A library of agent loops you can run.
Each loop is a reusable operating pattern — trigger, steps, model, tools, checks, budget, governance. Fork one into a durable runtime with a registry, telemetry, revision history, and a visible evidence label on every number that isn't wired to real data yet.
Data QA Triage
Classify defective rows, verify JSON shape, and surface drift before it reaches production.
- trigger
- batch upload or webhook
- runtime
- 2-6 min
- governance
- propose only
- agent
- Claude
CI Failure Triage
Collect logs, classify the likely root cause, propose a minimal patch plan, and open an approval item.
- trigger
- GitHub Actions failure webhook
- runtime
- 3-8 min
- governance
- human approval
- agent
- Claude
Release Prep
Turn merged work into a release candidate with changelog, smoke checks, backups, and rollback notes.
- trigger
- manual or tag candidate
- runtime
- 5-12 min
- governance
- human approval
- agent
- Claude
Incident Follow-up
Convert an incident into a timeline, root cause draft, action items, and recurrence checks.
- trigger
- incident closed
- runtime
- 6-15 min
- governance
- propose only
- agent
- Claude
Memory Proposal Review
Let agents propose memory updates without letting them silently rewrite the floor.
- trigger
- session end or operator request
- runtime
- 1-4 min
- governance
- propose only
- agent
- Claude
Code Review QA
Review a diff against a frozen checklist, grade each finding by severity, and post a structured review — no rubber stamps.
- trigger
- pull request opened
- runtime
- 3-9 min
- governance
- propose only
- agent
- Claude
Kill Flaky Tests
Find tests that pass and fail on the same commit, quarantine them, and propose a deflake or a fix.
- trigger
- CI run or schedule
- runtime
- 4-12 min
- governance
- propose only
- agent
- Claude
PR Babysitter
Watch an open PR: keep it rebased, re-run checks, answer review nits, and nudge it toward merge.
- trigger
- pull request opened
- runtime
- ongoing
- governance
- human approval
- agent
- Claude
Ship a PR Until Green
Drive a branch from red to all-green: fix failing checks one at a time, under a budget and a stop gate.
- trigger
- manual or failing PR
- runtime
- 5-20 min
- governance
- human approval
- agent
- Claude
Reach Coverage Target
Add tests toward a coverage target where it matters — uncovered branches first, no assertion-free padding.
- trigger
- manual or schedule
- runtime
- 6-18 min
- governance
- propose only
- agent
- Claude
De-Sloppify Pass
Sweep a diff or module for AI slop — dead code, vague names, redundant comments — and tighten it.
- trigger
- manual or pre-merge
- runtime
- 3-10 min
- governance
- propose only
- agent
- Claude
Migrate an API
Move call sites from an old API to a new one across the repo, in reviewable batches with a rollback note.
- trigger
- manual
- runtime
- 8-25 min
- governance
- human approval
- agent
- Claude
Morning Issue Triage
Start the day with new issues sorted: label, dedupe, draft a first response, and flag the urgent ones.
- trigger
- daily schedule
- runtime
- 4-10 min
- governance
- propose only
- agent
- Claude
Docs Drift Sentinel
Watch code changes for docs that fall out of date, and open a doc-fix proposal when they drift.
- trigger
- on merge or schedule
- runtime
- 3-9 min
- governance
- human approval
- agent
- Claude
Several loop patterns adapted from Awesome Agent Loops (CC BY 4.0). Step definitions and copy are original.