again only costs startup time.
androidResources {
noCompress += ["json", "woff2", "ttf", "woff", "txt", "svg"]
}
lint {
abortOnError true
warningsAsErrors false
checkReleaseBuilds true
}
}
dependencies {
implementation "androidx.core:core-ktx:1.15.0"
implementation "androidx.appcompat:appcompat:1.7.0"
implementation "androidx.activity:activity-ktx:1.9.3"
implementation "androidx.webkit:webkit:1.12.1"
implementation "androidx.lifecycle:lifecycle-runtime-ktx:2.8.7"
implementation "org.jetbrains.kotlinx:kotlinx-coroutines-android:1.9.0"
// Scanning the pairing code a computer shows.
//
// CameraX for the preview and the frame stream, ZXing for the decode, and
// deliberately not ML Kit: this app installs on phones whose HarmonyOS
// compatibility layer has no Google Play Services, so a decoder that needs
// GMS would work on the developer's phone and fail on the user's. ZXing
// core is pure Java with no runtime dependency at all, which is what makes
// the scanner work on both.
implementation "androidx.camera:camera-core:1.4.1"
implementation "androidx.camera:camera-camera2:1.4.1"
implementation "androidx.camera:camera-lifecycle:1.4.1"
implementation "androidx.camera:camera-view:1.4.1"
implementation "com.google.zxing:core:3.5.3"
}
// Fail before the packaging step with something a person can act on, rather
// than shipping an APK whose kernel is missing or whose page is a directory the
// build never filled. The two messages name the script that produces each.
tasks.register("requireBuildInputs") {
doFirst {
if (!kernelSo.exists()) {
throw new GradleException(
"内核二进制缺失:${kernelSo}\n先运行 ./scripts/build-kernel.sh(在 android/ 目录下)交叉编译内核。")
}
if (!pageDir.exists() || !new File(pageDir, "index.html").exists()) {
throw new GradleException(
"前端页面缺失:${pageDir}/index.html\n先运行 ./scripts/sync-page.sh(在 android/ 目录下)构建并同步页面。")
}
}
}
tasks.configureEach { task ->
if (task.name.startsWith("merge") && (task.name.contains("Assets") || task.name.contains("JniLibFolders"))) {
task.dependsOn("requireBuildInputs")
}
}
-32
View File
@@ -1,32 +0,0 @@
# R8 rules for the shell.
#
# There is very little to keep here, and that is the point: the app's own code
# is reached from the manifest and from Android's own callbacks, both of which
# R8 already traces, and nothing in it is loaded reflectively. What follows is
# the two places where a name is looked up rather than called.
# The WebView's JavaScript bridge, if one is ever added, is resolved by name on
# the Java side and by the string the page uses. Keeping annotated members makes
# that survive shrinking rather than failing at runtime with a method-not-found
# that only shows up on a release build.
-keepclassmembers class * {
@android.webkit.JavascriptInterface <methods>;
}
# The Kernel's child process is started by absolute path, and its own log is
# read as text; neither is reflective, so nothing else needs keeping.
# Line numbers make a stack trace from a release build readable, which is the
# only way a crash report from a phone can be acted on.
-keepattributes SourceFile,LineNumberTable
-renamesourcefileattribute SourceFile
# Kotlin's `@Metadata` annotation is present on every class the compiler touched
# and nothing in this app reflects over it. It is tempting to `-assumenosideeffects`
# it away, and that was tried: any pattern broad enough to name its members also
# matches the methods every class inherits from `Object` — `equals`, `hashCode`,
# `toString` — and R8 then treats those as removable, which is a silent
# behaviour change for a few kilobytes. Left as a warning suppression instead:
# the annotation is dropped from the shrunk output anyway, because nothing reads
# it.
-dontwarn kotlin.Metadata
-112
View File
@@ -1,112 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!--
The shell's manifest.
There is no INTERNET-adjacent permission beyond INTERNET itself: the page and
the kernel talk over loopback, and reaching a server is the kernel's job over
a socket it owns. Reading and writing files goes through the system's document
picker (SAF), which hands back a grant for one file at a time — so no
storage permission is declared and none is needed.
`usesCleartextTraffic` is off here, and cleartext is re-enabled in the network
security config only because the kernel serves its page on `127.0.0.1` and a
paired computer serves the same page over its LAN address — both plain HTTP.
The config says what that costs and what bounds it.
-->
<manifest xmlns:android="http://schemas.android.com/apk/res/android"
xmlns:tools="http://schemas.android.com/tools">
<uses-permission android:name="android.permission.INTERNET" />
<uses-permission android:name="android.permission.ACCESS_NETWORK_STATE" />
<!-- Only to tell the person the kernel is still running while the screen is
off; it is a foreground service with a dismissible notification, not a
background wake-up. -->
<uses-permission android:name="android.permission.FOREGROUND_SERVICE" />
<!-- From API 34 every foreground-service *type* needs its own permission as
well as the general one: naming dataSync in the manifest without asking
for it is a SecurityException at startForeground, on the one code path
that keeps a long task alive. -->
<uses-permission android:name="android.permission.FOREGROUND_SERVICE_DATA_SYNC" />
<uses-permission android:name="android.permission.POST_NOTIFICATIONS" />
<!-- Only to read the pairing code a computer shows. Requested at the
moment the person asks to scan, never at launch. -->
<uses-permission android:name="android.permission.CAMERA" />
<!-- The kernel is a child process, and a task left running is one the person
expects to finish with the screen off. Android's own doze can freeze the
whole app group instead, which is what the shell asks to be exempt from
— once, when a turn is actually running, never at launch. It is a
request the person can refuse, and refusing it only costs battery life
during a long turn. -->
<uses-permission android:name="android.permission.REQUEST_IGNORE_BATTERY_OPTIMIZATIONS" />
<!-- Every-file access, for opening a folder on this phone as a workspace.
The kernel is a plain Go binary: it opens paths with `open(2)`, and a
document URI is not a path it can open. SAF would hand back exactly
that unusable URI, so the one honest permission is the broad one —
granted by the person in system settings, which is the only place
Android offers it. Nothing here reads storage for any other purpose. -->
<uses-permission
android:name="android.permission.MANAGE_EXTERNAL_STORAGE"
tools:ignore="ScopedStorage" />
<!-- Below API 33 the same folder walk needs the older pair; from 33 they are
superseded by the permission above and asking for them would be a lie. -->
<uses-permission android:name="android.permission.READ_EXTERNAL_STORAGE" android:maxSdkVersion="32" />
<uses-permission android:name="android.permission.WRITE_EXTERNAL_STORAGE" android:maxSdkVersion="29" />
<!-- A phone without a touchscreen cannot drive this interface; saying so
keeps the listing honest rather than letting it install on a TV. -->
<uses-feature android:name="android.hardware.touchscreen" android:required="true" />
<!-- The camera is optional: a phone without one is still a fine client for
a kernel it already knows, and it can still pair by opening the link the
computer shows. Requiring it would take the app off those devices. -->
<uses-feature android:name="android.hardware.camera" android:required="false" />
<uses-feature android:name="android.hardware.camera.autofocus" android:required="false" />
<application
android:name=".App"
android:allowBackup="false"
android:dataExtractionRules="@xml/data_extraction_rules"
android:fullBackupContent="false"
android:icon="@mipmap/ic_launcher"
android:label="@string/app_name"
android:networkSecurityConfig="@xml/network_security_config"
android:supportsRtl="true"
android:theme="@style/Theme.Reasonix"
android:usesCleartextTraffic="false"
tools:targetApi="35">
<activity
android:name=".MainActivity"
android:exported="true"
android:configChanges="orientation|screenSize|screenLayout|keyboardHidden|uiMode|density|fontScale|smallestScreenSize"
android:launchMode="singleTask"
android:windowSoftInputMode="adjustResize"
android:theme="@style/Theme.Reasonix">
<intent-filter>
<action android:name="android.intent.action.MAIN" />
<category android:name="android.intent.category.LAUNCHER" />
</intent-filter>
</activity>
<!-- The kernel outlives the screen: a turn that takes minutes would
otherwise be killed when the phone is pocketed. The service is
`stopWithTask="false"` so swiping the app away ends it deliberately
rather than silently, and `foregroundServiceType="dataSync"` is the
closest honest description of "an agent is working". -->
<service
android:name=".KernelService"
android:exported="false"
android:foregroundServiceType="dataSync"
android:stopWithTask="false" />
<!-- Reading the pairing code a computer shows. Its own activity rather
than a dialog, because the camera preview needs the whole screen to
be worth pointing at anything. Not exported: only this app starts
it, and a scanner another app could open is one that could be made
to read a code on someone else's behalf. -->
<activity
android:name=".ScanActivity"
android:exported="false"
android:screenOrientation="portrait"
android:theme="@style/Theme.Reasonix.Scan" />
</application>
</manifest>
@@ -1,59 +0,0 @@
package io.reasonix.mobile.app
import android.app.Application
import android.app.NotificationChannel
import android.app.NotificationManager
import android.content.Context
/**
* The process-wide facts this app needs.
*
* The kernel is one child process, so there is exactly one of it no matter how
* many activities exist: it is held here, lazily, and the activity and the
* service both reach the same instance. Nothing else is stored — the theme, the
* language and every preference belong to the page and live in its localStorage.
*/
class App : Application() {
/**
* The kernel, created on first use.
*
* Lazy rather than eager so that merely launching the process to answer a
* notification does not unpack a 15MB page.
*/
val kernel: Kernel by lazy { Kernel(this) }
override fun onCreate() {
super.onCreate()
createChannel()
}
/**
* A channel the person can mute without muting the app, and one they can
* find: an agent that runs for minutes on a phone has to say so somewhere
* the system will not kill, and a notification is the only place Android
* offers. `IMPORTANCE_MIN` because the notification is a receipt, not an
* interruption — the work it reports is on another machine.
*/
private fun createChannel() {
// No SDK_INT guard: minSdk is 26, which is the release NotificationChannel
// arrived in, so every device this app installs on has it.
val manager = getSystemService(Context.NOTIFICATION_SERVICE) as NotificationManager
if (manager.getNotificationChannel(CHANNEL_KERNEL) != null) return
val channel = NotificationChannel(
CHANNEL_KERNEL,
getString(R.string.channel_kernel),
NotificationManager.IMPORTANCE_MIN,
).apply {
description = getString(R.string.channel_kernel_desc)
setShowBadge(false)
enableVibration(false)
setSound(null, null)
}
manager.createNotificationChannel(channel)
}
companion object {
const val CHANNEL_KERNEL = "reasonix-kernel"
}
}
@@ -1,699 +0,0 @@
package io.reasonix.mobile.app
import android.graphics.Bitmap
import android.graphics.Canvas
import android.os.Handler
import android.os.Looper
import android.util.Base64
import android.util.Log
import android.webkit.WebView
import org.json.JSONArray
import org.json.JSONObject
import java.io.BufferedInputStream
import java.io.ByteArrayOutputStream
import java.io.OutputStream
import java.net.ServerSocket
import java.net.Socket
import java.net.SocketException
import java.util.concurrent.CountDownLatch
import java.util.concurrent.TimeUnit
import java.util.concurrent.atomic.AtomicBoolean
import java.util.concurrent.atomic.AtomicReference
/**
* The kernel's browser, as a loopback HTTP server.
*
* The Go kernel (`internal/platform/browser/shell.go`) drives a WebView by
* POSTing JSON to a handful of paths on 127.0.0.1. This class is the other end
* of that conversation: it accepts connections on a random port, parses the
* minimal HTTP the kernel sends, and translates each path into an operation on
* the overlay WebView.
*
* The server is deliberately tiny — no connection pooling, no keep-alive, no
* chunked encoding — because the kernel is its only client and sends one
* request per connection. A `ServerSocket` accept loop on a daemon thread is
* the whole transport; each request is handled synchronously on its own thread.
*
* WebView operations must run on the UI thread. Each handler posts to the main
* looper and waits on a latch, with a timeout that bounds the worst case (a
* page that hangs the JS thread) to something the kernel's own 30-second client
* timeout will not outlast.
*
* Memory note on `/screenshot`: a full-page `WebView.draw` into an ARGB_8888
* bitmap at device resolution can reach tens of MB for a long page. The bitmap
* is downsampled by a factor of 2 in each dimension before PNG encoding, which
* keeps the peak allocation under ~10 MB for a typical 1080p viewport. The
* original is recycled immediately after the downsample.
*/
/** One thing a page asked the network for, as the WebView's request
* interceptor reports it. The interceptor is handed the method and the URL
* and nothing about the response, so a caller may show exactly this much and
* must not read it as a completed exchange. */
data class BrowserRequest(val method: String, val url: String, val kind: String)
class BrowserShellServer(
/** The overlay WebView, when one already exists. The routes that act on an
* open page — back, forward, reload, close — use this: with no page there
* is nothing to act on, and creating one to do nothing to it would start a
* renderer nobody asked for. */
private val existingView: () -> WebView?,
/** The overlay WebView, created on first use. The routes that drive a page
* use this, because the kernel can open one before the panel has ever been
* shown: the page's own panel asks for the view when it is displayed, and
* an agent's `browser_open` can arrive first. Requiring the view to exist
* already answered every such call with "browser view not ready", which
* reads as a broken engine rather than as a view that was not created yet.
* Called on the UI thread. */
private val ensureView: () -> WebView,
/** Why the page in the view is the browser's own error page rather than a
* site, as `(url, chromium error)`; null when the last load succeeded.
* `/read` asks this, because extracting `innerText` from the error page
* returns the browser's own "this page could not be loaded" notice, which
* the kernel would hand to the model as though the site had said it. */
private val loadFailure: () -> Pair<String, String>? = { null },
/** Begin a navigation: forget the load failure recorded for the page being
* left, drop its request list, and start the wait a following `/read`
* joins. Called on the UI thread, before the new address is loaded, so the
* wait is in place before any callback for the new page can fire. */
private val startNavigation: () -> Unit = {},
/** What the page currently in the view has asked the network for, oldest
* first. The WebView reports a request's method and URL to its interceptor
* but nothing about the response, so that is all a caller may show. */
private val requests: () -> List<BrowserRequest> = { emptyList() },
/** Wait for the navigation last asked for to settle — the page arrived, or
* its load failed — up to the given milliseconds. Called on the request's
* own thread, never the UI thread: the settle is signalled from the UI
* thread, so waiting on it there would deadlock. `/load` starts a
* navigation and returns before the page has been visited; without this a
* `/read` right behind it still sees the URL of the page before, and a
* failure recorded for that page was handed back as the new one's answer. */
private val awaitNavigation: (Long) -> Unit = {},
) {
companion object {
private const val TAG = "BrowserShellServer"
/** How long a UI-thread operation may take before the handler gives up.
