GeezerVison

Part of CODAGT-1341.

The chat composer's / menu gets a third group, "Organization skills", between Personal and Workspace.

• ChatMessageInput takes organizationId (from AgentChatInput) and fetches organizationSkills(organizationId) only once the trigger opens, with the personal skills' 60 second staleTime.
• Org skills go through resolveSkillTriggers, so a name shared with a personal or workspace skill is qualified in every source (/org/x), as chatd requires. Selecting one inserts its trigger text like a personal skill. Status and empty copy covers three sources.

Behavior change for personal skills: disabled personal and org skills are hidden from the menu and no longer shadow a built-in command, matching chatd, which drops them before computing aliases. The submit intercept in AgentChatPage/submitChatTurn counts enabled org skills too, so an org skill named compact makes /compact send as a message. A personal or org list that fails with no cached data stays unknown: triggers stay qualified, /clear and /compact stay hidden, and submitting one shows the pending toast instead of clearing or compacting.

Testing

• ChatMessageInput.test.tsx "slash menu skill sources": a bare org trigger chosen with the arrow keys, a personal and org collision qualified as /org/reviewer, disabled skills ignored when qualifying, a built-in hidden by an org skill, and built-ins hidden and triggers kept qualified when the org list fails. AgentChatPage.test.tsx: /clear and /compact run neither while the org list fails. AgentChatInput.test.tsx and submitChatTurn.test.ts cover the chat organization's skills and the /compact case; five new stories cover the groups and states.
• Red-green: dropping the org list from the intercept, an empty org list, no enabled filter, and ignoring org skills for commands each fail a test.
• pnpm lint, vitest under src/pages/AgentsPage and touched stories passed at the original tip.

Review record

• frontend-review passed; coder-frontend-review (up to 4 rounds per role) fixed SD1, SD3, CD, TS3 and DF5 (below).
• Declined: one shared group renderer in SkillsTriggerMenu (RO3), because scripts/check-compiler.mjs then reported unmemoized closures; filtering once for both consumers (RO3), because the intercept and the menu own different queries; splitting ChatMessageInput (RP6), already over the size guideline before about 20 more lines.
• Remote UAT round 6 through Coder Agents: PASS on the final stack tip afdd4a9f954 (chat); its low notes are in docs(docs/ai-coder/agents): document organization skills #30604.
• Test audit: the bare org trigger test now picks the skill with the arrow keys instead of a click, so it also fails when org skills drop out of the keyboard selection list or move ahead of personal skills.
• coder-frontend-review DF5 (blocking) made a failed list with no data count as empty for built-ins, and a second fix kept triggers qualified; Codex round 1 reversed the built-in half (41cc34b), so a transient error can't run /clear or /compact instead of a same-named skill.
• coder-frontend-review of the Codex fixes: fixed CD6, TS2, SD1 (566bfc0, e501a1d, c236bb4); declined TS6, as new spies match each story's spy style.

This stack adds organization skills to Coder Agents: admins publish a skill once per organization, share it with groups or members, and chats in that organization load it next to personal and workspace skills.
Stack, bottom to top: #30584, #30585, #30586, #30587, #30588, #30589, #30590, #30591, #30592, #30593, #30594, #30595, #30596, #30597, #30598, #30599, #30600, #30601, #30602, #30603, #30604.

Xum opened this PR on behalf of @ibetitsmike.