Babysitting PRs

Monitor open pull requests, respond to reviewer comments, fix CI failures, and keep PRs moving toward merge — without human intervention.

Workflow

runPipeline(
  listOpenPRs
    .iterate()
    .map(
      loop((recur, done) =>
        checkPRStatus.then(classifyStatus).branch({
          CIFailing: diagnoseCIFailure.then(applyFix).then(pushAndWait).then(recur),
          ReviewComments: addressComments.then(pushAndWait).then(recur),
          Approved: done,
          Stale: rebase.then(pushAndWait).then(recur),
        }),
      ),
    )
    .collect(),
);

Stages

1. List open PRs — query GitHub for PRs opened by the bot or assigned to it.
2. Check status — for each PR, check CI status, review comments, and staleness.
3. Classify — route to the appropriate handler:
   • CI failing: diagnose the failure from logs, apply a fix, push, wait for CI to re-run.
   • Review comments: address each comment, push the changes.
   • Approved: exit the loop — the PR is ready.
   • Stale: rebase onto the latest base branch and re-check.
4. Loop — after each action, re-check the PR status and loop until approved.

Key points

• Each PR is babysit independently and concurrently via .iterate().map().
• The CI diagnosis agent only sees the failure logs — it doesn't know about review comments or the PR's history.
• loop + branch creates a state machine that handles the full lifecycle of a PR.
• Consider adding withTimeout around the outer loop to abandon PRs that can't be fixed within a time budget.