* The kernel's client timeout is 30 s; staying under it means the kernel
* sees a clean error rather than a dropped connection. */
private const val UI_TIMEOUT_MS = 25_000L
/** The maximum text `/read` returns. Matches the kernel's `maxReadText`
* (400 << 10). The shell truncates here and says so, rather than letting
* the kernel truncate silently. */
private const val MAX_READ_TEXT = 400 * 1024
/** Screenshot downsample factor. 2 means half width and half height. */
private const val SCREENSHOT_SCALE = 2
/** The address the kernel dials. Both ends must agree on the family:
* see [start] for what goes wrong when they do not. */
private const val LOOPBACK_V4 = "127.0.0.1"
/** How long `/read` waits for a navigation it follows to settle before
* reading whatever the view is on. A page that never finishes loading
* must not hold the tool call open to the kernel's own timeout. */
private const val NAV_SETTLE_MS = 10_000L
/** How long `/read` then waits for the page's own script to answer.
* `NAV_SETTLE_MS + READ_SCRIPT_MS` has to stay under [UI_TIMEOUT_MS],
* and that under the kernel's 30-second client timeout: a read that
* waited out both budgets would outlast the caller, which would see a
* dropped connection instead of this shell's answer. 10 + 15 = 25 s. */
private const val READ_SCRIPT_MS = 15_000L
}
@Volatile
private var serverSocket: ServerSocket? = null
/** The port the server is listening on. Valid after [start] returns. */
@Volatile
var port: Int = 0
private set
/**
* The bearer credential every request must carry. Minted by the activity
* and handed to the kernel through the child's environment, so the server
* answers only the process this shell started.
*
* A phone's loopback is not a boundary: any other app may connect to
* 127.0.0.1, and the routes here run script in and read text out of
* whatever page is loaded. Binding to `getLoopbackAddress()` keeps other
* machines out but not other apps, so the credential is what makes "only
* the kernel this shell started" true.
*/
@Volatile
var token: String = ""
private set
private val mainHandler = Handler(Looper.getMainLooper())
/**
* Start listening on 127.0.0.1 with an ephemeral port. The port is available
* via [port] immediately after this returns, and the credential the kernel
* must present via [token].
*/
fun start() {
token = newToken()
// The loopback address is spelled out as IPv4, not taken from
// `InetAddress.getLoopbackAddress()`. That call is allowed to answer
// either family and Android's libcore answers IPv6 (`::1`), while the
// kernel's client is hardcoded to `http://127.0.0.1:<port>`
// (`internal/assembly/boot/browser.go`). A socket bound to `::1` does
// not answer a dial to `127.0.0.1`, so every browser call failed with
// "connection refused" while the port was bound and reported — the
// failure looked like a server that never started. Both ends have to
// name the same family, and IPv4 is the one the kernel names.
val ss = ServerSocket(0, 16, java.net.InetAddress.getByName(LOOPBACK_V4))
serverSocket = ss
port = ss.localPort
Log.i(TAG, "browser shell server listening on $LOOPBACK_V4:$port")
Thread({
while (!ss.isClosed) {
try {
val client = ss.accept()
Thread({ handleClient(client, token) }, "browser-shell-conn").apply { isDaemon = true }.start()
} catch (_: SocketException) {
// Server was closed; exit the loop.
}
}
}, "browser-shell-accept").apply { isDaemon = true }.start()
}
fun stop() {
runCatching { serverSocket?.close() }
serverSocket = null
}
/** A fresh 256-bit credential, hex-encoded, for one launch of the server. */
private fun newToken(): String {
val bytes = ByteArray(32)
java.security.SecureRandom().nextBytes(bytes)
return bytes.joinToString("") { "%02x".format(it) }
}
// ---- HTTP plumbing --------------------------------------------------------
private fun handleClient(client: Socket, expected: String) {
try {
client.soTimeout = 30_000
val input = BufferedInputStream(client.getInputStream())
val output = client.getOutputStream()
// Parse the request line.
val requestLine = readLine(input) ?: return
val parts = requestLine.split(" ", limit = 3)
if (parts.size < 2) return
val method = parts[0]
val path = parts[1].substringBefore('?')
// Read headers.
var contentLength = 0
var bearer: String? = null
while (true) {
val header = readLine(input) ?: break
if (header.isEmpty()) break
val lower = header.lowercase()
if (lower.startsWith("content-length:")) {
contentLength = header.substringAfter(':').trim().toIntOrNull() ?: 0
} else if (lower.startsWith("authorization:")) {
bearer = header.substringAfter(':').trim().removePrefix("Bearer ").trim()
}
}
// Nothing behind this server is safe to hand out on the strength of
// an address: the port is on loopback, which any other app on the
// phone can dial, and the routes run script in and read text out of
// the loaded page. The credential the kernel was started with is the
// only thing that says a caller is that kernel. Every route is
// refused without it, /health included — a probe needs no exemption.
if (expected.isEmpty() || bearer != expected) {
writeResponse(output, 401, """{"error":"unauthorized"}""")
return
}
// Read body.
val body = if (contentLength > 0) {
val buf = ByteArray(contentLength)
var read = 0
while (read < contentLength) {
val n = input.read(buf, read, contentLength - read)
if (n == -1) break
read += n
}
String(buf, 0, read, Charsets.UTF_8)
} else ""
// Route.
val (status, responseBody) = route(method, path, body)
writeResponse(output, status, responseBody)
} catch (e: Exception) {
Log.w(TAG, "connection error: ${e.message}")
} finally {
runCatching { client.close() }
}
}
private fun route(method: String, path: String, body: String): Pair<Int, String> {
return when {
method == "GET" && path == "/health" -> 200 to """{"ok":true}"""
method == "POST" && path == "/load" -> handleLoad(body)
method == "POST" && path == "/close" -> handleClose()
method == "POST" && path == "/read" -> handleRead()
method == "POST" && path == "/screenshot" -> handleScreenshot()
method == "POST" && path == "/click" -> handleClick(body)
method == "POST" && path == "/eval" -> handleEval(body)
method == "POST" && path == "/back" -> handleBack()
method == "POST" && path == "/forward" -> handleForward()
method == "POST" && path == "/reload" -> handleReload()
method == "POST" && path == "/requests" -> handleRequests()
else -> 404 to """{"error":"not found"}"""
}
}
private fun writeResponse(out: OutputStream, status: Int, body: String) {
val reason = when (status) {
200 -> "OK"
404 -> "Not Found"
500 -> "Internal Server Error"
else -> "Error"
}
val bytes = body.toByteArray(Charsets.UTF_8)
val header = "HTTP/1.1 $status $reason\r\n" +
"Content-Type: application/json\r\n" +
"Content-Length: ${bytes.size}\r\n" +
"Connection: close\r\n" +
"\r\n"
out.write(header.toByteArray(Charsets.UTF_8))
out.write(bytes)
out.flush()
}
private fun readLine(input: BufferedInputStream): String? {
val sb = StringBuilder()
while (true) {
val b = input.read()
if (b == -1) return if (sb.isEmpty()) null else sb.toString()
if (b == '\r'.code) {
val next = input.read()
if (next == '\n'.code) return sb.toString()
sb.append(b.toChar())
if (next != -1) sb.append(next.toChar())
continue
}
if (b == '\n'.code) return sb.toString()
sb.append(b.toChar())
}
}
// ---- endpoint handlers ----------------------------------------------------
/** POST /load {"url":"..."} → {"url":"...","title":"..."} */
private fun handleLoad(body: String): Pair<Int, String> {
val json = runCatching { JSONObject(body) }.getOrNull()
?: return 500 to """{"error":"invalid request body"}"""
val url = json.optString("url", "")
if (url.isEmpty()) return 500 to """{"error":"missing url"}"""
val result = AtomicReference("")
val latch = CountDownLatch(1)
mainHandler.post {
val wv = ensureView()
// The failure recorded for the page being left stops being this
// tab's, and the wait a following `/read` joins is armed before the
// new address is loaded, so no callback for the new page can fire
// ahead of it. `/load` answers before the page arrives, which is
// what left the old failure to be read back as the new page's.
startNavigation()
wv.loadUrl(url)
// The title and final URL are not available synchronously after
// loadUrl; answer with what was requested. The kernel stores this
// as the tab's URL and reads the real one on the next /read.
result.set(JSONObject().apply {
put("url", url)
put("title", "")
}.toString())
latch.countDown()
}
if (!latch.await(UI_TIMEOUT_MS, TimeUnit.MILLISECONDS)) {
return 500 to """{"error":"timeout waiting for UI thread"}"""
}
return 200 to result.get()
}
/** POST /close {} → 200 */
private fun handleClose(): Pair<Int, String> {
val latch = CountDownLatch(1)
mainHandler.post {
val wv = existingView()
if (wv != null) {
wv.loadUrl("about:blank")
wv.clearHistory()
}
latch.countDown()
}
if (!latch.await(UI_TIMEOUT_MS, TimeUnit.MILLISECONDS)) {
return 500 to """{"error":"timeout"}"""
}
return 200 to "{}"
}
/**
* POST /read {} → {"title":"...","url":"...","text":"..."}
*
* Injects a script that extracts readable text from the page, preferring
* `<article>` or `<main>` and falling back to `body.innerText`. The result
* is truncated to 400 KB (matching the kernel's `maxReadText`); when
* truncated, a note is appended so the model knows the text was cut.
*/
private fun handleRead(): Pair<Int, String> {
// The page the caller just asked for has not necessarily been visited
// yet: `/load` returns as soon as it has started the navigation. Waiting
// here, off the UI thread, is what makes this read answer about that
// page instead of the one before it.
awaitNavigation(NAV_SETTLE_MS)
val result = AtomicReference("")
val latch = CountDownLatch(1)
mainHandler.post {
val wv = ensureView()
// A page that did not load is not content. Whether the view is on
// the browser's own error page is read from the view itself rather
// than from a remembered failure: a failure recorded for the page
// before a new navigation would otherwise be handed back as this
// page's answer, long after the page the caller asked for had
// loaded. The remembered failure is still what names the address
// that failed and why, since the error page's own URL does not.
if ((wv.url ?: "").startsWith("chrome-error:")) {
val (url, reason) = loadFailure() ?: ((wv.url ?: "") to "ERR_FAILED")
result.set(JSONObject().apply {
put("url", url)
put("title", "")
put("reason", reason)
put("text", "")
}.toString())
latch.countDown()
return@post
}
// The extraction script: prefer article/main for content pages,
// fall back to body. innerText preserves visual line breaks. It
// also stamps a ref on every element a person could act on, so the
// kernel can name one in an act step the way the desktop backend
// names an accessibility node. A ref already on an element is kept,
// so it stays the same across snapshots of one document; a fresh
// document starts the numbering over, and the kernel only accepts a
// ref from the snapshot it last read.
val script = """
(function(){
var seq = window.__rxRefSeq || 0;
var used = {};
var marked = document.querySelectorAll('[data-rx-ref]');
for (var i = 0; i < marked.length; i++) { used[marked[i].getAttribute('data-rx-ref')] = true; }
function roleOf(el) {
var tag = el.tagName.toLowerCase();
var role = (el.getAttribute('role') || '').toLowerCase();
if (role) return role;
if (tag === 'a') return 'link';
if (tag === 'button') return 'button';
if (tag === 'select') return 'combobox';
if (tag === 'textarea') return 'textbox';
if (tag === 'input') {
var t = (el.getAttribute('type') || 'text').toLowerCase();
if (t === 'checkbox' || t === 'radio' || t === 'submit' || t === 'button' || t === 'reset') return t;
return 'textbox';
}
return 'generic';
}
function nameOf(el) {
var n = el.getAttribute('aria-label') || el.getAttribute('placeholder') || el.getAttribute('name') || '';
if (!n) { n = (el.innerText || el.textContent || '').trim(); }
return n.replace(/\s+/g, ' ').slice(0, 160);
}
var sel = 'a[href],button,input,select,textarea,[role=button],[role=link],[role=checkbox],[role=radio],[role=tab],[role=menuitem],[contenteditable=true]';
var els = document.querySelectorAll(sel);
var elements = [];
for (var j = 0; j < els.length; j++) {
var el = els[j];
var ref = el.getAttribute('data-rx-ref');
if (!ref) {
do { seq++; ref = 'e' + seq; } while (used[ref]);
el.setAttribute('data-rx-ref', ref);
used[ref] = true;
}
var raw = (el.value != null && String(el.value) !== '') ? String(el.value).slice(0, 160) : '';
elements.push({ref: ref, role: roleOf(el), name: nameOf(el), value: raw});
}
window.__rxRefSeq = seq;
var el2 = document.querySelector('article') || document.querySelector('main') || document.body;
var text = el2 ? (el2.innerText || '') : '';
var title = document.title || '';
var url = location.href;
return JSON.stringify({title: title, url: url, text: text, elements: elements});
})()
""".trimIndent()
wv.evaluateJavascript(script) { value ->
// evaluateJavascript returns a JSON-encoded string (the result
// is double-encoded). Strip the outer quotes.
val parsed = parseJsResult(value)
if (parsed != null) {
try {
val obj = JSONObject(parsed)
var text = obj.optString("text", "")
if (text.length > MAX_READ_TEXT) {
text = text.substring(0, MAX_READ_TEXT) +
"\n\n[text truncated at 400 KB]"
}
result.set(JSONObject().apply {
put("title", obj.optString("title", ""))
put("url", obj.optString("url", ""))
put("text", text)
put("elements", obj.optJSONArray("elements") ?: JSONArray())
}.toString())
} catch (e: Exception) {
result.set(JSONObject().apply {
put("title", "")
put("url", wv.url ?: "")
put("text", parsed.take(MAX_READ_TEXT))
}.toString())
}
} else {
result.set(JSONObject().apply {
put("title", "")
put("url", wv.url ?: "")
put("text", "")
}.toString())
}
latch.countDown()
}
}
if (!latch.await(READ_SCRIPT_MS, TimeUnit.MILLISECONDS)) {
return 500 to """{"error":"timeout waiting for page script"}"""
}
return 200 to result.get()
}
/**
* POST /screenshot {} → {"data":"<base64>","mime":"image/png"}
*
* Draws the WebView into a bitmap, downsamples by SCREENSHOT_SCALE to keep
* memory bounded, and encodes as PNG → base64. For a 1080×2400 viewport the
* full bitmap is ~10 MB; the downsampled one is ~2.5 MB. The full bitmap is
* recycled immediately after scaling.
