Why did the queue stop, and how do I read the reason?
The queue stopped by itself
The queue stops on its own and says why. It does not retry forever, and it does not stop in silence.
The reasons
- Nothing left to admit. Every remaining task overlaps something, or there are no tasks. This is the ordinary end of a run.
- The project folder is gone — renamed, moved, or on a disk that unmounted.
- Three tasks failed in a row. This is the guard. Three consecutive failures almost always means something environmental rather than three unrelated coding problems: a test command that does not run, a missing dependency, an expired credential. Running the rest of the queue into it would spend real tokens producing identical blocks.
Reading it
The reason is on the queue screen, in words, where the Run queue button was. The tasks keep their own states: what closed stayed closed, what blocked kept its output.
After three failures
Do not restart the queue first. Open the blocked tasks and read the attached output — three blocks with the same error is the environment telling you something. Run that acceptance command yourself, in the project folder, in a shell pane. Fix what it shows, then start the queue again.
Pausing versus stopping
Pause stops new fronts from opening and leaves the running ones alone: they finish, their commands run, their verdicts land. There is no button that kills a working agent mid-edit, because that leaves the repository in a state nobody asked for.