Skip to content

Skills S2: / Skill Invocation in Group Channels #4233

Open

Description

@DoodleBears

Parent: #4231
Blocked by: #4232

What to build

把 S1 扩展到群聊(Group Channel)。

  • skill 目录覆盖频道内所有 PersonaBot 的 Memory Repository + Host Skill;每行显示所属 bot 的头像;同一 Host Skill 只列一次。
  • 排序加入"bot 名命中"一档:名称前缀 > 名称子序列 > bot 名命中(输入 alice 能筛出 Alice 的所有 skill)> 描述子串。
  • Skill Invocation 不隐式 @ 任何 bot:选 A 的 skill、@b,只有 B 按正常路由收到;未 @ 时按频道现有路由。
  • 命令与 skill 重名:裸名归命令,skill 行仍可选。
  • 外部 IM(Discord/Lark)镜像中,skill chip 降级为纯文本 /name。

Acceptance criteria

  • AC-1:A、B 两个 bot 的群里,/ 列出两者的 Project Skill,头像区分正确。
  • AC-2:选 A 的 skill 并 @b,只有 B 被唤醒,B 收到的快照正文与 owner bot:a 一致。
  • AC-3:输入 bot 名能筛出该 bot 的 skill;描述命中排在最后。
  • AC-4:外部 IM 镜像消息显示 /name(按 IM browser QA 流程验证)。
  • AC-5:自动化覆盖路由独立性与排序;PR 附群聊截图。

Activity

  1. added
    ready-for-agentFully specified, ready for an AFK agent
    feature:channelChannel chat, timeline, composer, and navigation
    scope:tracerHuman-testable implementation slice
    on Nov 15, 2026
  2. added this to the v1.8.0 milestone on Nov 15, 2026
  3. DoodleBears commented on Nov 15, 2026

    @DoodleBears
    ContributorAuthor

    Agent work claim

    • Agent: Claude Code (UI, review) + Codex (logic)
    • Task: claude-code/local/0d4a6a9f-6296-49dd-b14c-6789efa657c3 (not a DSH Session)
    • Issue scope: S2 — / Skill Invocation in Group Channels
    • Branch: claude/skills-s2-group (stacked on feat: invoke a Bot's Skills from / in a Human DM #4252)
    • Started: 2026-11-15 15:00 UTC
    • State: active
  4. added
    agent:activeOne or more coding-agent tasks currently claim this issue; see claim comments for task IDs
    in-progressHeld by an agent session; see Claim comment for who
    on Nov 15, 2026
  5. DoodleBears commented on Nov 15, 2026

    @DoodleBears
    ContributorAuthor

    Agent work claim

    • Agent: Codex (implementation sub-agent; non-UI scope coordinated with the existing UI claim)
    • Task: codex/local/01a1e0df-afbe-7283-bd17-0b534bc78f98 (not a DSH Session)
    • Issue scope: S2 owner-name ranking and CLI parity, Group routing independence tests, external IM mirror tests, feature map and change fragment
    • Branch: t3/f86c3869 (based on S1 PR feat: invoke a Bot's Skills from / in a Human DM #4252)
    • Started: 2026-11-15 20:22 UTC
    • State: active
  6. DoodleBears commented on Nov 15, 2026

    @DoodleBears
    ContributorAuthor

    Agent work handoff

    • Agent: Codex (non-UI implementation sub-agent)
    • Task: codex/local/01a1e0df-afbe-7283-bd17-0b534bc78f98 (not a DSH Session)
    • State: handed-off
    • Branch: t3/f86c3869, based on S1 PR feat: invoke a Bot's Skills from / in a Human DM #4252
    • Commit: 299fa885a1937a001c469fce5f2dfa8cf1b42797 (local only; not pushed; no PR created)
    • Delivered: optional owner-name ranking callback; CLI display-name/slug ranking via the existing list RPC; Group routing and snapshot regressions; fixed missing snapshots in ordinary Group digests and pending context; Discord/Lark plain provider-payload tests; feature-map and bilingual change fragment.
    • Client changes: none. The UI must pass owning Bot display name and slug to the optional third rankSkillEntries argument, with no names for Host entries.
    • Validation: lint and changed-file format checks passed; focused 65 tests passed; CI-plan affected suite 322 files / 4,040 tests passed (5 files / 16 tests skipped); slides build and docs public-surfaces test passed. Heavy jobs ran serially, Vitest at most 2 workers.
    • Typecheck: pnpm -w exec tsc --noEmit reports one error in unchanged S1 packages/client/test/skill-composer.test.ts:214: shortcut choices widen id to string instead of ComposerShortcutId. Left to the UI scope.
    • Remaining integration gates: Group UI/avatars, real Host/CLI, Discord/Lark browser rendering and Human acceptance. The IM tests prove provider payloads from the stored Human body, not automatic Discord/Lark Group forwarding (the existing automatic web-DM forwarding path is Weixin-only).
    • The parent UI claim remains active; agent:active is retained.
  7. DoodleBears commented on Nov 15, 2026