*/
private fun handleScreenshot(): Pair<Int, String> {
val result = AtomicReference("")
val failed = AtomicBoolean(false)
val latch = CountDownLatch(1)
mainHandler.post {
val wv = ensureView()
try {
val width = wv.width.takeIf { it > 0 } ?: 1080
val height = wv.height.takeIf { it > 0 } ?: 1920
val bmp = Bitmap.createBitmap(width, height, Bitmap.Config.ARGB_8888)
val canvas = Canvas(bmp)
wv.draw(canvas)
// Downsample to bound the PNG size and the base64 string.
val scaled = Bitmap.createScaledBitmap(
bmp,
width / SCREENSHOT_SCALE,
height / SCREENSHOT_SCALE,
true,
)
if (scaled !== bmp) bmp.recycle()
val baos = ByteArrayOutputStream()
scaled.compress(Bitmap.CompressFormat.PNG, 85, baos)
scaled.recycle()
val b64 = Base64.encodeToString(baos.toByteArray(), Base64.NO_WRAP)
result.set(JSONObject().apply {
put("data", b64)
put("mime", "image/png")
}.toString())
} catch (e: Exception) {
// A 200 with an error body would be read by the kernel as a
// screenshot whose data is empty, and it would build a `data:`
// URI out of nothing. The status is what tells it this failed.
result.set("""{"error":"screenshot failed: ${e.message}"}""")
failed.set(true)
}
latch.countDown()
}
if (!latch.await(UI_TIMEOUT_MS, TimeUnit.MILLISECONDS)) {
return 500 to """{"error":"timeout"}"""
}
if (failed.get()) return 500 to result.get()
return 200 to result.get()
}
/** POST /click {"selector":"..."} → {"ok":true|false} */
private fun handleClick(body: String): Pair<Int, String> {
val json = runCatching { JSONObject(body) }.getOrNull()
?: return 500 to """{"error":"invalid request body"}"""
val selector = json.optString("selector", "")
if (selector.isEmpty()) return 500 to """{"error":"missing selector"}"""
val result = AtomicReference("")
val latch = CountDownLatch(1)
mainHandler.post {
val wv = ensureView()
// Escape the selector for embedding in a JS string.
val escaped = selector.replace("\\", "\\\\").replace("'", "\\'")
val script = """
(function(){
try {
var el = document.querySelector('$escaped');
if (!el) return JSON.stringify({ok: false});
el.click();
return JSON.stringify({ok: true});
} catch(e) {
return JSON.stringify({ok: false});
}
})()
""".trimIndent()
wv.evaluateJavascript(script) { value ->
val parsed = parseJsResult(value)
result.set(parsed ?: """{"ok":false}""")
latch.countDown()
}
}
if (!latch.await(UI_TIMEOUT_MS, TimeUnit.MILLISECONDS)) {
return 500 to """{"error":"timeout"}"""
}
return 200 to result.get()
}
/** POST /eval {"script":"..."} → {"result":"..."} */
private fun handleEval(body: String): Pair<Int, String> {
val json = runCatching { JSONObject(body) }.getOrNull()
?: return 500 to """{"error":"invalid request body"}"""
val script = json.optString("script", "")
if (script.isEmpty()) return 500 to """{"error":"missing script"}"""
val result = AtomicReference("")
val latch = CountDownLatch(1)
mainHandler.post {
val wv = ensureView()
wv.evaluateJavascript(script) { value ->
// The kernel expects a plain string in "result". The value from
// evaluateJavascript is JSON-encoded; unwrap it.
val unwrapped = parseJsResult(value) ?: ""
result.set(JSONObject().apply {
put("result", unwrapped)
}.toString())
latch.countDown()
}
}
if (!latch.await(UI_TIMEOUT_MS, TimeUnit.MILLISECONDS)) {
return 500 to """{"error":"timeout waiting for eval"}"""
}
return 200 to result.get()
}
/** POST /back → 200 */
private fun handleBack(): Pair<Int, String> {
val latch = CountDownLatch(1)
mainHandler.post {
existingView()?.let { if (it.canGoBack()) it.goBack() }
latch.countDown()
}
if (!latch.await(UI_TIMEOUT_MS, TimeUnit.MILLISECONDS)) {
return 500 to """{"error":"timeout"}"""
}
return 200 to "{}"
}
/** POST /forward → 200 */
private fun handleForward(): Pair<Int, String> {
val latch = CountDownLatch(1)
mainHandler.post {
existingView()?.let { if (it.canGoForward()) it.goForward() }
latch.countDown()
}
if (!latch.await(UI_TIMEOUT_MS, TimeUnit.MILLISECONDS)) {
return 500 to """{"error":"timeout"}"""
}
return 200 to "{}"
}
/** POST /reload → 200 */
private fun handleReload(): Pair<Int, String> {
val latch = CountDownLatch(1)
mainHandler.post {
val wv = existingView()
// A view sitting on the browser's own error page is not on the site,
// and reloading it would retry the error page itself. The address
// that failed is loaded again instead — which is the page a caller
// means by "reload" — read before `startNavigation` clears it.
val failed = if (wv != null && (wv.url ?: "").startsWith("chrome-error:"))
loadFailure()?.first else null
startNavigation()
if (wv != null) {
if (failed != null) wv.loadUrl(failed) else wv.reload()
}
latch.countDown()
}
if (!latch.await(UI_TIMEOUT_MS, TimeUnit.MILLISECONDS)) {
return 500 to """{"error":"timeout"}"""
}
return 200 to "{}"
}
/** POST /requests {} → {"requests":[{"method","url","kind"},…]} */
private fun handleRequests(): Pair<Int, String> {
val arr = JSONArray()
for (r in requests()) {
arr.put(JSONObject().apply {
put("method", r.method)
put("url", r.url)
put("kind", r.kind)
})
}
return 200 to JSONObject().apply { put("requests", arr) }.toString()
}
// ---- helpers ----------------------------------------------------------------
/**
* Unwrap the result of `evaluateJavascript`. The callback receives a
* JSON-encoded string: a JS string "hello" arrives as `"\"hello\""`, a
* number 42 as `"42"`, null as `"null"`. Strip the outer JSON encoding to
* get the raw value.
*/
private fun parseJsResult(value: String?): String? {
if (value == null || value == "null") return null
// The value is a JSON literal. If it is a JSON string, decode it.
return try {
if (value.startsWith("\"") && value.endsWith("\"")) {
JSONObject().apply { }.let {
// Use JSONObject's string decoding: wrap in an object.
JSONObject("""{"v":$value}""").optString("v", "")
}
} else {
value
}
} catch (_: Exception) {
value
}
}
}
@@ -1,119 +0,0 @@
package io.reasonix.mobile.app
import android.content.Context
import android.net.ConnectivityManager
import android.net.LinkProperties
import android.os.Build
/**
* The phone's DNS servers, as the kernel needs them.
*
* The kernel is a CGO-free Go binary, and Go's own source explains why that
* matters:
*
* ```
* // DNS requests don't work on Android, so prefer the cgo resolver.
* // Issue #10714.
* case "android":
* ```
*
* A normal Android program resolves names through `getaddrinfo`, which reaches
* the `netd` daemon and gets the servers of the network that is actually up.
* A static binary has no cgo resolver, so Go reads `/etc/resolv.conf` instead —
* and Android has no such file. Go then falls back to `127.0.0.1:53`, where
* nothing listens, and every lookup fails: the setup screen reports
* "连接失败" for every provider, including the official ones, on a phone whose
* browser reaches them fine.
*
* So they are read here and handed to the kernel through `REASONIX_DNS`. This is
* the same source the system itself uses, so a VPN, a private DNS hostname, or a
* network that changed since boot is all reflected without this code knowing
* anything about them.
*/
object Dns {
/**
* The active network's servers, comma separated, or "" when there is none.
*
* Read from the *active* network rather than from every link, because a phone
* commonly has several at once — Wi-Fi, mobile data, a VPN — and the servers
* of a network that is not carrying traffic are the wrong ones to ask.
*/
fun servers(context: Context): String {
val manager = context.getSystemService(Context.CONNECTIVITY_SERVICE) as? ConnectivityManager
?: return ""
val properties: LinkProperties = runCatching { manager.activeNetwork }.getOrNull()
?.let { network -> runCatching { manager.getLinkProperties(network) }.getOrNull() }
?: return ""
return properties.dnsServers
.mapNotNull { address -> address.hostAddress }
.filter { it.isNotBlank() }
.filter { routable(it) }
.sortedBy { preference(it) }
.joinToString(",")
}
/**
* Whether an address can actually carry a query.
*
* A link-local address (fe80::/10) without its zone cannot: the packet has
* no interface to leave by, so it is dropped, nothing answers, and the
* lookup hangs until the kernel's own timeout expires — which is reported as
* a timeout rather than as a failure to resolve, and so reads like the
* network is slow instead of like the address is unusable.
*
* Android reports the zone as a `%wlan0` suffix on `hostAddress`, and the
* address is only usable with it kept. Nothing here trims it: an earlier
* version called `substringBefore('%')`, which turned a usable link-local
* server into an unroutable one.
*/
private fun routable(host: String): Boolean =
!(host.startsWith("fe80:", ignoreCase = true) && !host.contains('%'))
/**
* IPv4 first, then IPv6.
*
* A phone commonly reports both, and its IPv6 entry is often a router's
* link-local address — reachable on one network and not the next. The
* kernel tries the servers in the order it is given them, and stops at the
* first that answers; putting the address family that shares the network's
* own numbering first means its timeout is spent on a server that works.
*/
private fun preference(host: String): Int = if (host.contains(':')) 1 else 0
/**
* The private-DNS setting, for the one case the address list cannot answer.
*
* Since Android 9 a phone may use DNS-over-TLS, and then `dnsServers` holds a
* name rather than an address — a name this kernel cannot resolve *because*
* resolving it is the thing that is broken. There is no way for the shell to
* do that lookup for it, so this reports the situation instead of handing
* over an unusable value: the caller says so in the log, and the person can
* turn private DNS off or set it to automatic.
*/
fun looksLikeDot(servers: String): Boolean =
servers.isNotBlank() && servers.split(',').any { part ->
val host = part.trim()
host.isNotEmpty() && host.none { it == ':' } && !host.all { it.isDigit() || it == '.' }
}
/** Whether the system requires a VPN or private DNS, which changes nothing
* here but is worth saying in a bug report. */
fun describe(context: Context): String {
val manager = context.getSystemService(Context.CONNECTIVITY_SERVICE) as? ConnectivityManager
?: return "no connectivity service"
val network = runCatching { manager.activeNetwork }.getOrNull() ?: return "no active network"
val properties = runCatching { manager.getLinkProperties(network) }.getOrNull()
?: return "no link properties"
// The DHCP server arrived in API 30, not 23: asked for below that it is a
// NoSuchMethodError on a device this app supports (minSdk 26), and the
// answer is only ever read out of this string by a person debugging.
val dhcp = if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.R) {
properties.dhcpServerAddress?.hostAddress
} else {
null
}
return "iface=${properties.interfaceName} dns=${properties.dnsServers.joinToString(",") { it.hostAddress ?: "?" }}" +
(dhcp?.let { " dhcp=$it" } ?: "")
}
}
@@ -1,87 +0,0 @@
package io.reasonix.mobile.app
import android.content.Context
import android.net.Uri
import android.webkit.CookieManager
import android.widget.Toast
import kotlinx.coroutines.Dispatchers
import kotlinx.coroutines.withContext
import java.io.FileNotFoundException
import java.net.HttpURLConnection
import java.net.URL
/**
* Putting a file the page produced onto the phone's disk.
*
* A WebView has no downloads shelf and cannot write to the system's document
* picker itself, so the two halves are joined here: the page asks, the picker
* answers with a `content://` the app has been granted, and this copies the
* bytes across.
*
* The request carries the WebView's own cookie, read out of the jar the page
* authenticated in. That is the only reason this is not a plain `openStream`:
* the kernel's download route is behind the session the page established, and a
* shell that fetched without it would get a 403 and no explanation.
*/
object Downloader {
suspend fun run(
context: Context,
url: String,
userAgent: String,
mime: String,
name: String,
target: Uri,
): Result<Unit> = withContext(Dispatchers.IO) {
runCatching {
val connection = (URL(url).openConnection() as HttpURLConnection).apply {
connectTimeout = 15_000
readTimeout = 60_000
requestMethod = "GET"
if (userAgent.isNotBlank()) setRequestProperty("User-Agent", userAgent)
// The download route answers with the same session cookie the
// page is using; without it the kernel refuses the request.
cookieFor(url)?.let { setRequestProperty("Cookie", it) }
}
try {
val code = connection.responseCode
if (code !in 200..299) {
throw IllegalStateException(
context.getString(R.string.download_failed_status, code),
)
}
val resolver = context.contentResolver
val output = resolver.openOutputStream(target, "w")
?: throw FileNotFoundException(target.toString())
output.use { out ->
connection.inputStream.use { input -> input.copyTo(out) }
}
// `copyTo` answers with a byte count; what the caller wants to
// know is that there were no bytes missing, not how many.
Unit
} finally {
connection.disconnect()
}
}
}
/**
* The cookie the page authenticated with, if any.
*
* `CookieManager` is the WebView's own store, so this asks the exact jar the
* request will be checked against rather than a copy this shell kept.
*/
private fun cookieFor(url: String): String? =
runCatching { CookieManager.getInstance().getCookie(url) }.getOrNull()?.takeIf { it.isNotBlank() }
/** A short receipt for a tap that otherwise does nothing visible. */
fun report(context: Context, name: String, result: Result<Unit>) {
val message = result.fold(
onSuccess = { context.getString(R.string.download_saved, name) },
onFailure = { failure ->
context.getString(R.string.download_failed, failure.message ?: "")
},
)
Toast.makeText(context, message, Toast.LENGTH_LONG).show()
}
}
@@ -1,741 +0,0 @@
package io.reasonix.mobile.app
import android.content.Context
import android.content.res.AssetManager
import org.json.JSONArray
import org.json.JSONObject
import android.util.Log
import kotlinx.coroutines.Dispatchers
import kotlinx.coroutines.delay
import kotlinx.coroutines.withContext
import java.io.File
import java.io.FileOutputStream
import java.net.HttpURLConnection
import java.net.URL
import java.security.SecureRandom
import java.util.Locale
/**
* The Reasonix kernel, as this app runs it: one child process serving loopback,
* plus the two things it needs on disk to start — the page it serves and a
* credential for this launch.
*
* The shape is the desktop host's (`cmd/reasonix-studio-host`), narrowed to what
* a phone does not have. That host binds a socket, mints a credential, speaks a
* handshake down a pipe and holds a lease; here the credential goes through
* `--token-file` and the bound address through `--port-file`, because a phone's
* app process has no pipe to hand a child and no reason to care which port was
* free. Everything after that — the kernel's own gate, its auth middleware, its
* CSRF check — is untouched, so the security of the surface is the one upstream
* tested, not one this shell invented.
*
* One instance lives in the app process. The activity asks it for a URL; the
* foreground service asks it to stay alive. Both see the same process, so there
* is nothing to synchronise beyond a lock on start/stop.
*/
class Kernel(private val context: Context) {
/** Where everything the kernel owns lives. Nothing here is backed up. */
private val home = File(context.filesDir, "home")
/**
* The folder the agent works in, which is not the folder the kernel lives in.
*
* These used to be the same directory, and that is what made the workbench
* unreadable: the kernel's home holds `config.toml`, `projects/`, `cache/`,
* `stats/` and the rest of its own bookkeeping, and a workspace pointed at
* it shows all of that as if the person had put it there. Reported as "why
* does the default workspace have so many files, the desktop version does
* not generate these" — the desktop does generate them, in a different
* directory, because its workspace is whatever folder it was opened on.
*
* Split apart, the workbench shows what the agent produced and nothing else.
* `write_roots` follows the workspace, so the file tools stay confined to
* the folder the person actually works in rather than to the kernel's own
* home — which is the stricter answer, and the one the desktop already has.
