# Setting Up Scheduled Tasks with /loop for Recurring Monitoring in Claude Code > Master scheduled tasks in Claude Code with /loop for recurring monitoring. Auto-expire jobs after 3 days. Learn commands and cancellation for efficient task management. - Repository: [Shayan Rais/claude-code-best-practice](https://github.com/shanraisshan/claude-code-best-practice) - Tags: how-to-guide - Published: 2026-03-12 --- **Use the `/loop` skill in Claude Code to create session-scoped cron jobs that auto-expire after 3 days, with commands like `/loop 5m "check status"` and cancellation via `cron cancel `.** The `shanraisshan/claude-code-best-practice` repository documents a powerful built-in skill for automating recurring workflows. Setting up scheduled tasks with `/loop` for recurring monitoring allows you to run commands at fixed intervals without leaving your Claude Code session, making it ideal for deployment checks, periodic reporting, or continuous integration monitoring. ## How the /loop Skill Works The `/loop` skill is implemented as a wrapper around an internal `CronCreate` tool, as detailed in [`implementation/claude-scheduled-tasks-implementation.md`](https://github.com/shanraisshan/claude-code-best-practice/blob/main/implementation/claude-scheduled-tasks-implementation.md). When you invoke `/loop ""`, the system translates your interval into a standard cron expression and registers a background job. The parsing logic converts shorthand like `1m` into `*/1 * * * *`, with **one minute** being the minimum granularity. Each job created through this mechanism is **session-scoped**, meaning it resides only in memory and automatically disappears when your Claude Code session terminates. For safety, all recurring jobs enforce a **hard 3-day auto-expiration** limit to prevent runaway background processes. ### Command Execution Flow When a scheduled tick fires, the command executes in a separate subprocess. The output streams back to Claude as a standard user prompt, which triggers the normal hook event lifecycle—including `UserPromptSubmit`, `Stop`, and sound notifications. This design ensures that scheduled tasks integrate seamlessly with existing Claude Code workflows and side-effects. ## Creating Recurring Monitoring Tasks To start monitoring, use the `/loop` command followed by an interval and the command string. Intervals support minute-level granularity using the `m` suffix. Monitor system status every minute: ```bash /loop 1m "date" ``` Run a Claude skill every 5 minutes: ```bash /loop 5m /simplify ``` Check deployment status every 10 minutes: ```bash /loop 10m "check deploy status" ``` Each invocation returns a unique job ID that you will need for cancellation. The repository's [`README.md`](https://github.com/shanraisshan/claude-code-best-practice/blob/main/README.md) references these examples in the "Scheduled Tasks" section as canonical patterns for recurring monitoring. ## Managing and Canceling Jobs Although jobs auto-expire after 72 hours, you can manually terminate them earlier using the `cron cancel` command followed by the job ID displayed at creation time. ```bash cron cancel abc123 ``` This immediate cancellation removes the job from the session's in-memory cron table. According to the implementation guide in [`implementation/claude-scheduled-tasks-implementation.md`](https://github.com/shanraisshan/claude-code-best-practice/blob/main/implementation/claude-scheduled-tasks-implementation.md), no cleanup is required on your filesystem because the scheduler never persists jobs to disk. ## Configuration and Limitations Administrators can disable the scheduled task functionality entirely by setting the environment variable defined in [`best-practice/claude-settings.md`](https://github.com/shanraisshan/claude-code-best-practice/blob/main/best-practice/claude-settings.md). Set `CLAUDE_CODE_DISABLE_CRON` to globally prevent the `CronCreate` tool from accepting new jobs. Key constraints to remember: - **Session-bound**: Jobs vanish when you close Claude Code - **3-day maximum**: Hard limit prevents indefinite scheduling - **One-minute floor**: Sub-minute intervals are not supported - **Subprocess isolation**: Each tick spawns a new process ## Summary - The `/loop` skill creates temporary cron jobs via the internal `CronCreate` tool - Jobs auto-expire after 3 days and disappear when the session ends - Use `/loop m ""` to schedule recurring monitoring - Cancel active jobs manually with `cron cancel ` - Disable globally via `CLAUDE_CODE_DISABLE_CRON` in [`best-practice/claude-settings.md`](https://github.com/shanraisshan/claude-code-best-practice/blob/main/best-practice/claude-settings.md) ## Frequently Asked Questions ### How long can a /loop task run in Claude Code? All `/loop` tasks automatically expire after 3 days (72 hours) as a safety mechanism to prevent runaway processes, regardless of whether the Claude Code session remains active. ### What happens to my scheduled tasks when I close Claude Code? Because jobs are session-scoped and stored only in memory, they immediately terminate and disappear when your Claude Code session ends. They do not persist across restarts. ### Can I schedule tasks to run more frequently than every minute? No. The `/loop` skill enforces a one-minute minimum granularity. Intervals like `30s` are not supported, as the cron translation logic in [`implementation/claude-scheduled-tasks-implementation.md`](https://github.com/shanraisshan/claude-code-best-practice/blob/main/implementation/claude-scheduled-tasks-implementation.md) maps inputs to standard cron expressions with minute-level precision. ### How do I disable scheduled tasks entirely in my Claude Code environment? Set the `CLAUDE_CODE_DISABLE_CRON` environment variable documented in [`best-practice/claude-settings.md`](https://github.com/shanraisshan/claude-code-best-practice/blob/main/best-practice/claude-settings.md). When this flag is present, the `CronCreate` tool will reject all new `/loop` requests.