GeezerVison

CODAGT-1135

CODAGT-1213

Remote dogfood UAT, round 1: PASS on fa60a1576782 (develop.sh with --experiments chat-automations --chat-max-queued-messages-per-chat 2, real model through the AI gateway). The later commits fix the minor UAT findings and the frontend self-review findings; a confirmatory round will run on the final head.

List of org A, with every column (org B showed only its own automations):

Missing target after archiving the target chat:

Run now, busy chat with when_busy=skip (409):

Run now, automation share of the queue used up (429):

View chats (chat-list filter): a chat the automation created, and chats it wrote to:

Experiment turned off for one user by a runtime rule: no nav entry, and the page is not found:

@codex review

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

@codex review

CI note: the ci pull_request workflow did not start for head 43fa1a5f23 (GitHub reports mergeable: UNKNOWN and no merge ref). This matches the known GitHub issue affecting pushes to existing stacked PRs in this repo since about 01:00Z (#30160, #30166 and #30167 show the same), which the stack orchestrator has escalated. No extra commits will be pushed to re-trigger it.

The previous head 8f2ee30d8f has a fully green ci run on record. 43fa1a5f23 adds the round 1 review fixes and a merge of the base branch, and passed the full local pre-push gate (make pre-push: Go tests, site build and tests).

@codex review

CI note: the ci pull_request workflow again did not start for head a307cbc190 (Codex round 2 fixes; mergeable: UNKNOWN, no merge ref), the same known GitHub issue as on 43fa1a5f23. No commits are pushed to re-trigger it. a307cbc190 passed the full local pre-push gate (make pre-push), and 8f2ee30d8f remains the last head with a fully green ci run.

@codex review

CI note: the ci pull_request workflow again did not start for head b37e9d99d0 (Codex round 3 fixes; mergeable: UNKNOWN), the same known GitHub issue as the previous two heads. No commits are pushed to re-trigger it. b37e9d99d0 passed the full local pre-push gate, and 8f2ee30d8f remains the last head with a fully green ci run.

Remote dogfood UAT, round 2: PASS (minor issues only) on the current head b37e9d99d0 (develop.sh test server on a Coder Agents workspace, two organizations plus an empty one, a real model through the AI gateway). The runner downloaded and hash-checked all 45 attachments and reviewed every screenshot. It repeated every round 1 criterion and verified each change since round 1:

• the nil UUID filter returns 400;
• a malformed id is ignored when the experiment is off;
• deleted messages no longer match;
• 30 chats page as 25 plus Load more;
• non-owner rows get no Run now or View chats and the switch cannot re-enable;
• Run now is disabled for a missing target and while a run is pending;
• buttons have per-row accessible names;
• the mobile link opens a full-width page;
• the next run advances without a reload;
• a failed refetch keeps the table.

Minor findings are tracked in CODAGT-1246.

Org A list:

Run now 409 (busy, skip), 202 (queued), 429 (queue share), 202 (idle):

Experiment off for one user by a runtime rule:

Org admin viewing another owner's automations:

A pending Run now disables every Run now:

Mobile (390x844):

Next run advances with the page open:

Blocked refetch keeps the table:

Readiness record for head b37e9d99d0a23b41771a2e7f137706f436ecf6be.

• Decision: ready with tracked follow-ups. The PR stays a draft for the stack owner.
• Independent advisory pass (a separate reviewer given the goal, diff, findings and test results): ready with tracked follow-ups. No deferred item is a security, data-loss or core-behavior blocker.
• Codex review: 4 rounds with 8, 4, 4 and 6 P2 findings. All threads are resolved. There is no clean verdict on this head.
  • Rounds 1 to 3 were fixed.
  • The round 3 model-availability item and all of round 4 were deferred: CODAGT-1244 and CODAGT-1246 (Linear, under CODAGT-1135).
  • Stopped because findings were not converging and none was a blocker. The advisor agreed, and the stack orchestrator accepted stopping at round 4.
• UAT: round 2 PASS on this exact head.
• Tests: local make pre-push passed on this head, and ci run 36703512580 (https://github.com/coder/coder/actions/runs/36703512580), started with workflow_dispatch, is green on this exact head.
• Pre-merge condition: met for this head. A later head (for example after a base merge) needs a new green ci run.

CI on the current head: ci run 36703512580 (https://github.com/coder/coder/actions/runs/36703512580), started with gh workflow run ci --ref tk/codagt-1213-automations-list-page because pushes to existing stacked PRs do not start the pull_request ci run right now. The run's headSha is b37e9d99d0a23b41771a2e7f137706f436ecf6be, and it completed with success: all 22 jobs succeeded or were skipped, including required.

Fixed the Codex P2 on automation chat history: the chat search query accepts archived:any (archived and active chats), and the automation chats dialog requests it. Go tests cover parsing and the API returning archived automation chats; the Vitest request assertions fail without the frontend change.

• Head: 356447f0d810854e675c36361e7fd982996611f0
• CI run (pull_request): https://github.com/coder/coder/actions/runs/36976442095

Generated with xum • Model: anthropic:claude-opus-5-5 • Thinking: high

@codex review

Documentation Check

Updates Needed

• docs/ai-coder/agents/chat-search-syntax.md (open) - Add any to the archived filter row and note it matches archived and active chats, while omitting the term still hides archived chats. This was addressed.

Automated review via Coder Agents

Codex Review: Didn't find any major issues. Can't wait for the next one!

Reviewed commit: 356447f0d8

Docs preview

Check off each page once it's been reviewed. If a page changes in a later push, its checkbox clears automatically so it gets a fresh look. Pages not yet wired into the docs navigation aren't listed here.

• docs/ai-coder/agents/chat-search-syntax.md

Review fixes: the automation chats dialog returns focus to the View chats button on close (with a test); AutomationsList calls useDashboard itself; AUTOMATIONS_PATH and useAutomationsEnabled moved to automationsFlag.ts; the dialog reuses ChatTitleLink; chat-search-syntax.md documents archived:any. Merged #30167.

• Head: 4b2a0afffccf40f6a1d2015a2a97780d9c7a7b56
• CI run (pull_request): https://github.com/coder/coder/actions/runs/36981473079

Generated with xum • Model: anthropic:claude-opus-5-5 • Thinking: high

Review follow-ups in head 4b2a0af:

• Documentation Check: 5f9cdb3 adds any to the archived row of docs/ai-coder/agents/chat-search-syntax.md: it matches archived and active chats, and omitting the term still hides archived chats.
• Folded nits from Mike's review, all in 1b6d9ff (doc wording tidied in fe4cf27):
  • AutomationsList calls useDashboard() itself; AutomationsListProps and the threaded props are gone.
  • AUTOMATIONS_PATH and useAutomationsEnabled moved to components/Automations/automationsFlag.ts, mirroring exp/chatBoard/chatBoardFlag.ts; every importer up the stack now imports from there.
  • The chats dialog reuses ChatTitleLink, now an exported link-only component; AutomationRow keeps the loading and error handling in a private ChatTitle wrapper.
  • (isAnyRunPending was already renamed earlier.)
• Focus restore: see the reply on the thread.

Generated with xum • Model: anthropic:claude-opus-5-5 • Thinking: high

@codex review