*/
private val workspaceDir = File(context.filesDir, "workspace")
/**
* The directory the kernel calls `REASONIX_HOME`, for the one caller that
* has to put a file where the kernel can read it — a private key the person
* picked from their Documents folder.
*/
fun homeDir(): File = home
/** The folder the agent works in. */
fun workspaceDir(): File = workspaceDir
/** The built page, unpacked once per version. */
private val pageDir = File(context.filesDir, "page")
private val tokenFile = File(home, "remote/launch.token")
private val portFile = File(home, "remote/serve.port")
private val pidFile = File(home, "remote/serve.pid")
private val lock = Any()
@Volatile
private var process: Process? = null
/** The port the shell's browser HTTP server listens on, set by the
* activity before `start()` is called. Zero means no server is up, and
* the kernel leaves its browser tools suspended (the correct degradation
* on a phone with no Chromium). */
@Volatile
var browserShellPort: Int = 0
/** The bearer credential for that server, set beside the port. Sent on
* every browser call so the server can refuse anything that did not come
* from the kernel this shell started — loopback on a phone is reachable
* by every other app. Empty when there is no server. */
@Volatile
var browserShellToken: String = ""
/** What the WebView should load, once the kernel has bound a port. */
@Volatile
var origin: String? = null
private set
/**
* The launch credential, as the URL fragment the page trades for a cookie.
* Kept in memory only: the file on disk belongs to the kernel's own
* bookkeeping, and a phone's app data is not a place to leave a second copy.
*/
@Volatile
private var token: String? = null
/**
* The URL the WebView loads first. It carries the credential in the
* fragment, which the page exchanges for an HttpOnly cookie before it makes
* any request — the same path a shared `reasonix web` link takes, so the
* token never enters a request line, a history entry or a log.
*/
fun entryUrl(): String? {
val origin = origin ?: return null
val token = token ?: return null
return "$origin/#token=$token"
}
/** The plain origin, for reloads once the cookie is set. */
fun plainUrl(): String? = origin
fun isRunning(): Boolean = process?.isAlive == true
/**
* Bring the kernel up, or answer the URL of the one already up.
*
* Starting is idempotent by intent: the activity and the service both call
* this, and whichever arrives second joins the running kernel instead of
* racing it for the port.
*/
suspend fun start(): Result<String> = withContext(Dispatchers.IO) {
synchronized(lock) {
origin?.let { return@withContext Result.success(it) }
runCatching {
val binary = kernelBinary()
publishPage()
prepareHome()
val launchToken = mintToken()
token = launchToken
val child = launch(binary, launchToken)
// The port is the kernel's answer, not this shell's guess. It
// writes the file after binding, so waiting on the file is
// waiting on the bind.
val bound = awaitPort(child)
val url = "http://$bound"
// A kernel that bound a port and then died would look healthy
// here and broken in the WebView; the two facts are checked
// together, and only a kernel that is answering is adopted.
awaitServing(child, url)
origin = url
process = child
url
}.onFailure { origin = null; process = null }
}
}
/**
* Stop the kernel.
*
* The graceful signal first: the kernel drains its listeners and closes the
* runtimes it opened, which is what lets a later launch adopt the sessions
* rather than find them locked. `destroyForcibly` is the second attempt, and
* only after a wait — a phone that is being shut down should not keep a
* half-killed agent.
*/
fun stop() {
synchronized(lock) {
val p = process ?: return
process = null
origin = null
token = null
runCatching {
p.destroy()
if (!p.waitFor(3, java.util.concurrent.TimeUnit.SECONDS)) {
p.destroyForcibly()
}
}
portFile.delete()
pidFile.delete()
}
}
// ---- the two inputs that ship inside the APK -------------------------
/**
* The kernel binary.
*
* It is packaged as a native library, so Android has already extracted it to
* a directory it allows execution from and given it the executable bit —
* which is the one thing a file under app storage does not get for free.
*/
private fun kernelBinary(): File {
val dir = File(context.applicationInfo.nativeLibraryDir)
val binary = File(dir, "libreasonix.so")
if (!binary.exists()) {
throw IllegalStateException(
"内核未随 APK 安装:${binary.path}\n这是一次构建问题,请重新安装完整 APK。",
)
}
return binary
}
/**
* Write the bundled page where the kernel can serve it.
*
* It goes to app storage rather than being served out of the APK because the
* kernel's `--page` is a directory on a real filesystem; there is no
* archive-aware variant, and adding one would be a change to the kernel for
* this shell's convenience alone.
*
* Unpacking is versioned: the marker holds the app's version name, so an
* update replaces the page and a relaunch does not. A partially written
* directory — the app killed mid-copy — is caught by the same marker never
* having been written, so the next launch rewrites it.
*/
private fun publishPage() {
val marker = File(pageDir, ".unpacked")
val stamp = "${BuildConfig.VERSION_NAME}:${BuildConfig.VERSION_CODE}"
if (marker.isFile && marker.readText().trim() == stamp) return
pageDir.deleteRecursively()
pageDir.mkdirs()
copyAssetTree(context.assets, "page", pageDir)
if (!File(pageDir, "index.html").isFile) {
throw IllegalStateException("随 APK 分发的页面不完整:缺少 index.html")
}
marker.writeText(stamp)
}
/**
* A directory tree out of assets.
*
* `AssetManager.list` on a file answers with an empty array, so a "file" is
* told from a directory by trying to open it — which is the only signal the
* API offers and the reason this is not a one-liner.
*/
private fun copyAssetTree(assets: AssetManager, from: String, to: File) {
val children = assets.list(from).orEmpty()
if (children.isEmpty()) {
// A leaf: open it as a stream. A directory that happens to be empty
// is indistinguishable and harmless to skip.
runCatching {
assets.open(from).use { input ->
FileOutputStream(to).use { output -> input.copyTo(output) }
}
}
return
}
to.mkdirs()
for (child in children) {
copyAssetTree(assets, "$from/$child", File(to, child))
}
}
/**
* Prepare the kernel's home.
*
* `REASONIX_HOME` is what every path in the kernel hangs off — config,
* sessions, the remote state directory its token file lands in — so pointing
* it at app storage is what keeps a phone's Reasonix separate from anything
* else and away from global storage entirely.
*/
private fun prepareHome() {
File(home, "remote").mkdirs()
// The workspace directory not existing yet is what marks the first run
// after this build split it from the kernel's home. That is the one
// moment a migration is due, and doing it here means it happens before
// the kernel is asked for anything.
val firstSplit = !workspaceDir.exists()
workspaceDir.mkdirs()
if (firstSplit) {
adoptSessionsFromHome()
}
// Every launch, not only the first. The rewrite is cheap and
// idempotent, and an entry that survived the first pass would otherwise
// survive forever: `firstSplit` is spent the moment the directory
// exists, so a build that cleaned the list once and missed leaves the
// home on the sidebar for good. Running it unconditionally costs one
// small read and makes the whole thing self-correcting.
forgetHomeFromWorkspaceList()
}
/**
* Drop the kernel's own home from the sidebar's remembered projects.
*
* The list is written whenever a pane opens, and until this build the pane
* opened on the home itself — so an upgraded install carries a "project"
* that is really the kernel's own directory. Leaving it there would put the
* person back where they started: one tap on it and the workbench is full
* of `config.toml`, `cache/` and `projects/` again, which is the complaint
* this whole change answers. Only that one path is removed; every folder
* someone added on purpose stays.
*
* Parsed rather than pattern-matched, because the file's two shapes — a bare
* array of paths and an object with `paths` and `launch` — both have to keep
* whatever else they hold.
*/
private fun forgetHomeFromWorkspaceList() {
val file = File(home, "serve-workspaces.json")
if (!file.isFile) return
val text = runCatching { file.readText() }.getOrNull() ?: return
val wrong = runCatching { homeNames() }.getOrNull() ?: return
val rewritten = runCatching { withoutHome(text, wrong) }.getOrNull() ?: return
if (!rewritten.dropped) return
runCatching { file.writeText(rewritten.text) }
.onSuccess { Log.i(TAG, "dropped the kernel home from the project list") }
.onFailure { Log.w(TAG, "could not rewrite the project list: ${it.message}") }
}
/**
* Every name of the kernel's home on this device — not just the one this
* process happens to write.
*
* Android shows one directory under two names: `/data/data/<pkg>` and
* `/data/user/0/<pkg>` are two mounts of the same tree, the same inode
* either way, and which name a path carries depends on how it was written
* rather than on where it points. The list is written by the kernel, whose
* working directory this shell set from a path it built itself; a lookup
* that compared against only one spelling would silently keep the entry
* this exists to remove. Both names are compared, so either spelling is
* recognised.
*/
private fun homeNames(): Set<String> {
val path = home.absolutePath
val names = mutableSetOf(path)
val user = "/data/user/0/"
val data = "/data/data/"
when {
path.startsWith(user) -> names.add(data + path.removePrefix(user))
path.startsWith(data) -> names.add(user + path.removePrefix(data))
}
return names
}
/** What a rewrite produced, and whether it actually removed anything. */
private class Rewrite(val text: String, val dropped: Boolean)
/**
* The list without the home, in whichever shape it arrived, and whether
* anything was actually dropped from it.
*
* The flag matters because `JSONObject` re-serialises: it escapes every
* `/` and does not preserve key order, so a round trip never reproduces
* the original bytes. Deciding to write on "the text differs" alone would
* therefore rewrite the kernel's file on every single launch, log a drop
* that never happened, and race the kernel for a file it also writes.
*/
private fun withoutHome(text: String, wrong: Set<String>): Rewrite {
if (text.trimStart().startsWith("[")) {
val paths = JSONArray(text)
val kept = JSONArray()
var dropped = false
for (i in 0 until paths.length()) {
val path = paths.optString(i)
if (path in wrong) {
dropped = true
continue
}
kept.put(path)
}
return if (dropped) Rewrite(kept.toString(), true) else Rewrite(text, false)
}
val list = JSONObject(text)
val kept = JSONArray()
var dropped = false
val paths = list.optJSONArray("paths")
if (paths != null) {
for (i in 0 until paths.length()) {
val path = paths.optString(i)
if (path in wrong) {
dropped = true
continue
}
kept.put(path)
}
if (dropped) list.put("paths", kept)
}
if (list.optString("launch") in wrong) {
list.remove("launch")
dropped = true
}
return if (dropped) Rewrite(list.toString(), true) else Rewrite(text, false)
}
/**
* Move the conversations that were filed under the kernel's own home.
*
* The transcript store is keyed by the workspace root — one directory per
* project, under `<state>/projects/<slug>/sessions` — and until this build
* the root was the home itself. Splitting the two therefore moves every
* existing conversation to a slug nothing looks under, which reads as "my
* history is gone" rather than as a reorganisation. So the old directory is
* renamed onto the new slug once.
*
* Renaming the project directory rather than just `sessions/` carries the
* rest of it too: `memory/` is filed the same way, and a project's memory
* lost while its transcripts were kept would be a stranger failure than
* either alone.
*
* The slug is the absolute path with its separators flattened, which is
* what `config.WorkspaceSlug` does on the kernel's side. Only the home's own
* slug is touched: a slug for some other folder is left exactly where it is,
* because those are projects someone added on purpose.
*/
private fun adoptSessionsFromHome() {
val projects = File(home, "projects")
if (!projects.isDirectory) return
val from = projects.resolve(slugOf(home.absolutePath))
if (!from.isDirectory) return
val to = projects.resolve(slugOf(workspaceDir.absolutePath))
// A target already there means a previous run half-finished, or the
// person opened this folder before. Either way its contents win:
// merging two transcript directories is not a thing this can do safely,
// and the old one is still on disk to be looked at.
if (to.exists()) {
Log.i(TAG, "${to.name} already exists; leaving ${from.name} in place")
return
}
if (!from.renameTo(to)) {
Log.w(TAG, "could not move ${from.name} to ${to.name}; history stays under the old slug")
} else {
Log.i(TAG, "moved conversations from ${from.name} to ${to.name}")
}
}
/** The kernel's `config.WorkspaceSlug`, for the two paths that matter here. */
private fun slugOf(path: String): String = path.replace('/', '-').replace('\\', '-').replace(':', '-')
/**
* The launch credential.
*
* Minted here rather than read from configuration: a credential a person can
* set is one anything that can read their files can present. 32 bytes, hex,
* the same shape the desktop host mints.
*/
private fun mintToken(): String {
val bytes = ByteArray(32)
SecureRandom().nextBytes(bytes)
val token = bytes.joinToString("") { "%02x".format(it) }
tokenFile.parentFile?.mkdirs()
tokenFile.writeText("$token\n")
// 0600 by intent, and the app's own private directory means it is
// already unreachable to every other app on the device.
runCatching {
tokenFile.setReadable(false, false)
tokenFile.setReadable(true, true)
tokenFile.setWritable(false, false)
tokenFile.setWritable(true, true)
}
return token
}
/**
* Start the child.
*
* `--addr 127.0.0.1:0` lets the kernel pick a free port; `--port-file`
* makes it announce which. `--auth token` is the mode a phone needs: the
* page authenticates, and a state-changing request without the credential
* is refused. `--no-open` is belt and braces — a serve never opens a
* browser, and a phone has no browser to open.
*/
private fun launch(binary: File, token: String): Process {
portFile.delete()
pidFile.delete()
// The folder the agent works in, resolved before the command line is
// built so it can be named on it. See the note under the environment
// block for why naming it is not the same as setting the directory.
val workspace = launchWorkspaceDir()
val command = listOf(
binary.absolutePath,
"serve",
"--addr", "127.0.0.1:0",
"--auth", "token",
"--token-file", tokenFile.absolutePath,
"--port-file", portFile.absolutePath,
"--pid-file", pidFile.absolutePath,
"--no-open",
"-page", pageDir.absolutePath,
// The project root, said outright. Without this the kernel derives
// it from its working directory, and on this phone the two did not
// agree: the sidebar named the shared folder, `write_roots` confined
// the file tools to the app's own directory, and every write into the
// folder the person was looking at came back
// `workspace.write_outside_scope`. A flag is a fact the kernel was
// told rather than one it inferred from a mount it sees differently.
"--dir", workspace.absolutePath,
)
val builder = ProcessBuilder(command)
builder.environment().apply {
put("REASONIX_HOME", home.absolutePath)
put("HOME", home.absolutePath)
// The kernel's own language, for its tool failures, approval
// prompts and error sentences. Without this it detects from the
// POSIX locale variables, and Android sets none of them for an
// app: the kernel then answers in English while the page in front
// of the person is Chinese, and every sentence added for a Chinese
// reader was dead text. The tag, not the bare language, so a
// Traditional Chinese device resolves to zh-TW rather than zh.
