[steve8708]

What changed

When an inbox sweep sees an ordinary queued message while the current run still has in-flight work, release the receipt for the completion-triggered drain instead of putting it behind a 10-second retry window. Add a regression for that active-tool path while keeping the existing admission-response retry case.

Verification

• Queued submission inbox spec: 30/30 passed.
• pnpm guards: all 93 passed.
• PostgreSQL fencing integration suite: 27/27 passed against a task-local PostgreSQL 17 instance.
• The matching beta Chat deploy job and smoke succeeded for 320ba2b.
• The authenticated queued-message E2E on that deployed head failed with background_worker_failed after the direct API returned a durable queued receipt: run 40210677278. This result predates this change and remains under investigation. The busy 409 path remains; this PR makes no claim of a 409 improvement.

[github-actions]

Here's a visual recap of what changed:

Open the full interactive recap

[builder-io-integration]

Builder reviewed your changes and has a few items to flag 🟡

Review Details

Incremental review summary

The latest update moves providerDispatchStarted until after approve() returns, and adds regression coverage for a synchronous local rejection. That fixes the synchronous ACP-not-pending case. The existing approval-dispatch comment remains open: an asynchronous approve() rejection before the provider accepts the approval can still be classified as an unknown provider outcome. I did not repost it because the same issue is already covered by the open thread.

I found no other confirmed new issues in the reviewed update. The continuation/inbox-ordering issue from the prior round was already resolved and is not part of the current open comments. No files were changed.

🧪 Browser testing: Skipped — infra unavailable. pnpm install setup and the restart attempt fail on toolkit TypeScript TS2883 errors.