Drain artifact watcher before closing WS on end_session_message

[jackcbrown89]

Summary

• On end_session_message, ws.close() was called immediately and the 3s chokidar drain + pending-request wait ran in the close handler. Any artifact_upload_request_message produced by a late chokidar event during the drain was silently dropped because send() is gated on readyState === OPEN.
• Moved the drain into a memoized drain() helper invoked before ws.close() on the orderly path. The close handler still calls drain() for unexpected disconnects; memoization prevents double-draining.

Test plan

• npx vitest run src/session.test.ts — all 29 tests pass

---

Note

Medium Risk
Changes session shutdown ordering and introduces memoized draining, which can affect artifact upload completeness and session close timing if edge cases exist (e.g., multiple close paths or long-running uploads). Scope is contained to WebSocketSession teardown logic plus a patch version bump.

Overview
Ensures end_session_message drains the artifact watcher and waits for pending artifact uploads before calling ws.close(), so late chokidar events can still enqueue artifact_upload_request_message while the socket is open.

Refactors the shutdown sequence into a memoized drain() helper that is invoked from both the orderly end-session path and the WS close handler (for unexpected disconnects), preventing duplicate drain work.

Bumps runtimeuse version from 0.11.0 to 0.11.1.

^{Reviewed by Cursor Bugbot for commit 79977a7. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 79977a7. Configure here.}

[cursor]

Drain failure prevents ws.close() and caches rejected promise

Low Severity

If drain() rejects (e.g., stopWatching() / chokidar watcher.close() throws), ws.close() on the next line is never reached, leaving the WebSocket open. Worse, the memoized drainPromise caches the rejection permanently, so when the close handler later calls await this.drain(), it re-throws — resolve() is never called and run() hangs forever. Wrapping ws.close() in a finally (or catching drain errors) would ensure the socket is always closed. In the old code, ws.close() was called unconditionally before any drain logic.

Additional Locations (1)

• packages/runtimeuse/src/session.ts#L202-L221

 

^{Reviewed by Cursor Bugbot for commit 79977a7. Configure here.}