// An explicit `language` in the kernel's config.toml still wins:
// DetectLanguage applies the configured value over this.
val languageTag = Locale.getDefault().toLanguageTag()
if (languageTag.isNotBlank() && languageTag != "und") {
put("REASONIX_LANG", languageTag)
}
// Android has no /tmp. Everything that wants a scratch directory —
// the shell tool, tool output spooling, sqlite journals — reads this.
put("TMPDIR", context.cacheDir.absolutePath)
// The phone's own DNS servers. A CGO-free Go binary cannot read them
// for itself — Android keeps them in `netd`, and Go looks for a
// `/etc/resolv.conf` that does not exist — so without this the kernel
// resolves nothing and every provider probe reports "连接失败" on a
// device whose browser reaches the same address fine.
val dns = Dns.servers(context)
if (dns.isNotBlank()) {
put("REASONIX_DNS", dns)
Log.i(TAG, "kernel DNS: $dns (${Dns.describe(context)})")
if (Dns.looksLikeDot(dns)) {
// Private DNS (DNS-over-TLS) is on, and it hands back a
// hostname this kernel cannot resolve by definition. Saying
// so here is what turns "连不上" into something actionable.
Log.w(TAG, "kernel DNS looks like a private-DNS hostname; turn private DNS off or set it to automatic")
}
} else {
Log.w(TAG, "kernel DNS: none reported by the system")
}
// The shell's browser server, when one is listening. The kernel
// reads this to bind its browser tools to the shell's WebView
// instead of looking for a Chromium binary that does not exist on
// a phone. When absent (zero), the variable is not set and the
// kernel keeps the browser tools suspended — the correct
// degradation for a phone with no engine to drive.
val shellPort = browserShellPort
if (shellPort > 0) {
put("REASONIX_SHELL_BROWSER_PORT", shellPort.toString())
// The credential the server checks. Set only alongside the
// port: a token with nothing listening would be a variable the
// kernel carries for no reason.
val shellToken = browserShellToken
if (shellToken.isNotBlank()) {
put("REASONIX_SHELL_BROWSER_TOKEN", shellToken)
}
}
}
// The agent's folder, not the kernel's. `REASONIX_HOME` above is still
// where the config, the sessions and the transcripts live; this is what
// the workspace root resolves to, because the kernel reads the working
// directory for it.
//
// The folder the person last opened, when it is still usable — the same
// answer the desktop host gives (`launchWorkspace`), and the reason a
// phone session used to work in the app's own `files/workspace` while
// the sidebar said it was in the shared folder: the launch entry in
// `serve-workspaces.json` was written and then ignored, and `write_roots`
// is derived from the working directory, so the file tools were confined
// to the app's private folder and every write into the shared one was
// refused. `workspaceDir` is still the answer for a first run and for a
// folder since deleted, but the choice is never silent — see
// [launchWorkspaceDir]. `workspace` was resolved above so it could also
// be named on the command line.
builder.directory(workspace)
// `PWD` is what the kernel's own `os.Getwd` prefers: Go uses it whenever
// it names the same directory as `.`, which is exactly this case, and it
// is the only way to make the kernel resolve its root to the string this
// shell chose. Without it `os.Getwd` falls back to the kernel's view of
// the mount, and on Android `/storage/emulated/0/...` and the path the
// FUSE layer reports for it need not be spelled the same way. The kernel
// derives `write_roots` from that root, so the two spellings have to be
// the same one.
builder.environment().put("PWD", workspace.absolutePath)
builder.redirectErrorStream(true)
val process = builder.start()
// The kernel's own log is the only place a startup failure says why, and
// a phone has no terminal to read. It is drained on a thread and kept
// ring-buffered so the About screen can show the last of it.
Thread({ drain(process) }, "kernel-log").apply { isDaemon = true }.start()
return process
}
/**
* The folder the person last opened, or null when there is not one.
*
* `serve-workspaces.json` is the kernel's own record of the projects the
* sidebar lists, and its `launch` entry is which of them a session starts
* in. Both shapes the file has are read — a bare array of paths, and an
* object with `paths` and `launch` — because both are written in the wild
* and a reader that only knew one would silently fall back.
*
* The entry has to be one of the listed paths, which is the same rule the
* kernel applies when it reads this file (`workspacelist.Read` blanks a
* `launch` that is not in `paths`). It matters more here than there: the
* working directory decides `write_roots`, so whatever this answers is
* where the agent's file tools may write — and on a phone that permission
* reaches all of shared storage. A file with a `launch` outside its own
* list is one this shell did not write, and it gets the listed answer.
*
* Null for anything unusable: no file, unreadable JSON, a launch that is
* not among the listed folders, an entry that has since been deleted, or a
* path that is now a file. Every one of those means the same thing to the
* caller — start where this shell always started.
*/
private fun rememberedLaunchDir(): File? {
val file = File(home, "serve-workspaces.json")
if (!file.isFile) return null
val text = runCatching { file.readText() }.getOrNull() ?: return null
val launch = runCatching {
// A bare array is the file's other shape — the one older kernels
// wrote, and therefore the one a person upgrading is most likely to
// have. Without this branch `JSONObject(text)` throws on it and the
// whole function falls back, which is the same mismatch (the
// sidebar naming one folder, the session working in another) this
// exists to fix. `withoutHome` below reads both shapes the same way.
if (text.trimStart().startsWith("[")) {
val arr = JSONArray(text)
if (arr.length() == 0) return@runCatching null
arr.optString(0)
} else {
val json = JSONObject(text)
val paths = json.optJSONArray("paths")
?.let { arr -> (0 until arr.length()).mapNotNull { arr.optString(it).takeIf(String::isNotBlank) } }
.orEmpty()
if (paths.isEmpty()) return@runCatching null
json.optString("launch").takeIf { it.isNotBlank() && it in paths } ?: paths.first()
}
}.getOrNull()?.takeIf { it.isNotBlank() } ?: return null
val dir = runCatching { File(launch) }.getOrNull() ?: return null
if (!dir.isDirectory) {
// A launch entry this process cannot stat is exactly the failure the
// caller exists to avoid, and falling back silently is what made it
// take three releases to notice: the session opened in the app's own
// folder while the sidebar named the shared one. A phone has no
// terminal, so the log is the only place the reason can be said —
// and `exists`/`readable` tell apart "deleted" from "not permitted",
// which need different answers from the person.
Log.w(
TAG,
"launch entry $launch is not a usable directory " +
"(exists=${dir.exists()}, readable=${dir.canRead()})",
)
return null
}
return dir
}
/**
* The folder the kernel should start in.
*
* Either the launch entry — the folder the person last opened, which is
* also the one the sidebar shows — or the app's own folder when there is
* none. Whichever it is, it is logged, because the two disagreeing is the
* bug this exists to fix and a silent choice is what let it survive.
*/
private fun launchWorkspaceDir(): File {
val remembered = rememberedLaunchDir()
if (remembered != null) {
Log.i(TAG, "kernel workspace: ${remembered.absolutePath} (the remembered launch project)")
return remembered
}
Log.i(TAG, "kernel workspace: ${workspaceDir.absolutePath} (no remembered launch project)")
return workspaceDir
}
private val logLines = ArrayDeque<String>()
private fun drain(process: Process) {
runCatching {
process.inputStream.bufferedReader().useLines { lines ->
lines.forEach { line ->
Log.i(TAG, line)
synchronized(logLines) {
logLines.addLast(line)
while (logLines.size > LOG_KEEP) logLines.removeFirst()
}
}
}
}
}
/** The kernel's tail, newest last, for a bug report or the About screen. */
fun recentLog(): String = synchronized(logLines) { logLines.joinToString("\n") }
private fun awaitPort(process: Process): String {
val deadline = System.currentTimeMillis() + START_TIMEOUT_MS
while (System.currentTimeMillis() < deadline) {
if (!process.isAlive) {
throw IllegalStateException("内核启动即退出。\n${recentLog().takeLast(2000)}")
}
val text = if (portFile.isFile) runCatching { portFile.readText().trim() }.getOrNull() else null
// The file is written by the kernel on the way to serving, and the
// kernel rewrites it with the real port when it was given :0.
if (!text.isNullOrBlank() && text.contains(":")) return text
Thread.sleep(POLL_MS)
}
throw IllegalStateException("内核在 ${START_TIMEOUT_MS / 1000}s 内没有报告监听地址。\n${recentLog().takeLast(2000)}")
}
/**
* Wait until the bound port answers.
*
* A port that is open but not yet serving is the normal first second of a
* kernel's life; a WebView pointed at it would show a connection error and
* cache the failure, so the wait belongs here rather than in the page.
*/
private fun awaitServing(process: Process, origin: String) {
val deadline = System.currentTimeMillis() + START_TIMEOUT_MS
while (System.currentTimeMillis() < deadline) {
if (!process.isAlive) {
throw IllegalStateException("内核在开始服务前退出。\n${recentLog().takeLast(2000)}")
}
if (probe("$origin/")) return
Thread.sleep(POLL_MS)
}
throw IllegalStateException("内核已监听 $origin,但页面没有响应。\n${recentLog().takeLast(2000)}")
}
private fun probe(url: String): Boolean = runCatching {
(URL(url).openConnection() as HttpURLConnection).run {
connectTimeout = 1000
readTimeout = 1000
requestMethod = "GET"
try {
responseCode in 200..399
} finally {
disconnect()
}
}
}.getOrDefault(false)
private companion object {
const val TAG = "ReasonixKernel"
const val LOG_KEEP = 400
const val POLL_MS = 120L
const val START_TIMEOUT_MS = 30_000L
}
}
@@ -1,139 +0,0 @@
package io.reasonix.mobile.app
import android.Manifest
import android.app.Notification
import android.app.PendingIntent
import android.app.Service
import android.content.Context
import android.content.Intent
import android.content.pm.PackageManager
import android.content.pm.ServiceInfo
import android.os.Build
import android.os.IBinder
import androidx.core.app.NotificationCompat
import androidx.core.app.NotificationManagerCompat
import androidx.core.app.ServiceCompat
import androidx.core.content.ContextCompat
import kotlinx.coroutines.CoroutineScope
import kotlinx.coroutines.Dispatchers
import kotlinx.coroutines.SupervisorJob
import kotlinx.coroutines.cancel
import kotlinx.coroutines.launch
/**
* Keeps the kernel alive while the app is not on screen.
*
* The reason this exists at all: an agent turn that takes minutes is a normal
* turn, and Android reclaims a backgrounded app's child process within seconds
* of the screen going off. Without the service, pocketing the phone would kill
* a running task — the one failure mode that makes a phone a bad client for
* something that works on its own.
*
* A foreground service needs a notification the person can see, which is also
* the honest thing to show: the kernel is running and what it is doing lives on
* another machine.
*/
class KernelService : Service() {
private val scope = CoroutineScope(SupervisorJob() + Dispatchers.Main)
override fun onBind(intent: Intent?): IBinder? = null
override fun onStartCommand(intent: Intent?, flags: Int, startId: Int): Int {
// Promoted before `start()`: a service that starts work first and
// notifies after has a window in which the system may kill it, and the
// work it would have been protecting is exactly what is being started.
startForegroundNotification()
val kernel = (application as App).kernel
scope.launch {
kernel.start().onFailure { failure ->
// A kernel that will not start is a fact the person needs, and a
// notification that only ever said "running" would be a lie.
notify(SERVICE_NOTICE, getString(R.string.notify_failed_title), failure.message.orEmpty())
stopSelf()
}
}
// Sticky, so a kernel killed under memory pressure comes back when there
// is room — which is what a person who left a task running expects.
return START_STICKY
}
override fun onDestroy() {
scope.cancel()
super.onDestroy()
}
private fun startForegroundNotification() {
val notification = buildNotification(
getString(R.string.notify_running_title),
getString(R.string.notify_running_body),
)
// `dataSync` matches the manifest's declaration. On 34+ the type is
// mandatory and a mismatch throws, so it is named in both places.
if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.UPSIDE_DOWN_CAKE) {
ServiceCompat.startForeground(
this,
SERVICE_NOTICE,
notification,
ServiceInfo.FOREGROUND_SERVICE_TYPE_DATA_SYNC,
)
} else {
startForeground(SERVICE_NOTICE, notification)
}
}
private fun buildNotification(title: String, body: String): Notification {
val open = PendingIntent.getActivity(
this,
0,
Intent(this, MainActivity::class.java).apply {
flags = Intent.FLAG_ACTIVITY_NEW_TASK or Intent.FLAG_ACTIVITY_CLEAR_TOP
},
PendingIntent.FLAG_UPDATE_CURRENT or PendingIntent.FLAG_IMMUTABLE,
)
return NotificationCompat.Builder(this, App.CHANNEL_KERNEL)
.setSmallIcon(R.drawable.ic_notification)
.setContentTitle(title)
.setContentText(body)
.setContentIntent(open)
// The notification is a receipt, not an alarm; swiping it away
// leaves the kernel running, which is what the person meant by
// swiping it.
.setOngoing(false)
.setSilent(true)
.setPriority(NotificationCompat.PRIORITY_MIN)
.build()
}
private fun notify(id: Int, title: String, body: String) {
// The notification permission is granted at runtime from Android 13 and
// the person can refuse it. Without this check the call throws
// SecurityException and takes the service down with it — the opposite of
// what a notification that could not be shown should do. Below 13 the
// permission does not exist and posting always succeeds.
if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.TIRAMISU &&
ContextCompat.checkSelfPermission(this, Manifest.permission.POST_NOTIFICATIONS) !=
PackageManager.PERMISSION_GRANTED
) {
return
}
runCatching {
NotificationManagerCompat.from(this).notify(id, buildNotification(title, body))
}
}
companion object {
private const val SERVICE_NOTICE = 1
/** Idempotent: called on every start of the activity, and by the boot
* path, and safe to repeat. */
fun start(context: Context) {
ContextCompat.startForegroundService(context, Intent(context, KernelService::class.java))
}
fun stop(context: Context) {
context.stopService(Intent(context, KernelService::class.java))
}
}
}
File diff suppressed because it is too large. Load diff
@@ -1,744 +0,0 @@
package io.reasonix.mobile.app
import android.content.ClipData
import android.content.ClipboardManager
import android.content.ContentValues
import android.content.Context
import android.content.res.Configuration
import android.net.Uri
import android.os.Build
import android.os.Environment
import android.os.Handler
import android.os.Looper
import android.provider.MediaStore
import android.webkit.JavascriptInterface
import android.webkit.MimeTypeMap
import java.io.File
import java.util.concurrent.CountDownLatch
import java.util.concurrent.TimeUnit
import java.util.concurrent.atomic.AtomicReference
/**
* What the page can ask the phone for.
*
* This is the whole surface, and it is deliberately tiny. Android's
* `@JavascriptInterface` is a bridge into the app's process from a WebView whose
* content is a web page: every method here is reachable by anything that can run
* script in that page, so each one is written as a verb with no address
* parameter it does not need, and none of them can name a path outside the
* kernel's own home.
*
* It is deliberately *not* the Electron preload bridge. That one exposes window
* management, a filesystem walker, an editor launcher and a browser view host —
* all of which are meaningless on a phone, and the two that were not (open a URL,
* save a file) already have a way to happen here: `WebViewClient` routes external
* links to the browser, and the page's own download reaches the WebView's
* `DownloadListener`.
