You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The Nightly DMG failed four nights running (2026-09-16 → 09-19). CI Health Triage diagnosed it correctly every night — the #11 digest named the exact cause (release-macos.sh requires Xcode 27, nightly-dmg.yml pinned 26) and the exact fix. Nobody saw it: the digest is comment N on a rolling 74-comment issue, and a red scheduled run notifies no one. It was found on 09-19 by a session that happened to run gh run list. Fixed in #389.
Detection worked. Delivery did not.
Proposals (pick one or more)
Escalate on a streak. When the same workflow has been red ≥ 2 consecutive scheduled runs, the triage opens (or re-opens) a dedicated issue titled for that workflow, assigned to Kev — instead of only appending to 🩺 CI Health #11. One issue per streak; closed by the first green run.
Widen the conclusion filter. The prompt inspects only conclusion "failure". A job whose runner label no longer resolves (now a real risk: ci(nightly): build the DMG on GitHub's Xcode 27 image — the nightly has failed since 09-16 #389 moved the nightly to GitHub's previewxcode-27 label) never starts, sits queued, and ends cancelled — outside the filter. Include cancelled and timed_out for scheduled workflows, and flag "no Nightly DMG run in the last 26 h" as its own signal.
A second channel.pulse-weekly (Monday digest → M1K3 remember) could read gh run list --workflow "Nightly DMG" and pin a red streak to the top, like the COPPA verdict.
ci.yml:59-61,68 says Xcode 27 "is not on the runner image yet". True of macos-26, but xcode-27 exists now — reword so nobody waits on macos-26 to catch up. (Comment-only; ci.yml's pins themselves are fine: CI does not invoke the release script.)
When a macos-27 image goes GA, move nightly-dmg.yml off the preview label.
Watch the nightly's wall clock against timeout-minutes: 60 for the first week on the new image.
What happened
The Nightly DMG failed four nights running (2026-09-16 → 09-19). CI Health Triage diagnosed it correctly every night — the #11 digest named the exact cause (
release-macos.shrequires Xcode 27,nightly-dmg.ymlpinned 26) and the exact fix. Nobody saw it: the digest is comment N on a rolling 74-comment issue, and a red scheduled run notifies no one. It was found on 09-19 by a session that happened to rungh run list. Fixed in #389.Detection worked. Delivery did not.
Proposals (pick one or more)
conclusion "failure". A job whose runner label no longer resolves (now a real risk: ci(nightly): build the DMG on GitHub's Xcode 27 image — the nightly has failed since 09-16 #389 moved the nightly to GitHub's previewxcode-27label) never starts, sits queued, and endscancelled— outside the filter. Includecancelledandtimed_outfor scheduled workflows, and flag "no Nightly DMG run in the last 26 h" as its own signal.pulse-weekly(Monday digest → M1K3remember) could readgh run list --workflow "Nightly DMG"and pin a red streak to the top, like the COPPA verdict.Also carried from #389's review
ci.yml:59-61,68says Xcode 27 "is not on the runner image yet". True ofmacos-26, butxcode-27exists now — reword so nobody waits onmacos-26to catch up. (Comment-only; ci.yml's pins themselves are fine: CI does not invoke the release script.)macos-27image goes GA, movenightly-dmg.ymloff the preview label.timeout-minutes: 60for the first week on the new image.