●  Agent loops \u00b7 reusable playbooks \u00b7 copy-paste prompts
EST. 20241 makerBuilt in public
Playbooks \u00b7 5 loops

Loops that
finish the job.

Short, repeatable routines for working with coding agents \u2014 each loop is a tight cycle with copy-paste prompts, a clear stop rule, and notes on what usually goes wrong. Written from lab practice: small batches, quoted evidence, no open-ended \u201ckeep going\u201d. Pick a loop, run it top to bottom, stop when the exit criteria say so.

No installs, no backend \u2014 just prompts you hand to the agent you already use, plus guardrails that keep each cycle honest.

\u00a7 01

The playbooks

5 loopscopy \u00b7 paste \u00b7 run \u00b7 stop
Fix loop \u00b7 6 steps 001 Open \u2197
Build Until Green
Run the build, fix the first error, repeat until the build is green.
6 steps \u00b7 10–30 min typical \u2192
Test loop \u00b7 6 steps 002 Open \u2197
Coverage Until Threshold
Add small, focused tests until the coverage target is met — then stop.
6 steps \u00b7 20–45 min typical \u2192
Quality loop \u00b7 6 steps 003 Open \u2197
A11y Audit Until Clean
Run automated accessibility checks on changed routes, fix, and repeat until clean.
6 steps \u00b7 15–40 min typical \u2192
Ops loop \u00b7 6 steps 004 Open \u2197
CI Failure Watcher
Poll CI, read the red checks first, push small fixes until the pipeline is green.
6 steps \u00b7 15–60 min typical \u2192
Strategy loop \u00b7 7 steps 005 Open \u2197
Plan Then Execute
Write a short structured plan first, then execute it step by step with checkpoints.
7 steps \u00b7 30 min–2 hr typical \u2192
\u00a7 02

How to run one

same shape every time

Every loop follows the same contract: one prompt per turn, quoted evidence back, stop at the exit criteria. Open a loop, copy the first prompt box into your agent, paste the result where asked, and move to the next step only when the \u201cwhy\u201d line is satisfied. If a step fails twice, use the failure-mode guardrail instead of pushing through \u2014 that is what the guardrails are for.