Run Any Loop Template with continuous-claude
Short answer
Every loop template here maps directly to a continuous-claude run. The template's objective, validation, and boundaries become the task prompt; the suggested max iterations become --max-runs; you add --max-cost and --max-duration caps, then run.
Why it matters
Templates describe a safe loop on paper; continuous-claude (MIT) executes one. Mapping the two keeps the template's boundaries and stop rule while a real runtime enforces budget, time, and iteration caps — instead of an unbounded agent.
Practical checklist
- Copy the template's continuous-claude tab as your starting command
- Set --max-runs from the template's suggested iterations
- Set --max-cost from the Budget Calculator's per-run cap
- Add --max-duration so a stuck run cannot run forever
- Keep human approval before merge — do not auto-merge
Example
PR Babysitter → continuous-claude --max-runs 5 --max-cost 5 --max-duration 1h --review-prompt "…" "Keep CI green and resolve review comments; do not merge." The template's boundaries go in the task; the caps stop runaway cost.
Common failure modes
Running without --max-cost or --max-duration
Dropping the template's boundaries from the task prompt
Allowing auto-merge
Ignoring the stall threshold on repeated failures