Why a new issue: #6140's fix landed as #6156, but the symptom stands. Every run on #6140 resumes its landed pull and cancels itself, so this is the follow-up.
The evidence (steward, 2026-10-26):
21 to 22 approved forge pulls have been conflicted for over 3 hours (approved="27 approved: conflict=21 …").
The crew is idle, and no bring-up-to-date run has started.
None were handed to a person (0 hand-over notes in 12 h).
Every one truly conflicts with main (a merge-tree in the pod), and all their heads are in the base repo.
Commenting on the pulls doesn't help: it only re-runs reviews.
Done when (one small pull):
src/crew/finishing.ts logs one line per approved, conflicted pull it does NOT bring up to date, with the not(…) reason: finishing: skip pull=N why="…". Once per pull per reason change, not every tick.
The autonomy tick line gets finishing="queued Q, in flight F, skipped S: <reason>×k, …".
A test where a conflicted approved pull is skipped for each reason and the log names it.
That's all. Once it lands, the steward reads the reasons and files the real fix.
Why a new issue: #6140's fix landed as #6156, but the symptom stands. Every run on #6140 resumes its landed pull and cancels itself, so this is the follow-up.
The evidence (steward, 2026-10-26):
approved="27 approved: conflict=21 …").Done when (one small pull):
src/crew/finishing.tslogs one line per approved, conflicted pull it does NOT bring up to date, with thenot(…)reason:finishing: skip pull=N why="…". Once per pull per reason change, not every tick.finishing="queued Q, in flight F, skipped S: <reason>×k, …".That's all. Once it lands, the steward reads the reasons and files the real fix.