*
* The one thing a phone must add is `importIntoHome`: the kernel runs on this
* device and can only read paths under its own home, so a private key the person
* has in their Documents folder has to be copied across. Nothing else the page
* wants is out of its reach.
*/
class NativeBridge(
private val context: Context,
private val kernel: Kernel,
private val onOpen: (String) -> Unit,
) {
/**
* Whether the document calling may use this bridge.
*
* `addJavascriptInterface` makes the object reachable from whatever page the
* WebView holds, so a link the model produced — or a cleartext page on the
* LAN — would otherwise get `importIntoHome`, `pickFolder`, `hostCall` and
* the clipboard. The activity answers by comparing the main frame's origin
* against the kernel's and the one paired computer the person confirmed.
*
* Anything that reaches the device checks this first. The five methods that
* do not — version, systemTheme, safeArea, setLayerOpen, setRunning — report
* back what the page already knows about its own window and grant nothing.
*/
var trusted: () -> Boolean = { false }
/** The version the page reads to decide whether this bridge is the one it
* knows. Bumped when a method's contract changes, not when one is added. */
@JavascriptInterface
fun version(): String = BRIDGE_VERSION
/**
* Copy a file the person picks into the kernel's home, answering with its
* path there — or "" when they dismissed the picker.
*
* `kind` names the directory inside the home, because that is a fact about
* the kernel's layout and not about the file: an SSH private key belongs in
* `.ssh`, and a dotenv the kernel reads for passphrases belongs at the root
* of the home where `config.Load` looks for it.
*
* The path that comes back is *inside the home*, never the document
* provider's URI. That is the point: the kernel is a child process with no
* access to the provider, so a `content://` would be a path it cannot open.
*/
@JavascriptInterface
fun importIntoHome(kind: String, name: String): String {
if (!trusted()) return ""
val dir = when (kind) {
"identity" -> File(kernel.homeDir(), ".ssh")
"env" -> kernel.homeDir()
else -> return ""
}
val safeName = sanitise(name)
if (safeName.isEmpty()) return ""
val picked = pickFile() ?: return ""
return try {
dir.mkdirs()
val target = File(dir, safeName)
context.contentResolver.openInputStream(picked)?.use { input ->
target.outputStream().use { output -> input.copyTo(output) }
} ?: return ""
// A private key with group or world read is one OpenSSH refuses to
// use, and the refusal reads as "bad permissions" from a long way
// away. Setting it here is what makes the file usable at all.
if (kind == "identity") {
runCatching {
target.setReadable(false, false)
target.setReadable(true, true)
target.setWritable(false, false)
target.setWritable(true, true)
}
}
target.absolutePath
} catch (_: Exception) {
// A copy that failed leaves nothing behind; the caller sees "" and
// the person can try again.
""
}
}
/**
* Open a URL in the device's browser.
*
* The page asks for this when a link points somewhere outside the kernel —
* a provider's console, a documentation page — and loading it in place would
* strand the person in a window with no address bar and no way back.
*/
@JavascriptInterface
fun openExternal(url: String) {
if (!trusted()) return
// Only what a browser can open. A `file://` or `content://` here would
// be the page reaching into the device through a method named for
// leaving it.
if (!url.startsWith("http://") && !url.startsWith("https://")) return
onOpen(url)
}
/** Put text on the clipboard. A WebView's own `navigator.clipboard` write is
* gated behind a permission Android grants only to a focused document, so
* the one thing the page actually offers to copy — a token, a command, a
* path — has no other way out.
*
* Answers whether the write landed. A platform that refuses it raises
* rather than answering, and an exception must not cross
* `@JavascriptInterface` as a silent success — the page would say 已复制
* over an empty clipboard. Caught here, the answer is the one the caller
* can act on: false, and the button says 复制失败 instead. */
@JavascriptInterface
fun copyText(text: String): Boolean = if (!trusted()) false else runCatching {
val clipboard = context.getSystemService(Context.CLIPBOARD_SERVICE) as ClipboardManager
clipboard.setPrimaryClip(ClipData.newPlainText("reasonix", text))
}.isSuccess
/**
* Write text where the person picks, and answer with the path it went to.
*
* The page's own route to a file — a blob URL on an `<a download>` — reaches
* nothing here: a WebView has no downloads shelf, and its `DownloadListener`
* fires only for a real navigation, not for a blob the page just built. So
* the text crosses the bridge and the shell opens the system's document
* picker, which is also the only way to reach a folder the person can find
* again afterwards.
*
* "" is a dismissal, the same answer `importIntoHome` gives, and the page
* reads the two the same way.
*/
/**
* Put a copy of a file where a phone's own file manager will show it.
*
* `saveText` asks the person where to put it, which is the right question
* on a desktop and the wrong one here: the kernel's workspace is
* app-private storage, so the file someone wants is in a place no file
* manager can open, and a picker that starts at "recent" is another place
* to lose it. This answers the actual question — "where is it" — by
* putting it in Downloads, in a folder named after the app, under a name
* that is the file's own.
*
* The answer is the path relative to shared storage, so the page can say
* where the file went rather than only that it worked. "" means the write
* failed or the platform refused it.
*
* MediaStore on Android 10 and later, a plain file before that. The two
* are not interchangeable: from 10 the app may not write another app's
* files directly, and MediaStore is the only way in; before it there is no
* MediaStore collection for downloads and the public directory is writable
* with the storage permission this app does not ask for, so that branch is
* best-effort and says so by answering "".
*/
@JavascriptInterface
fun exportToDownloads(name: String, content: String): String {
if (!trusted()) return ""
val safeName = sanitise(name).ifEmpty { "reasonix.txt" }
return exportFile(safeName, content).orEmpty()
}
@JavascriptInterface
fun saveText(name: String, content: String): String {
if (!trusted()) return ""
val safeName = sanitise(name).ifEmpty { "reasonix.txt" }
return saveFile(safeName, content) ?: ""
}
/**
* Hand text to whatever the device shares with — a chat app, the mail
* client, a note app.
*
* The share sheet rather than a second save dialog: "where is that file"
* is the question this answers, and on a phone a document someone wants to
* keep or send usually goes through one of those apps rather than a folder.
*
* Shared as text with the filename as its subject, which is what every
* target understands. A file attachment would need a `FileProvider` and a
* cache copy for the same words to arrive.
*/
@JavascriptInterface
fun shareText(name: String, content: String): Boolean {
if (!trusted()) return false
val safeName = sanitise(name).ifEmpty { "reasonix.txt" }
val posted = java.util.concurrent.CountDownLatch(1)
var ok = false
main.post {
ok = runCatching { askForShare(safeName, content) }.getOrDefault(false)
posted.countDown()
}
runCatching { posted.await(SHARE_TIMEOUT_SECONDS, TimeUnit.SECONDS) }
return ok
}
/**
* The device's own light/dark answer.
*
* The page resolves its theme from `localStorage` first and only asks the
* system when nothing is saved, so this is what makes a fresh install match
* the phone instead of defaulting to light.
*/
@JavascriptInterface
fun systemTheme(): String {
val night = context.resources.configuration.uiMode and Configuration.UI_MODE_NIGHT_MASK
return if (night == Configuration.UI_MODE_NIGHT_YES) "dark" else "light"
}
/**
* The window's real safe-area insets, in CSS pixels, as
* `top,right,bottom,left`.
*
* A WebView does not implement `env(safe-area-inset-*)`: those are an iOS
* Safari feature, and on Android every one of them resolves to `0`. So a
* page that pads itself with them — which `mobile.css` does — is un-padded
* on exactly the platform this shell exists for, and the first line of the
* interface sits under the status bar.
*
* The values have to come from here for a second reason too: from Android 15
* the window is edge-to-edge whether the app asks for it or not, so the bars
* really are drawn over the page and there is no window-manager switch that
* could move the content back down.
*
* Answered as a string because that is what crosses the bridge; the page
* sets it as a CSS custom property and `mobile.css` reads it.
*/
@JavascriptInterface
fun safeArea(): String {
val insets = onInsets?.invoke() ?: return "0,0,0,0"
return "${insets.left},${insets.top},${insets.right},${insets.bottom}"
}
/**
* Read the pairing code a computer is showing.
*
* A phone cannot type what a computer renders as a QR code — it is a
* one-time code inside a URL, and the whole reason it is a picture is that
* nobody reads one aloud. So the page asks for the camera and gets the text
* back, which is the same direction the computer's own dialog describes.
*
* The answer is the code's own text, not a parsed result: what a pairing
* fragment means is the page's to decide, and it already knows how — the
* `#pair=` handling ships with the interface.
*/
@JavascriptInterface
fun scanPairingCode(): String = if (!trusted()) "" else scanWithCamera()
/**
* Load a URL in this window instead of handing it to the browser.
*
* The page asks for this once, right after a scan: the code it read points at
* the computer's own Studio, and the whole point of scanning is to end up
* looking at that computer. Going out to a browser would leave the app on one
* screen and the paired machine on another, with the back gesture an exit
* rather than a return.
*
* Only http and https are accepted. The page is this repository's own build
* and is trusted, but a scanner reads whatever is held in front of it, so the
* one value that arrives here from outside is checked: a `file://` or
* `javascript:` that happened to be on a printed code must not become a
* navigation.
*/
@JavascriptInterface
fun openInApp(url: String) {
if (!trusted()) return
val trimmed = url.trim()
if (!trimmed.startsWith("http://") && !trimmed.startsWith("https://")) return
main.post { onOpenInApp?.invoke(trimmed) }
}
/** How the activity loads a URL in the window it owns. Set by the activity. */
var onOpenInApp: ((String) -> Unit)? = null
// ---- the native browser view host bridge ----
//
// The page's `host.ts` looks for `window.reasonixHost` with Promise-based
// methods (showBrowserView, hideBrowserView, etc.). `@JavascriptInterface`
// is synchronous, so the round trip is: JS generates an id, calls
// `hostCall(id, method, args)`, Kotlin does the work asynchronously, then
// pushes the result back via `evaluateJavascript` calling
// `window.__reasonixHostReply(id, json)`. The injected script (in
// MainActivity) wires each Promise to that reply channel.
/** How the activity handles an async host call. Set by the activity. */
var onHostCall: ((id: String, method: String, args: String) -> Unit)? = null
/**
* Entry point for the page's `window.reasonixHost` Promise methods.
*
* `id` is a unique call identifier the JS side generated; `method` is the
* method name (e.g. "showBrowserView"); `args` is the JSON-encoded argument
* array. The activity handles the call asynchronously and replies through
* `evaluateJavascript` with `window.__reasonixHostReply(id, resultJson)`.
*/
@JavascriptInterface
fun hostCall(id: String, method: String, args: String) {
if (!trusted()) return
onHostCall?.invoke(id, method, args)
}
/**
* Whether the page has a panel up that the back gesture should close first.
*
* Written by the page, read by the activity's back handler. The page owns the
* answer — it is the only thing that knows what is open, and a second
* implementation of that question in the shell would be one that drifts — so
* the page reports the fact and the shell keeps the action.
*
* Volatile because the two ends are on different threads: the page sets it
* from the WebView's JavaScript thread, the back gesture reads it on the main
* one, and a stale read here is a panel that does not close.
*/
@Volatile
var layerOpen: Boolean = false
private set
@JavascriptInterface
fun setLayerOpen(open: Boolean) {
layerOpen = open
}
// ---- the picker, which is an Activity result in a synchronous world ----
private val pending = AtomicReference<Uri?>(null)
private val latch = AtomicReference<CountDownLatch?>(null)
/** The code the camera read, held between the activity's answer and the
* wait below — the same round trip the file picker takes. */
private val scanned = AtomicReference<String?>(null)
private val scanLatch = AtomicReference<CountDownLatch?>(null)
/** The main thread, because starting an Activity result must happen there. */
private val main = Handler(Looper.getMainLooper())
/**
* How to ask for a file. Set by the activity, which owns the Activity Result
* API; it answers whether the ask was dispatched, not whether the person
* picked anything — that arrives through `deliver`.
*/
var askForFile: () -> Boolean = { false }
/**
* How the activity reports the window's insets. Set by the activity, which is
* the only thing that can see them, and asked on every call rather than
* cached — a rotation, a fold, or the keyboard opening all change them, and a
* stale answer here is a stripe of interface under the status bar.
*/
var onInsets: (() -> Insets?)? = null
/**
* How to open the camera. Set by the activity, which owns the Activity
* Result API; like `askForFile` it answers whether the ask was dispatched,
* and the code itself arrives through `deliverScan`.
*/
var askForScan: () -> Boolean = { false }
/**
* How to put a file where the person picks. Set by the activity, which owns
* the Activity Result API; it answers whether the ask was dispatched, and
* the chosen place arrives through `deliverSave`.
*/
var askForSave: (name: String, content: String) -> Boolean = { _, _ -> false }
/**
* How to open the share sheet. Set by the activity, which is the only thing
* that can start an Intent; it answers whether anything on the device took
* the text.
*/
var askForShare: (name: String, content: String) -> Boolean = { _, _ -> false }
/** The place the save dialog answered with, held between the two halves. */
private val saveTarget = AtomicReference<Uri?>(null)
private val saveLatch = AtomicReference<CountDownLatch?>(null)
/**
* Ask where to put a file and write it there.
*
* The same latch round trip as `pickFile`, and for the same reason:
* `@JavascriptInterface` cannot suspend, so the WebView's thread waits while
* the picker is open — a modal panel the person is looking at.
*/
private fun saveFile(name: String, content: String): String? {
val gate = CountDownLatch(1)
saveLatch.set(gate)
saveTarget.set(null)
val posted = CountDownLatch(1)
var dispatched = false
main.post {
dispatched = runCatching { askForSave(name, content) }.getOrDefault(false)
posted.countDown()
}
val ready = runCatching { posted.await(PICK_TIMEOUT_SECONDS, TimeUnit.SECONDS) }.getOrDefault(false)
if (!ready || !dispatched) {
saveLatch.set(null)
return null
}
return try {
if (!gate.await(PICK_TIMEOUT_SECONDS, TimeUnit.SECONDS)) null
else saveTarget.get()?.toString()
} catch (_: InterruptedException) {
null
} finally {
saveLatch.set(null)
}
}
/** Called by the activity once the bytes are written. A null uri is a dismissal. */
fun deliverSave(uri: Uri?) {
saveTarget.set(uri)
saveLatch.get()?.countDown()
}
/**
* Read the pairing code a computer is showing, and return it.
*
* The same shape as `pickFile` and for the same reason: a
* `@JavascriptInterface` method is called on the WebView's thread and must
* return a value, so the camera is opened on the main thread and this one
* waits. The block lasts as long as the person is aiming, which is why the
* timeout is generous — a scan that gave up after a few seconds would be
* one nobody could complete.
*
* An empty string is "no code", which is what a refusal, a cancellation and
* a timeout all are. The caller acts on the text, so a distinct error would
* only be a fourth way to say the same thing.
