A safety net for the coding agent. In rare cases, the infrastructure orchestrating a fix run could drop a wake-up signal — the agent would finish its work, but the run would sit in "running" forever and the resulting pull request never got written back to the ticket.
A background sweeper now checks every 15 minutes for runs that have gone quiet. A run that has stalled briefly is nudged awake and resumes its normal path; a run that is truly dead is finalized directly — its result is recovered and written back, or it's cleanly marked as failed so the ticket isn't blocked. Waking a healthy run is harmless, so there's no risk to runs that are simply taking their time.
The guarantee this gives you: no fix run ever stays permanently undetermined. Worst case, a dropped signal costs a few minutes of latency instead of a stuck ticket. Live for all workspaces — nothing to do on your end.