Claude has added cloud scheduling for automated tasks, giving Claude Code users a way to run recurring work without leaving a local terminal or laptop active. Anthropic calls the feature routines.
A routine combines a prompt, a code repository and a trigger. It can run on a schedule, start after an API call or react to a GitHub event. Because the job runs on Anthropic’s infrastructure, it can continue even when the user’s computer is offline.
How Claude Cloud Scheduling Works
Users define the work Claude should perform, connect the relevant repository and choose when the routine should start. Each execution runs as a separate cloud task with its own logs and results, making the automation easier to review than an unattended local process.
- Scheduled routines: run recurring jobs such as daily triage, weekly reports or dependency checks.
- API triggers: start a routine from another application or workflow.
- GitHub events: react to repository activity such as issues or pull requests.
- Cloud execution: keep work running without an open terminal session.
Useful Tasks to Automate
Cloud scheduling is most useful for repeatable jobs with a clear output and bounded permissions. Examples include producing a morning summary of open issues, checking documentation for stale references, preparing release notes, reviewing dependency updates or generating a weekly engineering report.
The feature can also handle maintenance tasks that are easy to postpone. A team could ask Claude to scan for outdated tests every Friday, propose fixes in a branch and leave the result for human review.
Cloud Routines vs Local Scheduled Tasks
Claude Code offers more than one scheduling model. Local scheduled tasks depend on the user’s machine and environment. The /loop command is useful for repeated work inside an active session. Cloud routines are intended for jobs that need reliable execution when no local session is running.
| Method | Best for | Needs local machine |
|---|---|---|
| Cloud routine | Reliable recurring or event-driven work | No |
| Desktop scheduled task | Work tied to local files or apps | Yes |
/loop | Repeated checks in an active session | Yes |
Security and Cost Controls
Scheduled agents should receive the minimum permissions needed for the task. Teams should restrict repository access, keep secrets out of prompts, review proposed changes and set spending limits. A recurring error can otherwise consume tokens or create repeated unwanted changes.
Cloud scheduling moves Claude Code closer to a persistent engineering agent rather than an assistant used only on demand. The value will depend on reliable triggers, clear logs and human approval for actions that affect production systems.
Source: Anthropic’s Claude Code routines documentation. See also Claude Code Artifacts for sharing live pages.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
