Public: anyone can read this; the crew can change it.

Sentinel: The lane has stopped: nothing has merged and nothing said so #14298

Merged opened this

Check · Review · Land · Live done — Landed and live in production.

Merged

This issue stayed open after the lane moved, and that part is a code defect. Pulls merged at 05:25 and 20:28 on 2026-11-17, and the sentinel's close step (closeLaneMoving → closeCaughtUp) should have closed the issue at its next hunt.

Notes

shipped production is serving it deploy 02af203

opened this

This issue stayed open after the lane moved, and that part is a code defect. Pulls merged at 05:25 and 20:28 on 2026-11-17, and the sentinel's close step (closeLaneMoving → closeCaughtUp) should have closed the issue at its next hunt. The close skips any issue a person has commented on, and personSpoke counted the forge's own notes as that. Those notes are posted under the owner's account: "Live triage (automatic, not written by a person): …" and "Parked by the crew: …". The live-triage note came an hour after the issue was filed, so from then on the sentinel could never close it. The two parks then made it look like a person was holding the issue open.

src/crew/person_spoke.ts now leaves out comments that start with the park lead (PARK_LIKE) and comments that carry the automatic mark " (automatic, not written by a person):", which live triage and first look both write. A comment the owner actually writes still holds the issue for them. This affects every Sentinel close that uses personSpoke: lane-stopped, main-ahead and the shared-cause closes in src/crew/sentinel.ts. docs/features/lane-stopped.md says so.

Not done: the alarm itself still fires on a 90-minute merge gap while pulls are open. That threshold is working as designed, so I left it alone. The issue may fire again whenever merges pause, but from now on it closes at the next hunt after the lane moves, unless a person has written on it.

Tests: test/person-spoke.test.ts checks that a park note, a live-triage note and a first-look note under the owner's name do not count, and that the owner's own words do. It fails without the fix. The sentinel and lane-alarm suites pass (86 tests), and the gate and bun x tsc --noEmit are clean.

To try it: after this lands, the next sentinel hunt with a merge less than 90 minutes old and no failure line on five pulls closes this issue with "The lane moved: …".

🤖 Generated with Claude Code

Closes #14263

commented

This issue stayed open after the lane moved, and that part is a code defect. Pulls merged at 05:25 and 20:28 on 2026-11-17, and the sentinel's close step (closeLaneMoving → closeCaughtUp) should have closed the issue at its next hunt. The close skips any issue a person has commented on, and personSpoke counted the forge's own notes as that. Those notes are posted under the owner's account: "Live triage (automatic, not written by a person): …" and "Parked by the crew: …". The live-triage note came an hour after the issue was filed, so from then on the sentinel could never close it. The two parks then made it look like a person was holding the issue open.

src/crew/person_spoke.ts now leaves out comments that start with the park lead (PARK_LIKE) and comments that carry the automatic mark " (automatic, not written by a person):", which live triage and first look both write. A comment the owner actually writes still holds the issue for them. This affects every Sentinel close that uses personSpoke: lane-stopped, main-ahead and the shared-cause closes in src/crew/sentinel.ts. docs/features/lane-stopped.md says so.

Not done: the alarm itself still fires on a 90-minute merge gap while pulls are open. That threshold is working as designed, so I left it alone. The issue may fire again whenever merges pause, but from now on it closes at the next hunt after the lane moves, unless a person has written on it.

Tests: test/person-spoke.test.ts checks that a park note, a live-triage note and a first-look note under the owner's name do not count, and that the owner's own words do. It fails without the fix. The sentinel and lane-alarm suites pass (86 tests), and the gate and bun x tsc --noEmit are clean.

To try it: after this lands, the next sentinel hunt with a merge less than 90 minutes old and no failure line on five pulls closes this issue with "The lane moved: …".

🤖 Generated with Claude Code

Approved

Fixes the defect: park and automatic notes under the owner's name no longer count as a person speaking. The new test would fail without it and the doc line is in the right feature file. Follow-up, not blocking: the filter matches on text alone. A person who quotes an automatic note, or starts a comment with the park lead, is ignored. Other park queries pair the lead with BY_CREW (src/core/done_on_main.ts). Anchoring the exclusion to the lead position (LIKE for the automatic lead too) would be tighter than instr anywhere in the body.

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs