Scheduled Tasks and Automation

Schedule Claude Code to run at specific times

Use the /loop skill for periodic tasks

Create cron jobs that invoke Claude

Build notification systems for long-running tasks

Know when automation adds value versus complexity

Claude Code does not have to run only when you are watching. You can schedule it to run at specific times, loop on tasks indefinitely, or trigger it from external events. This opens up use cases like overnight batch processing, periodic code maintenance, and automated monitoring.

This lesson covers scheduling and automation patterns for Claude Code.

The simplest form of automation is the built-in /loop skill:

This runs the prompt every 5 minutes until you stop it.

Interval formats:

5m - 5 minutes

1h - 1 hour

30s - 30 seconds

Examples:

Check token usage every 10 minutes.

Hourly test runs.

Health monitoring.

Stopping loops:

Press Escape or Ctrl+C to stop a running loop.

Limitations:

The loop runs in your current session. If you close the terminal, the loop stops. For persistent automation, use cron or external schedulers.

For automation that persists beyond your session, use cron:

Create a shell script:

Schedule with cron:

Add:

This runs at 2 AM daily.

Cron format:

Examples:

0 * * * * - Every hour

0 9 * * 1-5 - 9 AM on weekdays

0 0 * * 0 - Midnight on Sundays

Build scripts that run Claude for specific tasks:

Daily lint fix:

Weekly dependency update:

Overnight batch processing:

For long-running or scheduled tasks, notifications help you stay informed:

Email notification:

Slack notification:

Desktop notification (macOS):

Log files:

Append output to a log file for later review.

Trigger Claude based on events rather than schedules:

File watcher:

Git hook:

Create .git/hooks/post-commit:

Webhook receiver:

Build a simple server that triggers Claude:

For production automation on Linux, use systemd:

Create a service file:

Enable and start:

Automated tasks can run up costs quickly:

Limit frequency:

Do not run expensive tasks more often than needed. Daily is usually enough for maintenance tasks.

Use cheaper models:

Haiku costs less for simple tasks.

Scope narrowly:

Narrow scope reduces tokens.

Set timeouts:

Kill after 5 minutes to prevent runaway costs.

Budget monitoring:

Check your Anthropic dashboard regularly. Set up billing alerts.

Automated scripts need robust error handling:

Check exit codes:

Retry logic:

Logging:

Use /loop for simple in-session periodic tasks

Use cron for persistent scheduled automation

Headless mode (-p and --yes) enables scripted execution

Add notifications for visibility into automated tasks

Event-driven automation responds to file changes, webhooks, or git hooks

Control costs with narrow scope, cheaper models, and timeouts

Handle errors with retries, logging, and alerts

Systemd provides production-grade service management

Create a simple automated task:

Create a script at ~/scripts/claude-status.sh:

Make it executable: chmod +x ~/scripts/claude-status.sh

Test it manually: ~/scripts/claude-status.sh

Check the log: cat ~/claude-status.log

Optionally add to cron for hourly runs:

Remove from cron when done experimenting.

This exercise introduces scripted automation.

https://code.claude.com/docs/en/automation - Automation guide

https://code.claude.com/docs/en/headless - Headless mode reference

https://code.claude.com/docs/en/skills#loop - Loop skill documentation