Revert "Migrate to shared claude-code-review.yml reusable workflow"

[jnasbyupgrade]

Reverts #50

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Advanced

Run ID: aec540f0-3f87-4241-bfdf-747955eb0c8d

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Claude finished @jnasbyupgrade's task in 4m 44s —— View job

---

Code review — PR #51 (revert of shared claude-code-review.yml workflow)

• Gather context (PR diff, CLAUDE.md, related PR Migrate to shared claude-code-review.yml reusable workflow #50)
• Determine request type: code review (/code-review:code-review ... --comment)
• Review .github/workflows/claude-code-review.yml across correctness/simplification/efficiency/reuse angles
• Post inline findings
• Post summary

This PR is a mechanical, GitHub-generated revert of a single file, .github/workflows/claude-code-review.yml, undoing the migration to a shared reusable workflow (commit 6b2257e) and restoring the prior hand-maintained, per-repo copy. No SQL/extension code is touched, so this repo's schema-independence testing conventions (CLAUDE.md) don't apply.

I reviewed the restored ~170-line file (trigger/concurrency config, the author-gate if:, the CI-gate polling script, checkout step, and the claude-code-action invocation). Two findings survived verification and were posted as inline comments:

1. .github/workflows/claude-code-review.yml:84-92 — gh api --paginate --jq '[...]' applies the jq filter per page rather than merging pages first. Once sibling check-runs exceed GitHub's 30-per-page default, $json becomes multiple concatenated JSON arrays, so jq 'length' emits multiple numbers into $total/pending/bad, breaking the -eq integer comparisons and corrupting the cost-gate decision. ci.yml currently produces ~19-20 sibling checks (under the boundary), but its comments anticipate adding a PG 18 leg to two 6-leg matrices, which would cross it.
2. .github/workflows/claude-code-review.yml:27 (altitude) — the revert reintroduces the per-repo copy-paste maintenance burden that the original migration's commit message cited as the reason to centralize this workflow in Postgres-Extensions/ai; any future fix to the shared workflow now has to be manually re-applied here.

Both findings were confirmed and posted inline via mcp__github_inline_comment__create_inline_comment.

No code changes were made (review only, as requested).

[github-actions]

gh api --paginate --jq concatenates per-page output instead of merging arrays, breaking the numeric checks once sibling check-runs span more than one page.

gh api --paginate --jq '[...]' applies the --jq filter to each page separately and prints the results one after another — it does not merge them into a single JSON array first. GitHub's check-runs endpoint defaults to 30 items per page.

This repo's ci.yml alone already produces on the order of ~19-20 sibling check-runs (lint, release-safety, changes, the 6-leg test matrix, the 3-leg pg-upgrade-test matrix, the 6-leg pg-tle-test matrix, all-checks-passed), plus whatever claude.yml contributes. That's under 30 today, but it's a boundary the script never guards: the very next PG-version addition to both the test and pg-tle-test matrices (both comments in ci.yml note PG 18 support is anticipated) pushes the sibling-check count past 30.

Once that happens, $json becomes two concatenated JSON array literals (e.g. [...][...]), and jq 'length' <<<"$json" emits multiple newline-separated numbers instead of one. total=$(...) then holds a multi-line string, so [ "$total" -eq 0 ] fails with "integer expression expected", silently takes the false branch, and the same problem recurs for pending/bad — corrupting the gate decision (e.g. permanently deciding skip, disabling the paid review with no visible error) instead of counting all pages.

Fix: slurp all pages into one array, e.g. gh api ... --paginate --slurp --jq '[.[].check_runs[]|select(...)] ' (or post-process with jq -s 'add'), so pagination doesn't change the shape of $json.

[github-actions]

Reverting to this hand-maintained copy reintroduces the exact maintenance-drift risk the migration it undoes was meant to eliminate.

The commit this PR reverts (Migrate to shared claude-code-review.yml reusable workflow) explicitly called out the cost of the per-repo copy this restores: "a future fix lands here without a manual copy" (i.e., without the migration, any subsequent fix to Postgres-Extensions/ai's shared workflow — bug fixes, the labeled/claude-debug on-demand trigger, security hardening — has to be re-applied by hand to every consuming repo, including this one, or it silently drifts out of sync). If this revert is meant to be permanent rather than a temporary rollback while investigating a regression in the shared workflow, it's worth tracking that the fix should eventually be pushed back to ai/ and this repo re-migrated, rather than leaving the fork in place indefinitely.