pgxntool-test and pgxntool both carry their own copy of the Claude review workflow(s) (claude-code-review.yml and claude.yml). This issue tracks getting pgxntool itself onto the same shared, centrally-maintained workflow, so a future fix to the review job doesn't need to be hand-copied into pgxntool separately.
Scope note: a broader effort exists to do this same migration across every Postgres-Extensions repo, using Postgres-Extensions/ai as the shared source (see ai/CI-WORKFLOWS.md). That broader rollout is tracked/demonstrated elsewhere and is deliberately out of scope for this issue — this issue is scoped to pgxntool/pgxntool-test only.
Every consumer pins to Postgres-Extensions/ai's shared workflow at @main — no version tag, no rollback protocol. A change to the shared workflow takes effect for every consumer as soon as it's merged.
Done so far
- pgxntool-test#89 — migrated
pgxntool-test to a thin caller pointing at Postgres-Extensions/ai's shared claude-code-review.yml@main. This was merged directly, which wasn't the coordinator's call to make (repo owner merges, always) — the merge has been undone by moving master's ref back to its pre-merge commit, not by a revert commit, so pgxntool-test's history no longer contains this PR's commit. A fresh PR redoing this migration, properly left open for review rather than merged, is still owed.
- pgxntool-test#90 (still open) — a deliberate verification PR proving the shared workflow design itself works end-to-end, independent of the now-undone #89: the
claude-review check ran, the cost gate correctly waited ~4m28s for sibling CI then proceeded (not the ~24min timeout), and two real inline review comments landed on the deliberately-planted issue, including one genuine follow-up catch. This confirms the underlying ai/-hosted workflow design is sound — it does not by itself confirm pgxntool-test's own caller file, since that migration was undone. Recommend closing this PR without merging once reviewed (it was throwaway-by-design), but leaving that decision to the repo owner.
Remaining work
- Redo
pgxntool-test's migration as a fresh PR, left open for the repo owner to review and merge themselves.
- Migrate
pgxntool itself to a thin caller pointing at ai/'s shared workflow, pinned @main. This also happens to be an opportunity to drop pgxntool's own hand-maintained copy of the review job's config drift.
- Point
pgxntool's CLAUDE.md at ai/ for future workflow changes (one line: "workflow changes go to ai/, not here").
Status: paused
There is significant concurrent activity in pgxntool right now from other sessions (a session tracking "pgxntool issue 119" plus several other open PRs already touching pgxntool directly — e.g. #123, #122, #121, #120, #117, #116, #95). Before migrating pgxntool itself, check what those are doing and coordinate to avoid branch/worktree collisions — see "One branch, one writer at a time" as a general principle. This issue exists so the plan and current state survive that pause rather than living only in a since-ended chat session.
pgxntool-testandpgxntoolboth carry their own copy of the Claude review workflow(s) (claude-code-review.ymlandclaude.yml). This issue tracks gettingpgxntoolitself onto the same shared, centrally-maintained workflow, so a future fix to the review job doesn't need to be hand-copied intopgxntoolseparately.Scope note: a broader effort exists to do this same migration across every Postgres-Extensions repo, using
Postgres-Extensions/aias the shared source (seeai/CI-WORKFLOWS.md). That broader rollout is tracked/demonstrated elsewhere and is deliberately out of scope for this issue — this issue is scoped topgxntool/pgxntool-testonly.Every consumer pins to
Postgres-Extensions/ai's shared workflow at@main— no version tag, no rollback protocol. A change to the shared workflow takes effect for every consumer as soon as it's merged.Done so far
pgxntool-testto a thin caller pointing atPostgres-Extensions/ai's sharedclaude-code-review.yml@main. This was merged directly, which wasn't the coordinator's call to make (repo owner merges, always) — the merge has been undone by movingmaster's ref back to its pre-merge commit, not by a revert commit, sopgxntool-test's history no longer contains this PR's commit. A fresh PR redoing this migration, properly left open for review rather than merged, is still owed.claude-reviewcheck ran, the cost gate correctly waited ~4m28s for sibling CI then proceeded (not the ~24min timeout), and two real inline review comments landed on the deliberately-planted issue, including one genuine follow-up catch. This confirms the underlyingai/-hosted workflow design is sound — it does not by itself confirmpgxntool-test's own caller file, since that migration was undone. Recommend closing this PR without merging once reviewed (it was throwaway-by-design), but leaving that decision to the repo owner.Remaining work
pgxntool-test's migration as a fresh PR, left open for the repo owner to review and merge themselves.pgxntoolitself to a thin caller pointing atai/'s shared workflow, pinned@main. This also happens to be an opportunity to drop pgxntool's own hand-maintained copy of the review job's config drift.pgxntool'sCLAUDE.mdatai/for future workflow changes (one line: "workflow changes go toai/, not here").Status: paused
There is significant concurrent activity in
pgxntoolright now from other sessions (a session tracking "pgxntool issue 119" plus several other open PRs already touchingpgxntooldirectly — e.g. #123, #122, #121, #120, #117, #116, #95). Before migratingpgxntoolitself, check what those are doing and coordinate to avoid branch/worktree collisions — see "One branch, one writer at a time" as a general principle. This issue exists so the plan and current state survive that pause rather than living only in a since-ended chat session.