Most developers still use Claude Code the way they use ChatGPT: type a prompt, wait for a response, type another prompt. The /loop command changes that pattern entirely.
Shipped as a built-in slash command, /loop lets you schedule a recurring prompt that fires automatically on an interval while your session stays open. Tell Claude to check your deployment every 5 minutes, review a PR every 20, or run your test suite every hour, and it just does it in the background. The syntax is dead simple:
/loop 5m check if the deployment finished and tell me what happened
/loop 20m /review-pr 1234
/loop check the build every 2 hours
No interval? It defaults to every 10 minutes. You can use seconds, minutes, hours, or days as units, and natural language intervals work too.
What Makes This More Than a Fancy Cron Job
The practical use cases go beyond "run this command repeatedly." Developers are chaining /loop with other slash commands and skills to build genuinely autonomous workflows. Deploy monitoring, CI babysitting, and iterative code review all become set-and-forget tasks. Each time the loop fires, Claude gets full context and can make decisions, not just report status.
Three tools power it under the hood: CronCreate, CronList, and CronDelete. You can manage up to 50 scheduled tasks per session, and Claude handles the cron expression conversion automatically.
The Constraints Are Real
A few limitations worth understanding before you rely on this:
- Session-scoped only. Close the terminal and everything stops. There is no persistence across restarts.
- Three-day expiry. Recurring tasks auto-delete after 72 hours to prevent runaway loops.
- No catch-up. If Claude is busy when a task comes due, it fires once when idle, not once per missed interval.
- Idle-only execution. Scheduled prompts wait for Claude to finish its current turn before firing.
For anything that needs to survive a restart, Anthropic points users toward GitHub Actions workflows or Desktop scheduled tasks instead.
Who Actually Benefits
This is most useful for developers who keep a Claude Code session running during their workday. Polling a staging environment, watching for CI failures, or periodically checking whether a long build finished - these are the small, annoying tasks that break your flow when you do them manually. /loop handles them quietly.
The broader shift is real though: Claude Code is moving from a tool you talk to into a tool that works alongside you. The difference between prompting an AI and delegating to one is exactly this kind of persistent, autonomous operation.