Automating Workflows with Claude Code Scheduler

Claude Code's scheduling features let you save a prompt and have it run automatically on a recurring cadence. Instead of opening a new session every morning to type the same "check error logs" prompt, you define the task once and Claude handles it on schedule — monitoring deployments, reviewing PRs, running health checks, or generating daily briefings while you focus on other work.

Two Types of Scheduling

Claude Code offers two distinct scheduling mechanisms, each suited to different workflows.

Desktop Scheduled Tasks

Desktop scheduled tasks are persistent. They survive restarts, run on a visual schedule, and fire as long as the Claude Code desktop app is open. These are for recurring automation that you want to keep running indefinitely — daily code reviews, weekly dependency checks, or continuous monitoring.

How to create a desktop scheduled task:

1. Open Claude Code desktop
2. Navigate to the scheduled tasks panel
3. Define your prompt and schedule (e.g., every 6 hours, daily at 9am)
4. Select the project directory for context
5. Save — the task runs automatically on schedule

CLI Session-Scoped Tasks with /loop

The /loop skill is the quickest way to schedule a recurring prompt from the command line. Pass an optional interval and a prompt, and Claude sets up a background job that fires while your session stays open.

# Check deployment status every 5 minutes
/loop 5m check the deployment at staging.example.com and report any errors

# Monitor PR checks every 10 minutes (default interval)
/loop review open PRs in this repo and flag any with failing checks

# Run tests every 15 minutes during a refactor
/loop 15m run the test suite and report any new failures

Session-scoped tasks die when you exit the CLI session. They're designed for temporary monitoring during active work, not long-term automation.

Practical Use Cases

Deployment Monitoring

Watch a deployment in progress without staring at a terminal:

/loop 3m check the Vercel deployment status for this project. If the
latest deployment failed, show me the error logs and suggest a fix.

PR Babysitting

Keep an eye on CI/CD checks across your open pull requests:

/loop 10m list all open PRs in this repo. For any with failing checks,
read the failure logs and add a comment with the likely cause.

Error Log Surveillance

Monitor application logs for emerging issues:

/loop 5m read the last 100 lines of logs/error.log. If there are any
new errors since the last check, categorize them and estimate severity.

Dependency Audit

Schedule periodic checks for security vulnerabilities:

# Desktop scheduled task — runs daily
Run npm audit and report any high or critical vulnerabilities.
If any are found, check if patch versions are available and list
the upgrade commands.

Daily Briefing

Generate a morning summary of what changed overnight:

# Desktop scheduled task — runs daily at 8:30am
Summarize all commits pushed to main since yesterday. Group by author.
Flag any changes to database schemas, API contracts, or security-sensitive
files. Write the summary to daily-briefings/today.md.

Test Suite Health

Continuously validate that your test suite passes during a refactor:

/loop 15m run pytest tests/ -x --tb=short. If any test fails, identify
which recent file change likely caused it by checking git diff.

How Scheduling Works Under the Hood

When you create a scheduled task, Claude Code:

1. Saves the prompt — your instruction is stored as a reusable task definition
2. Sets the trigger — cron-based for desktop tasks, interval-based for /loop
3. Runs in context — each execution happens in your project directory with access to your files and CLAUDE.md
4. Reports results — output appears in your Claude Code session or notification system

Each execution is an independent Claude session. The scheduled prompt runs fresh each time — Claude doesn't carry conversation history between executions, but it does read your project files and CLAUDE.md for context.

Task Expiration

Recurring tasks created with /loop automatically expire after 3 days. The task fires one final time, then deletes itself. This prevents forgotten polling tasks from running indefinitely.

Desktop scheduled tasks do not auto-expire — they run until you disable or delete them.

Configuration Tips

Choosing the Right Interval

• 1-3 minutes — active monitoring (deployment in progress, CI pipeline running)
• 5-10 minutes — semi-active monitoring (PR checks, error logs)
• 15-30 minutes — background health checks
• Hourly/Daily — audits, briefings, dependency checks

Shorter intervals consume more API credits. Match the interval to how quickly you need to know about changes.

Writing Effective Scheduled Prompts

Scheduled prompts run without your real-time guidance, so they need to be self-contained:

Be explicit about what to check:

# Good
Check if the latest deployment to production succeeded by reading
the output of `vercel ls --prod`. Report the status and URL.

# Too vague
Check the deployment.

Define what to do with results:

# Good
If any test fails, write the failure details to test-failures.md
and include the file and line number of the likely cause.

# Too vague
Run the tests and let me know.

Include thresholds and conditions:

# Good
Check the error rate in logs/app.log. Only alert me if there are
more than 10 errors in the last hour or any error contains "database".

# Too noisy
Report every error in the logs.

Combining with Other Claude Code Features

Scheduler + Browser Access

Schedule tasks that interact with web applications:

# Requires /chrome or browser MCP
Every 30 minutes, open the Grafana dashboard at monitoring.internal/api-latency,
take a screenshot, and report if p99 latency exceeds 500ms.

Scheduler + Git Automation

Automate routine git maintenance:

# Daily task
Check for any branches that haven't been updated in more than 7 days.
List them with their last commit date and author. Write the list to
stale-branches.md.

Limitations

• Session-scoped tasks (/loop) require an active CLI session — they stop when you close the terminal
• Desktop tasks require the app to be running — they don't fire when the app is closed
• No external triggers — you can't fire a task based on a webhook or external event (use CI/CD for that)
• No state between runs — each execution is a fresh session, so tasks can't track "what changed since last run" natively (but they can read files or git history to approximate this)
• /loop tasks expire after 3 days — recreate them if you need longer-running monitors

Related Resources

• Claude Code: Complete Getting Started Guide — installation and basics
• Claude Code Browser Access — browser automation for scheduled web checks
• Claude Dispatch: Asynchronous Task Management — async task assignment from mobile