Skip to content

Run Watch-and-Act Workflows with /loop

Use /loop when Claude should repeat a prompt on an interval.

This is useful for watch-and-act tasks where the next step depends on a condition changing.

  • Wait for CI to pass, then report status.
  • Check a deployment until it is healthy.
  • Poll a queue or dashboard.
  • Retry a flaky external check with a cap.
  • Monitor a long-running background process.
Every 5 minutes, check whether CI has finished.
If it passes, summarize the result and stop.
If it fails, inspect the failure and tell me the likely root cause.
Do not merge or deploy without approval.

Keep loop prompts narrow. Avoid giving broad write or deploy permissions to an unattended loop unless you have explicit safeguards.