Fix Codex Agent Chat Stop interrupt request

[teamleaderleo]

Fix Agent Chat Codex Stop so it interrupts the active turn with the protocol-required thread and turn IDs.

The adapter now waits for turn/started during the startup window, guards the late ID against completed or later generations, and surfaces interrupt request failures in the chat. The regression test covers the complete request shape, missing-ID handling, and generation guard.

Changelog

Fixed: Codex Agent Chat Stop now interrupts the active turn reliably.

Validation

• bun test/codex-stop.test.ts
• bun test/codex-route.test.ts
• bun test test/catalog.test.ts
• git diff --check
• Full bun run check could not start because this fresh worktree has no installed bun type definitions or React dependencies; fleet CI should provide the full check.

Closes #15779. Related terminal Stop UI work: #15234.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

---

Summary by cubic

Fixes Codex Agent Chat Stop so it interrupts the active turn with the protocol-required thread and turn IDs.

• Stop now waits for turn/started during startup and sends the interrupt with both IDs instead of sending an incomplete request.
• A late turn ID is discarded if the generation completed or a later one started.
• Interrupt request failures are surfaced in the chat instead of being silently swallowed.
• Adds regression tests that drive codexAdapter.stop() against a fake server, covering the request shape, missing-ID handling, and generation guard.

Closes #15779.

^{Written for commit 1f1c718. Summary will update on new commits.}

[coderabbitai]

Warning

Review limit reached

Next included review available in 9 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used all 10 included reviews currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: Repository: manaflow-ai/cmux/.coderabbit.yaml

Review profile: ASSERTIVE

Plan: Advanced

Run ID: dc7bb215-2655-426b-80f9-6c3718f4f26a

📥 Commits

Reviewing files that changed from the base of the PR and between 990efc1 and 1f1c718.

📒 Files selected for processing (2)

• agent-chat/adapters/codex.ts
• agent-chat/test/codex-stop.test.ts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[github-actions]

Dogfood build of 1f1c718e0d0b2b90cff08cd5eb913f287bc7a16d

cmux DEV pr-15837-1f1c718e.app

The link opens this exact commit in the cmux dev menu bar app; the page waits until the build is ready. Builds run only while this PR has the dev-build label. Under load the fleet builds the newest push each time a worker frees up, so some pushes are skipped. It signs in against production, so Cloud or backend changes still need a tagged build with a development backend.

Covers ab741838..1f1c718e (commits: 1) since the previous link, cmux DEV pr-15837-ab741838.app; if that push was skipped, its page names the newer build. To build a commit in between: cmux-ci build cmux --ref <sha> --tag bisect-<sha8> --workspace https://github.com/manaflow-ai/cmux/pull/15837.

[teamleaderleo]

Review subagent on the exact diff: the fix itself is correct. TurnInterruptParams in the codex app-server protocol (codex-rs/app-server-protocol/src/protocol/v2/turn.rs upstream) carries a required threadId and turnId, so the old request("turn/interrupt", { threadId }) could only ever be rejected, and the old .catch(() => {}) swallowed the rejection, which is why Stop looked like it did nothing. Three findings on the test, all about the test rather than the fix.

Review:

1. The new test could not fail on the bug it was fixed for. It only called the two pure helpers and never invoked codexAdapter.stop, so it passed with the pre-fix body restored. That is a source-shape test, not a behavior test.
2. The test file never runs in CI.
3. console.log("codex stop assertions passed") and export {} sat ahead of the last two assertions, so the file could report success before finishing.

Fixed (1f1c718):

• 1: the test now injects a fake app server and asserts on the params that actually reach it. Five cases: known turn id sends {threadId, turnId}, Stop pressed before turn/started sends nothing until the id arrives and then sends it complete, a late turn id does not interrupt a newer generation on the same thread, no thread or no active turn sends nothing, and a rejected interrupt surfaces as an error event instead of being swallowed. codexSetSharedServerForTest is the seam, because stop() reads the module-level shared connection directly.
  Proof it now pins the bug: with the pre-fix stop() body restored the test fails with Interrupt must carry both protocol IDs: {"threadId":"thread-1"}, and passes again with the fix in place.
• 3: all assertions now run before the console.log.

Left:

• 2 does not hold, so nothing changed for it. .github/workflows/ci-guards.yml:340 runs bun run test in agent-chat on the preflight group, and agent-chat/run-tests.ts discovers tests with new Glob("*.test.ts").scanSync(...) rather than a list of names, so this file was already picked up. Local full-suite run on this head: 21 scripts and 8 bun:test files ran, 0 fail, and bun x tsc --noEmit clean.

Auto-merge on squash; it lands when CI is green.

— Raindrop g2 🫧
Run: run_worker_20260930_3fc64ba6

[github-actions]

Merge receipt for 1f1c718e0d: every check was green at merge (13 verified; 15 skipped by policy). Full suite runs on main after merge.