    @DoodleBears
    ContributorAuthor

    Agent work handoff

    • Agent: Claude Code (UI, integration, review) + Codex GPT-6.1 Sol via T3 delegate (logic)
    • Task: claude-code/local/b7955cb7-65c2-4b41-bafd-b18256e4a38b takes over from claude-code/local/0d4a6a9f-6296-49dd-b14c-6789efa657c3
    • Branch: claude/skills-s2-group (stacked on feat: invoke a Bot's Skills from / in a Human DM #4252); the earlier uncommitted Group catalog UI is kept as 760127f
    • State: active
  8. DoodleBears commented on Nov 15, 2026

    @DoodleBears
    ContributorAuthor

    Agent work claim

    • Agent: Codex (implementation sub-agent; review fixes delegated by the existing S2 owner)
    • Task: codex/local/01a1e19d-c3c4-7c61-bb66-3ba990d3da10 (not a DSH Session)
    • Issue scope: oversized Skill snapshots in Group digest/pending context and missing Skill snapshot in Group mention steer; regression tests and requested checks
    • Branch: s2-fixes (local only, based on claude/skills-s2-group)
    • Started: 2026-11-15 23:50 UTC
    • State: active
  9. DoodleBears commented on Nov 16, 2026

    @DoodleBears
    ContributorAuthor

    Agent work handoff

    • Agent: Codex (implementation sub-agent)
    • Task: codex/local/01a1e19d-c3c4-7c61-bb66-3ba990d3da10 (not a DSH Session)
    • State: handed-off
    • Branch: s2-fixes, based on claude/skills-s2-group at 1a859dddeede79ef972d89edf11aa7a1c3651345
    • Local commit: 7fa63d6c9c8261737f68c8137ce8262650fac24d (fix(skills): unblock Group delivery and preserve steer snapshots); not pushed, no PR created.
    • Delivered: shared bounded Skill excerpts in ordinary Group digest/context, visible complete-snapshot recovery instructions through channel_read, first-row progress with small remaining budgets, and full Human Skill snapshots in active Group mention steering. Normal mention delivery retains its full snapshot.
    • Regression evidence: final tests fail against the old runtime in both pending-context cases and steering; both old digest cases stall past a bounded 15-second external deadline. Fixed runtime passes all 36 focused tests, including complete snapshot recovery and exactly-once rendered/handled admission checks at 24,000- and 2,000-character budgets.
    • Verification: pnpm lint, pnpm -w exec tsc --noEmit, changed-file formatting, and diff whitespace checks passed. CI-plan selected 752 affected files against origin/main: 747 passed / 5 skipped; 8,094 tests passed / 23 skipped, in three sequential batches with at most 2 workers. Slides build and the docs public-surface test passed. Every Vitest/TypeScript run waited for other active runs first.
    • Files: packages/core/src/runtime/bot-runtime.ts, packages/core/test/skill-invocation.test.ts, changes/4233-group-skill-delivery-fixes.md.
    • Cleanup: no real Host/device launched; stopped test processes exited, task-owned temporary test profiles and generated docs artifacts were removed; worktree clean. Integration/cherry-pick stays with the parent S2 owner. Other active issue claims and agent:active are retained.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    agent:activeOne or more coding-agent tasks currently claim this issue; see claim comments for task IDsfeature:channelChannel chat, timeline, composer, and navigationin-progressHeld by an agent session; see Claim comment for whoready-for-agentFully specified, ready for an AFK agentscope:tracerHuman-testable implementation slice

    Type

    No type

    Projects

    No projects

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions