[kyle-sexton]

No related issue: regression from #8562 reported by the GitHub Support hub

Summary

maintenance-reconcile-lost-runs has failed every scheduled run since
#8562 merged. The plan step lists pr-verify runs for the last 24 hours
in one page and throws when total_count is over 100. pr-verify now has
more than 100 runs a day. The plan counted the failure but still
printed "nothing lost", and the act step exited 1 without saying why.

Fix

• The lane run listing pages with the existing paginate helper, up to
  LANE_RUN_PAGES (5 pages, 500 runs). Past the cap it still throws and
  counts as a failure.
• The summary says how many listings failed, not "nothing lost", when
  failed > 0.
• The act step prints one ::error title=reconcile:: line when it exits
  1, with the plan's listing failures and its own failed re-runs.
• A changelog fragment for pr-pipeline (patch).

Verification

• Failing run:
  https://github.com/melodic-software/claude-code-plugins/actions/runs/39456940340
• New tests: a 150-run listing over 2 pages finds a lost run on page 2;
  a 501-run listing is a failure and never "nothing lost"; act prints
  the reason when the plan's failed > 0. All three fail on
  origin/main (29 pass, 3 fail) and pass here (32 of 32).
• scripts/check-affected.sh --base origin/main --tests: 10 passed,
  4 skipped.

Related

• feat(pr-pipeline): trust a reconciler-recorded re-run of a lost pr-verify run #8562 introduced the lane listing.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-20T16:32:41.038032Z	db7a569	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

claude-security-review reviewed db7a569a196adde378eb721e3a69b4da4aea3ae3: 0 finding(s), 0 as inline comments.

No security issues found in PR #8685 (head db7a569).

The PR touches only the lost-run reconciler script, its test and a changelog fragment, and reads only GitHub API data. I checked these points:

• Pagination: the new code reuses the shared paginate helper, which throws "has more than N pages" when the cap is hit. A day with more than 500 pr-verify runs is therefore still treated as a failure and never as a partial read. No re-run is issued from incomplete data, and the failure is now reported as "N listings failed; what was lost is unknown" instead of "nothing lost".
• Page shape: a response that is not a page of runs throws, so the listing fails closed.
• Run filtering: the lane's runs are still filtered by exact r.path === lanePath, so no other workflow's runs are pulled in.
• Input handling: the query parameters are still built with encodeURIComponent and a fixed timestamp format, and no untrusted text goes into a request path or shell.
• Logging: the new ::error line interpolates only integer counts, so it cannot inject workflow commands.
• Gate behavior: the act step's exit code is unchanged (1 on any failure, 0 otherwise). Only a diagnostic message is added.

The diff changes no instruction surface, so I did not apply the instruction-surface lens or consult the exception register.

[github-actions]

claude-review reviewed db7a569a196adde378eb721e3a69b4da4aea3ae3: 0 finding(s), 0 as inline comments.

No findings. I read the diff, the PR context, and paginate in .github/actions/check-trusted-trigger/check-trusted-trigger.mjs. I did not run the tests, so the PR's "32 of 32 pass" is author-claimed and unverified.

• Pagination: reconcile.mjs now reads the lane run listing through the existing paginate helper, capped at LANE_RUN_PAGES (5 pages, 500 runs). The wrapper checks that each page has a workflow_runs array. A malformed page or a listing past the cap throws inside the per-lane try and is counted as a failure, so a partial read is never used. The other listing in plan still has its old single-page check.
• Report wording: "nothing lost" is now said only when no listing failed. When one did, the message is "N listings failed; what was lost is unknown".
• Act step: it now prints an ::error title=reconcile:: line, with the plan's listing failures and its own failed re-runs, before it exits 1.
• Tests: the three new tests cover a lost run on page 2, a listing past the cap, and the act step's reason line. The fake GitHub handles the page= parameter. They cover the changed behaviour.
• Closing issue: the PR links none (No related issue: …), so there are no acceptance criteria to check.
• Path-scoped rules: no rule file under .claude/rules/ matches the changed files, apart from the changelog-fragment convention, which the PR follows.

One edge case is too minor to block. A listing of exactly 500 runs fills page 5 and paginate then throws "has more than 5 pages". It errs toward failing, never toward a wrong "nothing lost", and the 501-run test sits past it.