*/
private fun scanWithCamera(): String {
val gate = CountDownLatch(1)
scanLatch.set(gate)
scanned.set(null)
val posted = CountDownLatch(1)
var dispatched = false
main.post {
dispatched = runCatching { askForScan() }.getOrDefault(false)
posted.countDown()
}
val ready = runCatching { posted.await(SCAN_TIMEOUT_SECONDS, TimeUnit.SECONDS) }.getOrDefault(false)
if (!ready || !dispatched) {
scanLatch.set(null)
return ""
}
return try {
if (!gate.await(SCAN_TIMEOUT_SECONDS, TimeUnit.SECONDS)) "" else scanned.get().orEmpty()
} catch (_: InterruptedException) {
""
} finally {
scanLatch.set(null)
}
}
/** Called by the activity when the scanner answers. Null is a dismissal. */
fun deliverScan(text: String?) {
scanned.set(text)
scanLatch.get()?.countDown()
}
/**
* Ask for one file and wait for the answer.
*
* `@JavascriptInterface` methods cannot suspend: the WebView calls them on
* its own thread and expects a value back. So the ask is posted to the main
* thread and this one waits on a latch. The cost is a blocked JS thread for
* as long as the picker is open — which is a modal panel the person is
* looking at, with nothing else in the page running.
*
* A timeout is set rather than waiting forever: a picker that never answers
* (the activity destroyed underneath it) would otherwise pin a WebView
* thread until the process died.
*/
private fun pickFile(): Uri? {
val gate = CountDownLatch(1)
latch.set(gate)
pending.set(null)
val posted = CountDownLatch(1)
var dispatched = false
main.post {
dispatched = runCatching { askForFile() }.getOrDefault(false)
posted.countDown()
}
val ready = runCatching { posted.await(PICK_TIMEOUT_SECONDS, TimeUnit.SECONDS) }.getOrDefault(false)
if (!ready || !dispatched) {
latch.set(null)
return null
}
return try {
if (!gate.await(PICK_TIMEOUT_SECONDS, TimeUnit.SECONDS)) null else pending.get()
} catch (_: InterruptedException) {
null
} finally {
latch.set(null)
}
}
/** Called by the activity when the picker answers. A null uri is a dismissal. */
fun deliver(uri: Uri?) {
pending.set(uri)
latch.get()?.countDown()
}
/** The folder under Downloads that this app's exports land in. */
private fun exportFolder(): String = "Reasonix"
/** The MIME type this name declares, or null when the platform does not know
* it.
*
* Where the type is written, the name and the type agree, and that agreement
* is the whole fix: `.md` is answered `text/markdown` here, so MediaStore
* has nothing to correct and the export stays `报告.md` instead of being
* renamed `报告.md.txt` by the `text/plain` it used to be given. Where the
* type is unknown the field is left off, which leaves MediaStore to derive
* the same `application/octet-stream` it would have been told — the name
* still stands on its own, which is all this is for. */
private fun mimeFor(name: String): String? {
val ext = name.substringAfterLast('.', "").lowercase()
if (ext.isEmpty() || ext == name.lowercase()) return null
return runCatching { MimeTypeMap.getSingleton().getMimeTypeFromExtension(ext) }.getOrNull()
}
private fun exportFile(name: String, content: String): String? = runCatching {
val bytes = content.toByteArray(Charsets.UTF_8)
if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.Q) {
val resolver = context.contentResolver
val values = ContentValues().apply {
put(MediaStore.Downloads.DISPLAY_NAME, name)
// The type is derived here rather than declared, and it is
// written only when the name really has an extension that maps
// to one.
//
// It used to be `text/plain` for everything, and MediaStore
// answers a name whose extension disagrees with the declared
// type by appending its own: exporting `报告.md` landed in
// Downloads as `报告.md.txt`, and any source file came out with
// the wrong second extension. Deriving the type from the name
// instead keeps the two in agreement — which is what stops the
// rename — and a name the platform has no type for is left off
// entirely rather than declared as `application/octet-stream`,
// so nothing is corrected there either.
mimeFor(name)?.let { put(MediaStore.Downloads.MIME_TYPE, it) }
put(
MediaStore.Downloads.RELATIVE_PATH,
Environment.DIRECTORY_DOWNLOADS + "/" + exportFolder(),
)
// Hidden until the bytes are all there, so a file manager that
// reads the folder mid-write cannot show half a document.
put(MediaStore.Downloads.IS_PENDING, 1)
}
val uri = resolver.insert(MediaStore.Downloads.EXTERNAL_CONTENT_URI, values)
?: return@runCatching null
resolver.openOutputStream(uri)?.use { it.write(bytes) } ?: return@runCatching null
values.clear()
values.put(MediaStore.Downloads.IS_PENDING, 0)
resolver.update(uri, values, null, null)
"${Environment.DIRECTORY_DOWNLOADS}/${exportFolder()}/$name"
} else {
val dir = File(
Environment.getExternalStoragePublicDirectory(Environment.DIRECTORY_DOWNLOADS),
exportFolder(),
)
if (!dir.exists() && !dir.mkdirs()) return@runCatching null
val target = File(dir, name)
target.writeBytes(bytes)
target.absolutePath
}
}.getOrNull()
private fun sanitise(name: String): String {
// A name from the page is used as a filename, so it is reduced to one:
// no separators, no traversal, and nothing a filesystem treats as
// special. The default is the name the field it came from suggests.
val trimmed = name.trim().substringAfterLast('/').substringAfterLast('\\')
if (trimmed.isEmpty() || trimmed == "." || trimmed == "..") return ""
return trimmed.filter { it.isLetterOrDigit() || it in "._-" }.take(64)
}
/**
* Whether a turn is running, so the screen can be kept awake for one.
*
* The page owns the answer — it is the only thing that knows if the kernel
* is working — and the shell owns the flag, because a window flag is not
* something a page can set. Nothing is asked at launch: a window that is
* merely open is one the person may be reading, and holding the screen on
* for that is holding it on for nothing.
*/
@JavascriptInterface
fun setRunning(running: Boolean) {
onRunning?.invoke(running)
}
/** How the activity applies that flag. Set by the activity. */
var onRunning: ((Boolean) -> Unit)? = null
/**
* Whether the app may read every file, which is what lets the kernel open a
* folder on this phone as a workspace.
*
* Asked rather than assumed: the permission is granted by hand in system
* settings, so the page has to be able to say "not yet" rather than fail on
* the first `stat`. Answered `true` below API 30, where ordinary read
* permission is enough and the concept does not exist.
*/
@JavascriptInterface
fun hasAllFilesAccess(): Boolean = filesAccess?.invoke() ?: true
var filesAccess: (() -> Boolean)? = null
/** Open the system page where that permission is granted. The only place
* Android offers it — there is no dialog for this one. */
@JavascriptInterface
fun requestAllFilesAccess() {
if (!trusted()) return
askForFilesAccess?.invoke()
}
var askForFilesAccess: (() -> Unit)? = null
/**
* Whether this app is exempt from the system's battery optimisation.
*
* A turn runs in a child process under a foreground service. Under doze the
* whole app group is frozen anyway, so the turn stops making progress while
* the screen is off — which is the state the person left it in on purpose.
* Unrestricted is the answer, and it is a question Android answers.
*/
@JavascriptInterface
fun batteryUnrestricted(): Boolean = if (!trusted()) false else (batteryFree?.invoke() ?: true)
var batteryFree: (() -> Boolean)? = null
/** Ask for that exemption. Unlike the storage one, this has a real dialog. */
@JavascriptInterface
fun requestBatteryUnrestricted() {
if (!trusted()) return
askForBattery?.invoke()
}
var askForBattery: (() -> Unit)? = null
/**
* Ask for a folder on this phone and answer with its real filesystem path.
*
* "" is a dismissal, the same answer `importIntoHome` gives — and, unlike
* that one, an empty string is a normal outcome here rather than a failure.
* The `startIn` hint is a path the kernel knows; the picker may ignore it,
* which is the picker's business.
*
* The round trip is the same latch as `pickFile`, for the same reason: a
* document-tree picker is another activity, and this call cannot suspend.
*/
@JavascriptInterface
fun pickFolder(startIn: String): String {
if (!trusted()) return ""
val gate = CountDownLatch(1)
folderLatch.set(gate)
folderPending.set(null)
val posted = CountDownLatch(1)
var dispatched = false
main.post {
dispatched = runCatching { askForFolder(startIn) }.getOrDefault(false)
posted.countDown()
}
val ready = runCatching { posted.await(PICK_TIMEOUT_SECONDS, TimeUnit.SECONDS) }.getOrDefault(false)
if (!ready || !dispatched) {
folderLatch.set(null)
return ""
}
return try {
if (!gate.await(PICK_TIMEOUT_SECONDS, TimeUnit.SECONDS)) "" else folderPending.get().orEmpty()
} catch (_: InterruptedException) {
""
} finally {
folderLatch.set(null)
}
}
var askForFolder: (String) -> Boolean = { false }
private val folderPending = AtomicReference<String?>(null)
private val folderLatch = AtomicReference<CountDownLatch?>(null)
/** Called by the activity with the folder's real path, or null on dismissal. */
fun deliverFolder(path: String?) {
folderPending.set(path ?: "")
folderLatch.get()?.countDown()
}
companion object {
const val BRIDGE_VERSION = "4"
private const val PICK_TIMEOUT_SECONDS = 5L * 60
/** Long, because this is however long someone takes to aim. A shorter
* one would turn "still lining it up" into a failed scan. */
private const val SCAN_TIMEOUT_SECONDS = 5L * 60
/** Only as long as the sheet takes to appear: the answer is "something
* took it", which is known the moment the Intent resolves. */
private const val SHARE_TIMEOUT_SECONDS = 15L
}
}
/**
* The window's insets, in CSS pixels.
*
* A named type rather than four numbers because the page reads them as one
* thing: half an inset is a layout that is wrong in one direction, and a caller
* that passed the wrong pair would be indistinguishable from a device with no
* cutout at all.
*/
data class Insets(val left: Int, val top: Int, val right: Int, val bottom: Int)
@@ -1,66 +0,0 @@
package io.reasonix.mobile.app
import androidx.camera.core.ImageProxy
import com.google.zxing.BarcodeFormat
import com.google.zxing.BinaryBitmap
import com.google.zxing.DecodeHintType
import com.google.zxing.MultiFormatReader
import com.google.zxing.PlanarYUVLuminanceSource
import com.google.zxing.common.HybridBinarizer
/**
* Reading a QR code out of one camera frame.
*
* ZXing core and nothing else: it is pure Java, so it runs identically on a
* phone with Google Play Services and on one whose HarmonyOS compatibility
* layer has none. A decoder that leaned on ML Kit would work on the machine it
* was written on and fail on the one that needs it.
*
* The frame arrives as YUV_420_888, whose Y plane is already the greyscale
* bitmap ZXing wants — no conversion, no colour maths, just the bytes. That is
* the whole reason a phone can decode 30 frames a second without warming up.
*/
object Qr {
private val reader = MultiFormatReader().apply {
setHints(
mapOf(
// Only the form the computer actually renders. Leaving the
// reader open to every 1D barcode makes each frame slower for
// a case that cannot occur, and a slower frame is a code that
// has already left the viewfinder.
DecodeHintType.POSSIBLE_FORMATS to listOf(BarcodeFormat.QR_CODE),
DecodeHintType.TRY_HARDER to true,
),
)
}
/** The code's text, or null when this frame holds none. */
fun decode(image: ImageProxy): String? {
val plane = image.planes.firstOrNull() ?: return null
val buffer = plane.buffer
val data = ByteArray(buffer.remaining())
buffer.get(data)
// `rowStride` is padded to a hardware alignment and is usually wider
// than the picture. Passing the width as the row length is what tells
// ZXing where each row actually ends; getting it wrong shears the image
// and every code reads as garbage.
val source = PlanarYUVLuminanceSource(
data,
plane.rowStride,
image.height,
0,
0,
image.width,
image.height,
false,
)
return runCatching {
reader.decodeWithState(BinaryBitmap(HybridBinarizer(source))).text
}.getOrNull().also {
// The reader keeps per-frame state between calls; resetting is what
// stops a partial match in one frame from corrupting the next.
reader.reset()
}
}
}
@@ -1,146 +0,0 @@
package io.reasonix.mobile.app
import android.Manifest
import android.content.Intent
import android.content.pm.PackageManager
import android.os.Bundle
import android.view.Gravity
import android.view.ViewGroup
import android.widget.FrameLayout
import android.widget.TextView
import androidx.activity.result.contract.ActivityResultContracts
import androidx.appcompat.app.AppCompatActivity
import androidx.camera.core.CameraSelector
import androidx.camera.core.ImageAnalysis
import androidx.camera.core.ImageProxy
import androidx.camera.core.Preview
import androidx.camera.lifecycle.ProcessCameraProvider
import androidx.camera.view.PreviewView
import androidx.core.content.ContextCompat
import java.util.concurrent.Executors
/**
* The camera, held open just long enough to read one pairing code.
*
* A phone that wants to drive a computer cannot type the code that computer
* shows: it is a base64 fragment inside a URL, and the whole point of showing it
* as a QR code is that nobody reads one aloud. So this is the way in — the same
* gesture the computer's own dialog asks for, from the other side.
*
* The decode is ZXing rather than ML Kit on purpose. ML Kit needs Google Play
* Services, and this app installs on phones whose HarmonyOS compatibility layer
* has none: a GMS-backed scanner would work on a developer's phone and fail on
* the device that needs it. ZXing core is pure Java, so the scanner behaves the
* same on both.
*
* It returns the first code it reads and finishes. A pairing code is
* single-use and short-lived, so "keep scanning after a hit" is not a feature —
* it is a way to spend a code that a later frame would have discarded.
*/
class ScanActivity : AppCompatActivity() {
private lateinit var preview: PreviewView
private lateinit var hint: TextView
private val executor = Executors.newSingleThreadExecutor()
/** Set once a code is found, so the frames in flight behind it do not each
* try to finish the activity. */
@Volatile
private var settled = false
private val requestCamera = registerForActivityResult(ActivityResultContracts.RequestPermission()) { granted ->
if (granted) start() else refuse()
}
override fun onCreate(savedInstanceState: Bundle?) {
super.onCreate(savedInstanceState)
val root = FrameLayout(this)
preview = PreviewView(this).apply {
layoutParams = FrameLayout.LayoutParams(
ViewGroup.LayoutParams.MATCH_PARENT,
ViewGroup.LayoutParams.MATCH_PARENT,
)
// The frames are analysed, not composed: filling the view and
// cropping is right here and would be wrong for a camera app.
scaleType = PreviewView.ScaleType.FILL_CENTER
}
hint = TextView(this).apply {
text = getString(R.string.scan_hint)
setTextColor(0xFFFFFFFF.toInt())
textSize = 14f
gravity = Gravity.CENTER
val pad = (16 * resources.displayMetrics.density).toInt()
setPadding(pad, pad, pad, pad)
// Readable over whatever the camera is pointing at, which is the
// one background a design system cannot choose.
setBackgroundColor(0x99000000.toInt())
layoutParams = FrameLayout.LayoutParams(
ViewGroup.LayoutParams.MATCH_PARENT,
ViewGroup.LayoutParams.WRAP_CONTENT,
Gravity.BOTTOM,
)
}
root.addView(preview)
root.addView(hint)
setContentView(root)
if (ContextCompat.checkSelfPermission(this, Manifest.permission.CAMERA) ==
PackageManager.PERMISSION_GRANTED
) {
start()
} else {
requestCamera.launch(Manifest.permission.CAMERA)
}
}
private fun refuse() {
// No dialog of our own: the permission prompt already said no, and a
// second window repeating it is one more tap to dismiss. The caller
// gets the same answer as a cancelled scan.
setResult(RESULT_CANCELED)
finish()
}
private fun start() {
val providerFuture = ProcessCameraProvider.getInstance(this)
providerFuture.addListener({
val provider = runCatching { providerFuture.get() }.getOrNull() ?: return@addListener refuse()
val previewUseCase = Preview.Builder().build().also { it.setSurfaceProvider(preview.surfaceProvider) }
val analysis = ImageAnalysis.Builder()
// Only the newest frame matters: a stale one describes where the
// camera was, and a code that has already left the viewfinder
// must not win over one that is in it now.
