Starter kits

pr-babysitter — the PR-status polling loop

Polls every open PR on a fast schedule and flags what needs a human: merge conflicts, red CI, stale no-review PRs, threads awaiting a response. It…

Polls every open PR on a fast schedule and flags what needs a human: merge conflicts, red CI, stale no-review PRs, threads awaiting a response. It never comments, nudges, or merges — a wrong beat costs a re-read, not an unrecallable action on a shared system.

Category: B · PR & review · Heartbeat: schedule (every 10m, business hours) · Level: L1 (report-only) · Cost: High

Quickstart

1. Read LOOP.md — the six parts, the prompt, the limits, the three stops.
2. pr-babysitter-state.md is your spine — already committed with an empty
   per-PR status table.
3. Claude Code: point a Routine or `claude -p` cron (every 10m) at
   .claude/skills/loop-task/SKILL.md. OpenCode: rename opencode.json.example →
   opencode.json, add the AGENTS.md line, wire skills/loop-task.md into a cron
   + wrapper script.
4. Rehearse once, by hand, before the schedule ever fires:
   "Run the pr-babysitter skill now, exactly as the schedule would."
5. Grade the rehearsal with the loop-verifier agent (.claude/agents/loop-verifier.md).
   PASS → arm the schedule. FAIL → fix the skill, never the report.

What's in this kit

File	Role
`LOOP.md`	the definition — six parts, prompt, limits, ownership, three stops
`pr-babysitter-state.md`	the spine — per-PR last-seen status, so a beat only reports what changed
`loop-budget.md`	caps: 60 runs/day, 350k tokens/day, 80% tripwire
`loop-constraints.md`	the constitution — report-only, no comments/merges
`loop-run-log.md`	append-only, one line per beat
`.claude/skills/loop-task/SKILL.md`	the procedure (Claude Code)
`.claude/agents/loop-verifier.md`	the read-only reviewer that grades `pr-status-report.md`'s format
`opencode.json.example` · `skills/loop-task.md`	the OpenCode twin — same skill text, cron + wrapper heartbeat

Loop Ready checklist

Provable success condition — pr-status-report.md exists, ranked, template-matched, every open PR considered
Run limit — 60/day (10-minute cadence, business hours)
Spine written first, committed empty
Report-only (L1) start — writes only pr-status-report.md + spine + log
Human gate placed — you read the report; nothing acts until you do
One log line per beat, no silent runs
Kill switch — loop-pause-all in loop-budget.md

How this relates to the course

This kit is the scheduled twin of the "PR-review loop" case study in Part 2 · Step 7 — Event-Driven Loops: that page's check-yourself question is literally about a PR-review loop that silently skipped PRs during a burst, because events are dropped, not queued. The fix the page prescribes — a slow scheduled reconciliation sweep asking "any open PRs without a review from me?" — is this loop's entire job, run standalone instead of paired with an event-driven pr-reviewer. The heartbeat, level, and cost are set by the Part V §15B catalog entry.

Source: Panaversity's Loop Engineering: A Crash Course (S1). Full attribution: resources/sources.md.