Anthropic has officially launched Routines, an automated task feature for Claude Code, currently available in research preview. By configuring a prompt, code repository, and connectors just once, Claude can now work fully autonomously in the cloud. Since these tasks run entirely on Anthropic's cloud infrastructure, you don't need to keep your computer running. Even after you clock out and shut down, it can handle code backlogs, review code, and respond to cloud events on schedule.
This feature fundamentally transforms the cumbersome processes developers previously faced, such as managing cron jobs, setting up infrastructure, and maintaining MCP servers. The system supports three highly flexible triggering mechanisms that comprehensively cover daily development scenarios.
Three Trigger Mechanisms
Routines supports three trigger types, which can be used individually or combined.
The first is time-based triggering. You set a frequency (hourly, daily, weekly, etc.), and the routine executes automatically at the scheduled time. For example, every night at 2 AM, it can pull the highest-priority bug from Linear, attempt a fix, and open a draft PR. If you previously used /schedule in the CLI, those tasks have now automatically migrated to scheduled Routines with no action required on your part.
The second is API triggering. Each Routine has its own HTTP endpoint and authentication token. Sending a POST request to the endpoint starts a new session and returns a session URL. This allows you to integrate Claude Code into alert systems, deployment pipelines, internal tools, or any system capable of sending HTTP requests. A practical use case: point a Datadog alert to the Routine endpoint; Claude will pull stack traces, correlate them with recent commits, and prepare a fix draft before the on-call engineer even opens the page.
The third is GitHub event triggering. Subscribe to repository events such as PR creation, pushes, Issue updates, or workflow runs. Whenever an event occurs, the Routine launches automatically. Claude opens a dedicated session for each matching PR, continuously following up on subsequent comments and CI failures. For instance, you can configure it to monitor all PRs involving the /auth-provider module and automatically post change summaries to the #auth-changes channel. Support for more event sources will be added in the future.
A single Routine can bind to multiple trigger types simultaneously. For example, a PR review Routine could be configured to run nightly on a schedule, be called by deployment scripts, and respond to every new PR.
Anthropic is Already Using It Internally
Anthropic has already deployed Routines internally for documentation maintenance and backlog management. A typical internal use case involves automatically triggering documentation synchronization whenever a PR is merged into the release branch.
Based on early user patterns, common scenarios include:
For scheduled Routines: Categorizing, tagging, and assigning owners to new Issues every night, with results summarized in Slack; scanning merged PRs weekly to identify documentation referencing modified APIs and automatically opening PRs to update them.
For API Routines: Triggering after every CD pipeline deployment, where Claude runs smoke tests on the new build, scans error logs, and posts go/no-go results to the release channel; connecting documentation feedback or internal dashboard reports to a Routine, allowing Claude to open the corresponding repository session and draft modification proposals.
For GitHub Routines: Triggering a Routine for every PR merged into the Python SDK to automatically port changes to the Go SDK and open a corresponding PR; running custom team security and performance checklists automatically when a PR is opened, posting inline comments before human review begins.
How to Create Routines
Routines can be created via the web interface, the Desktop client, or the CLI. All three entry points write to the same cloud account and stay synchronized.
On the web, visit https://claude.ai/code/routines, click "New routine," and fill in the prompt, repository, cloud environment, triggers, and MCP connectors. The prompt is the most critical component; since Routines run fully autonomously without permission pop-ups, the instructions must be exceptionally clear, explicitly stating what needs to be done and what constitutes success.
In the CLI, running /schedule allows you to create scheduled Routines via conversation, or you can create them directly with a description like /schedule daily PR review at 9am. Note that API and GitHub triggers can only be configured via the web interface.
Permissions and Isolation
Routines belong to an individual claude.ai account and are not shared with team members. The Routine executes operations using your GitHub identity; commits and PRs appear under your account, and connector actions like Slack messages or Linear tickets use your bound credentials.
By default, Claude can only push code to branches with the claude/ prefix to prevent accidental modifications to protected branches. If you need to lift this restriction, you can enable "Allow unrestricted branch pushing" for specific repositories within the Routine configuration.
Usage and Limits
Routines consume subscription usage limits just like interactive sessions. Additionally, there is a daily cap on Routine runs per account: Pro users are limited to 5 runs per day, Max users to 15, and Team/Enterprise users to 25. Once the limit is exceeded, you can continue running Routines by enabling extra usage billing (pay-as-you-go); otherwise, the counter resets the following day.
Routines are available starting today to all paying Claude Code users (Pro, Max, Team, Enterprise) who have the Claude Code web interface enabled.
Entry point:
https://claude.ai/code/routines
Documentation:
https://code.claude.com/docs/en/routines
Major Upgrade to Claude Code Desktop
Additionally, Anthropic has released a significant upgrade to the Claude Code Desktop application today, featuring a complete redesign from the ground up.
Parallel Sessions and Centralized Management: You can now run multiple Claude sessions side-by-side in a single window, managed via a new sidebar, with significantly improved speed.
All-in-One Workspace: New features include an integrated terminal, file editor, HTML/PDF preview capabilities, and a faster code diff viewer.
Highly Customizable: Supports drag-and-drop layout, allowing you to tailor the interface to your specific workflow preferences.
Seamless Compatibility: All your existing CLI plugins work perfectly here, offering an experience completely consistent with the command line.
--end--