Most developers know the drill. You write a script. You set up a cron job. You wire it into some monitoring tool and pray it runs cleanly overnight. Then something breaks, your laptop was asleep, and the task never ran.
Anthropic is trying to fix that with Claude Code Routines, now available in research preview for Pro, Max, Team, and Enterprise plan subscribers.
What are Routines?
A routine is a saved Claude Code configuration — a prompt, one or more repositories, and a set of connectors — packaged once and run automatically. Think of it as putting your most repetitive development tasks on autopilot.
The key differentiator: Routines run on Claude Code’s web infrastructure, so your machine doesn’t need to be online for each task. That alone removes one of the biggest friction points for scheduled automation.
Three Ways to Trigger a Routine
Routines support three trigger types, and you can combine them on a single routine.
Scheduled triggers run on a recurring cadence—hourly, daily, weekday, or weekly. Times are entered in your local zone and converted automatically, so the routine runs at that wall-clock time regardless of where the cloud infrastructure is located.
API triggers give each routine a dedicated HTTP endpoint. You POST to it with a bearer token, and a new session starts. This makes it easy to connect Claude Code to alerting systems, deploy pipelines, or internal tools.
GitHub triggers run automatically in response to repository events — pull requests, releases, and more. You can narrow exactly which events fire a routine using filters for PR author, branch name, labels, draft status, and more.
A single routine can combine all three. For example, a PR review routine can run nightly, be triggered by a deploy script, and also respond to every new PR.
Real-World Use Cases
Anthropic has been clear about what routines are designed to handle: unattended, repeatable tasks tied to a clear outcome.
Some examples from the documentation:
Backlog maintenance. A nightly schedule trigger reads issues opened since the last run, applies labels, assigns owners based on the referenced area of code, and posts a summary to Slack. The team starts the day with a groomed queue, no manual effort required.
Alert triage. When an error threshold is crossed, your monitoring tool calls the routine’s API endpoint. The routine pulls the stack trace, correlates it with recent commits, and opens a draft pull request with a proposed fix. On-call engineers review the PR instead of starting from a blank terminal.
Documentation drift. A weekly-scheduled routine scans merged pull requests, flags documentation that references changed APIs, and opens update PRs against the docs repo for an editor to review.
Library porting. A GitHub trigger fires when a PR merges to one SDK repository, then ports the change to a parallel SDK in another language and opens a matching PR. Two codebases stay in sync without a developer having to re-implement each change by hand.
Mitch Ashley, VP and practice lead for software lifecycle engineering at The Futurum Group, believes, “Claude Code Routines shift agent execution to Anthropic’s infrastructure, positioning the vendor as the execution environment for autonomous coding work. Scheduled, API, and GitHub triggers remove supervised sessions entirely, letting agents act on repositories without runtime approval.”
Ashley continues, “For DevOps teams, governance shifts from reviewing actions to constraining scope before runtime: repository access, connector selection, branch prefix rules. Observability into what the agent decided becomes the audit trail for pipeline owners. Teams treating routines as cron jobs will miss the governance surface they now”
How Routines Work Under the Hood
Routines run autonomously as full Claude Code cloud sessions: there are no permission-mode pickers and no approval prompts during a run. That autonomy is intentional, but it carries responsibility. The prompt you write is the most important part. Because the routine runs without supervision, it has to be self-contained and explicit about what to do and what success looks like.
Scope matters, too. What a routine can reach is determined by the repositories you select, the environment’s network access and variables, and the connectors you include. Anthropic recommends limiting each to what the routine actually needs.
Connectors — MCP-based integrations with tools like Slack, Linear, and Google Drive — are included by default when you create a routine. You can remove any that aren’t relevant to keep the access footprint small.
By default, Claude can only push to branches prefixed with “claude/”. This prevents routines from accidentally modifying protected or long-lived branches. You can disable that restriction on a per-repository basis if your workflow requires it.
Availability and Limits
Routines are available on Pro, Max, Team, and Enterprise plans with Claude Code on the web enabled. Routines draw down subscription usage just like interactive sessions do. In addition to the standard subscription limits, routines have a daily cap on the number of runs that can start per account.
You can manage routines from the web at claude.ai/code/routines, from the CLI using /schedule, or from the Claude Code desktop app via New Remote Task. All three surfaces write to the same cloud account, so a routine created via the CLI appears immediately in the web interface.
This is a research preview, which means behavior, limits, and the API surface may change. But the direction is clear. Anthropic is pushing Claude Code beyond an interactive tool into something closer to a persistent engineering resource — one that keeps working whether or not anyone is sitting at a keyboard.
For development teams already running Claude Code, routines are worth a look. The infrastructure lift is minimal. The time savings potential is not.
