/loop is the simplest way to make Claude do the same thing over and over - poll for status, re-run a check, keep monitoring an endpoint.

What it does

Run /loop <interval> <prompt> and Claude re-executes the prompt on that cadence. Omit the interval and Claude picks a sensible pace based on the task. You can stop the loop at any time. Common patterns: watch a deploy, re-run tests, check for new emails.

When to use it

Polling status pages, health endpoints, or CI jobs.
Recurring checks on a long-running process.
Keeping a dashboard-like view running in a terminal.
Dev workflows where "keep checking until X" is the shape of the work.

Gotchas

Loops burn tokens on every iteration. Set a sensible interval and a stop condition.
One-off tasks don't benefit from loops - use them only for genuinely recurring work.
Loops stop when the session ends. For durable scheduling, use routines or scheduled tasks.

Official docs: https://code.claude.com/docs/en/scheduled-tasks.md#run-a-prompt-repeatedly-with-loop