.setBackpressureStrategy(ImageAnalysis.STRATEGY_KEEP_ONLY_LATEST)
.build()
.also { it.setAnalyzer(executor, ::analyse) }
runCatching {
provider.unbindAll()
provider.bindToLifecycle(this, CameraSelector.DEFAULT_BACK_CAMERA, previewUseCase, analysis)
}.onFailure { refuse() }
}, ContextCompat.getMainExecutor(this))
}
private fun analyse(image: ImageProxy) {
try {
if (settled) return
val text = Qr.decode(image) ?: return
if (settled) return
settled = true
setResult(RESULT_OK, Intent().putExtra(EXTRA_TEXT, text))
// Finishing from the analysis thread is the whole reason `settled`
// exists; the activity teardown has to happen on the main one.
runOnUiThread { finish() }
} finally {
image.close()
}
}
override fun onDestroy() {
super.onDestroy()
executor.shutdown()
}
companion object {
const val EXTRA_TEXT = "text"
}
}
@@ -1,18 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- The mark, drawn rather than shipped as a bitmap: one R in the app's own ink
on the palette's ground, which scales to every launcher's mask without a
density bucket per size. The hole in the R is the glyph's own counter, cut
with evenOdd rather than painted over. -->
<vector xmlns:android="http://schemas.android.com/apk/res/android"
android:width="108dp"
android:height="108dp"
android:viewportWidth="108"
android:viewportHeight="108">
<path
android:fillColor="#F4F1EA"
android:fillType="evenOdd"
android:pathData="M39,30 L39,78 L47,78 L47,60 L54.5,60 L65,78 L74.5,78 L62.5,58 C68.6,55.8 72.6,50.6 72.6,43.6 C72.6,35.4 66.8,30 57.6,30 Z M47,37 L56.4,37 C61.2,37 64.4,39.4 64.4,43.8 C64.4,48.2 61.2,50.9 56.4,50.9 L47,50.9 Z" />
<path
android:fillColor="#A6680C"
android:pathData="M62,66 L74,66 L74,78 L62,78 Z" />
</vector>
@@ -1,14 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- The status-bar glyph, not the launcher icon: Android renders a notification
icon as a mask, so anything that is not a solid silhouette on transparency
comes out as a grey blob. -->
<vector xmlns:android="http://schemas.android.com/apk/res/android"
android:width="24dp"
android:height="24dp"
android:viewportWidth="24"
android:viewportHeight="24"
android:tint="#FFFFFF">
<path
android:fillColor="#FFFFFF"
android:pathData="M3,4h18v2H3zM3,9h18v2H3zM3,14h12v2H3zM3,19h12v2H3zM17.5,13.2l1.6,2.6 1.6,-2.6h2.3l-2.8,4.3V21h-2.2v-3.5L15.2,13.2z" />
</vector>
@@ -1,62 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!--
The screen between the tap and the page.
It exists for a reason that is not cosmetic: the kernel has to be unpacked and
started before there is a page to show, and on a cold start that is about a
second. A blank window for a second reads as a broken app; a line of text that
says what is happening does not.
The progress bar is indeterminate and driven by nothing: the WebView's own
progress is mirrored onto it while the page loads, and before that it simply
turns, which is the honest description of "we do not know how long".
-->
<LinearLayout xmlns:android="http://schemas.android.com/apk/res/android"
android:layout_width="match_parent"
android:layout_height="match_parent"
android:background="@color/page"
android:orientation="vertical"
android:gravity="center"
android:paddingStart="32dp"
android:paddingEnd="32dp">
<TextView
android:id="@+id/splash_wordmark"
android:layout_width="wrap_content"
android:layout_height="wrap_content"
android:text="@string/app_name"
android:textColor="#8A857B"
android:textSize="22sp"
android:letterSpacing="0.08"
android:fontFamily="sans-serif-medium" />
<ProgressBar
android:id="@+id/splash_bar"
style="?android:attr/progressBarStyleHorizontal"
android:layout_width="132dp"
android:layout_height="wrap_content"
android:layout_marginTop="26dp"
android:indeterminate="true" />
<TextView
android:id="@+id/splash_text"
android:layout_width="wrap_content"
android:layout_height="wrap_content"
android:layout_marginTop="18dp"
android:gravity="center"
android:text="@string/splash_connecting"
android:textColor="#8A857B"
android:textSize="13sp"
android:lineSpacingMultiplier="1.5" />
<Button
android:id="@+id/splash_retry"
style="?android:attr/borderlessButtonStyle"
android:layout_width="wrap_content"
android:layout_height="wrap_content"
android:layout_marginTop="14dp"
android:minHeight="44dp"
android:text="@string/splash_retry"
android:textColor="#A6680C"
android:visibility="gone" />
</LinearLayout>
@@ -1,13 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- The adaptive launcher icon.
minSdk is 26, which is the release that introduced adaptive icons, so this
is the only launcher asset the app needs: there is no device running it
that would ask for a bitmap fallback. -->
<adaptive-icon xmlns:android="http://schemas.android.com/apk/res/android">
<background android:drawable="@color/ic_launcher_background" />
<!-- The safe zone is the middle 66dp of 108dp; the glyph is drawn inside it
so no launcher's mask can clip it. -->
<foreground android:drawable="@drawable/ic_launcher_foreground" />
<monochrome android:drawable="@drawable/ic_launcher_foreground" />
</adaptive-icon>
@@ -1,26 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<resources>
<string name="app_name">Reasonix</string>
<string name="channel_kernel">Kernel running</string>
<string name="channel_kernel_desc">Shown while the Reasonix kernel runs in the background. Can be muted on its own in system settings.</string>
<string name="notify_running_title">Reasonix is running</string>
<string name="notify_running_body">A task is still going. Switch back to see it.</string>
<string name="notify_failed_title">Reasonix could not start</string>
<string name="err_kernel_unreachable">Could not reach the kernel. Reopen the app from your home screen.</string>
<string name="download_default_name">reasonix-download</string>
<string name="download_saved">Saved: %1$s</string>
<string name="download_failed">Save failed: %1$s</string>
<string name="download_failed_status">Save failed: the server answered %1$d</string>
<string name="splash_connecting">Starting the kernel…</string>
<string name="splash_retry">Retry</string>
<string name="scan_title">Scan the code on the computer</string>
<string name="scan_hint">Point at the QR code in the computer\'s device panel to connect to it.</string>
<string name="scan_button">Scan to connect</string>
<string name="scan_denied">No camera permission, so scanning is unavailable. Allow it in system settings.</string>
</resources>
@@ -1,7 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- The dark palette's page colour. `DayNight` on the theme picks this up when
the system is dark, which is what keeps the frame before the WebView from
flashing light at someone using a dark phone at night. -->
<resources>
<color name="page">#0E1014</color>
</resources>
@@ -1,9 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- The light palette's page colour, and only what the window needs before the
WebView paints. The interface's own colours are `tokens.css`'s and are the
page's business; these two exist so the first frame is the app's ground
rather than the platform default white. -->
<resources>
<color name="page">#F4F1EA</color>
<color name="boot">#07080C</color>
</resources>
@@ -1,6 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- The launcher ground: the interface's dark page colour. It is the same in
both themes because a launcher icon is not themed by the app. -->
<resources>
<color name="ic_launcher_background">#14161C</color>
</resources>
@@ -1,35 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- Every string the shell itself shows. The interface's own text belongs to
the page and lives in its i18n bundle; these are only the few lines that
exist before it, or when it cannot load. -->
<resources>
<string name="app_name">Reasonix</string>
<string name="channel_kernel">内核运行中</string>
<string name="channel_kernel_desc">Reasonix 内核在后台运行时显示,可在系统设置中单独关闭。</string>
<string name="notify_running_title">Reasonix 正在运行</string>
<string name="notify_running_body">任务继续进行,切回应用即可查看。</string>
<string name="notify_failed_title">Reasonix 无法启动</string>
<string name="err_kernel_unreachable">连接内核失败。请返回桌面重新打开应用。</string>
<string name="download_default_name">reasonix-download</string>
<string name="download_saved">已保存:%1$s</string>
<string name="download_failed">保存失败:%1$s</string>
<string name="download_failed_status">保存失败:服务器返回 %1$d</string>
<string name="splash_connecting">正在启动内核…</string>
<string name="splash_retry">重试</string>
<string name="wsperm_title">需要「所有文件访问」权限</string>
<string name="wsperm_body">要打开手机上的文件夹作为工作区,需要先允许本应用访问所有文件。点「去设置」,在列表里打开 Reasonix 的开关,然后返回这里。</string>
<string name="wsperm_go">去设置</string>
<string name="wsperm_later">以后再说</string>
<string name="scan_title">扫描电脑上的二维码</string>
<string name="scan_hint">对准电脑上「设备访问」里的二维码,识别后即可连接这台电脑。</string>
<string name="scan_button">扫码连接电脑</string>
<string name="scan_denied">没有相机权限,无法扫码。请在系统设置里允许。</string>
<string name="share_title">分享文件</string>
</resources>
@@ -1,28 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!-- The shell's own furniture: a window background matching the page's light
palette before the WebView paints, and a transparent system bar so the
page's own safe-area padding is what the notch meets. The page themes
itself; this exists only so the first frame is not a white flash. -->
<resources xmlns:tools="http://schemas.android.com/tools">
<style name="Theme.Reasonix" parent="Theme.AppCompat.DayNight.NoActionBar">
<item name="android:windowBackground">@color/page</item>
<item name="android:statusBarColor">@android:color/transparent</item>
<item name="android:navigationBarColor">@android:color/transparent</item>
<!-- The page draws its own edge-to-edge layout and pads itself with the
safe-area insets from mobile.css. Laying the window out edge to edge
while leaving the bars transparent is what makes those insets real
numbers instead of zero. -->
<item name="android:windowLayoutInDisplayCutoutMode" tools:targetApi="p">shortEdges</item>
<item name="android:enforceStatusBarContrast" tools:targetApi="q">false</item>
<item name="android:enforceNavigationBarContrast" tools:targetApi="q">false</item>
</style>
<!-- The camera. A black ground because the preview has not drawn yet and a
light window behind a dark viewfinder is a flash in the dark; the bars
stay transparent so the preview reaches the edges, which is what makes
a code near the top of the screen scannable. -->
<style name="Theme.Reasonix.Scan" parent="Theme.Reasonix">
<item name="android:windowBackground">@android:color/black</item>
</style>
</resources>
@@ -1,31 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!--
Backups are off, and this file says why rather than leaving it implicit.
A Reasonix home holds provider keys, an account token and every session
transcript. Android's auto-backup would copy all of it to the person's cloud
drive and, on a device-to-device transfer, to a new phone — which is a
credential leaving the device by a route nobody chose. `allowBackup="false"`
in the manifest turns that off; these two stanzas are the belt to its braces,
excluding everything explicitly so that a later change to `allowBackup` cannot
quietly start copying credentials.
What a person wants to move between phones, the kernel already moves: its own
export writes a file the person hands to the system's picker.
-->
<data-extraction-rules>
<cloud-backup>
<exclude domain="root" />
<exclude domain="file" />
<exclude domain="database" />
<exclude domain="sharedpref" />
<exclude domain="external" />
</cloud-backup>
<device-transfer>
<exclude domain="root" />
<exclude domain="file" />
<exclude domain="database" />
<exclude domain="sharedpref" />
<exclude domain="external" />
</device-transfer>
</data-extraction-rules>
@@ -1,27 +0,0 @@
<?xml version="1.0" encoding="utf-8"?>
<!--
Cleartext HTTP is permitted, and the platform has no way to permit less.
Two kinds of page load over plain HTTP: the kernel's own page on
`127.0.0.1:<port>`, and the same page served by a paired computer on whatever
address its router handed out. The port is chosen when the kernel starts and
the LAN address when a code is scanned, so neither can be written here — and a
`domain-config` matches literals and domains, never a subnet.
So the boundary is not this file. `MainActivity` decides which cleartext origin
may become the window: `shouldOverrideUrlLoading` sends every http(s) origin
that is not the kernel's or the one just paired out to the browser, and the
native bridge answers only those two. A plaintext page on the LAN therefore
cannot become the document holding the bridge, which is what the reviewer's
"deliberate path for LAN pairing" asks for.
HTTPS is untouched: nothing here downgrades a request, and `trust-anchors`
keeps the system roots for every TLS connection that is made.
-->
<network-security-config>
<base-config cleartextTrafficPermitted="true">
<trust-anchors>
<certificates src="system" />
</trust-anchors>
</base-config>
</network-security-config>
-15
View File
@@ -1,15 +0,0 @@
// The build itself lives in app/build.gradle; this file declares the plugin
// versions once so the app module can be read on its own.
//
// AGP 8.11.1 is the pairing for the Gradle 8.14 toolchain this machine has:
// AGP 8.7 and older refuse a Gradle that new, and AGP 8.13+ wants a Gradle this
// one does not have. Both numbers are pinned rather than ranged, because a
// build that silently moved either one would be a different build.
plugins {
id "com.android.application" version "8.11.1" apply false
id "org.jetbrains.kotlin.android" version "2.1.20" apply false
}
tasks.register("clean", Delete) {
delete rootProject.layout.buildDirectory
}
-15
View File
@@ -1,15 +0,0 @@
# AndroidX is the only copy of the support libraries this project uses; the
# non-transitive R class keeps the app's own resources from aliasing library
# ones, which is what makes `R.string.*` unambiguous here.
android.useAndroidX=true
android.nonTransitiveRClass=true
# The daemon is what keeps a second `gradle assembleDebug` from re-reading the
# whole SDK, and the heap is sized for the R8 pass over the WebView and the
# packaged page rather than for this handful of Kotlin files.
org.gradle.jvmargs=-Xmx3g -Dfile.encoding=UTF-8
org.gradle.daemon=true
org.gradle.parallel=true
org.gradle.caching=true
kotlin.code.style=official
File diff suppressed because it is too large. Load diff
Binary file not shown.
Loaded 100 of 12269 files, more files were not shown because too many files have changed in this diff. Show more