diff --git a/.claude/skills/open-pr/SKILL.md b/.claude/skills/open-pr/SKILL.md index f1f4ef5fc2..281add364e 100644 --- a/.claude/skills/open-pr/SKILL.md +++ b/.claude/skills/open-pr/SKILL.md @@ -183,10 +183,19 @@ When such paths merged, watch until the build succeeds — a red Cloud Build means production did NOT update even though the PR was green: ```bash -gcloud builds list --limit 3 -gcloud builds log # on failure +gcloud builds list --region=europe-west4 --limit 3 \ + --format='table(id,status,createTime,substitutions.TRIGGER_NAME,substitutions.SHORT_SHA)' +gcloud builds log --region=europe-west4 # on failure ``` +The `--region` flag is load-bearing: the `deploy-app` / `deploy-api` +triggers are **regional** (europe-west4), and the global `gcloud builds +list` answers with a handful of months-old global builds — all +`SUCCESS`, none from today — which reads like "nothing was triggered" +while both deploys are already done (2026-08-28, #10808 follow-through: +a 20-minute poll on the global list never saw the builds). Match the +`SHORT_SHA` column against the merge commit before trusting a row. + ## Gotchas - **`isOutdated` ≠ `isResolved`.** A fix-push can outdate a Copilot diff --git a/CHANGELOG.md b/CHANGELOG.md index 517a8c4267..0e8380333b 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -211,6 +211,11 @@ aggregate instead: an italic *Catalog* line at the end of the version section an ### Changed +- **`open-pr` skill: the post-merge Cloud Build watch names the region** — `gcloud builds + list` without `--region=europe-west4` answers with months-old global builds (all `SUCCESS`, + none from today), which reads like "nothing was triggered" while both regional deploys are + already done; the #10808 follow-through polled that list for 20 minutes. The command now + carries the region and a `SHORT_SHA` column to match against the merge commit (#10809). - **`babysit-pipeline`'s gap-backfill guidance corrected against a completed run** — the section shipped in #10628 was written mid-backfill and three of its claims did not survive the rest of it. The halt-on-cluster threshold counted raw failed runs, but one