Claude Code adds routines with GitHub and API triggers

• Anthropic put Claude Code routines into research preview on April 14, with official docs saying a routine bundles a prompt, repositories, and connectors, then runs on Anthropic-managed infrastructure via schedule, API call, or GitHub event in the cloud, even when your laptop is closed (Introducing routines in Claude Code, Automate work with routines).
• The trigger mix is the real unlock: the launch post on X points to GitHub events and API activation, while Trigger a routine via API says the /fire endpoint is aimed at CI pipelines, alerting systems, and internal tools.
• Anthropic also shipped a rebuilt desktop app the same day, and the desktop launch post plus Felix Rieseberg's repost frame it as a ground-up redesign for reviewing and steering multiple tasks instead of living inside a code editor.
• Aakash Gupta's walkthrough adds the clearest product read: the new desktop layout centers a task list and live preview, with parallel jobs isolated in separate git worktrees.
• The release did not stop at UI. v2.1.108 on GitHub also added session recap, 1-hour prompt-cache controls, and built-in slash-command discovery in the same build.

You can read Anthropic's announcement, skim the routines docs, and inspect the experimental API trigger docs. The same docs quietly note that the Desktop app now shows both local scheduled tasks and cloud routines in one grid, while the GitHub release bundles recap and caching controls into the same update.

Anthropic's official framing is simple: configure a routine once, then let it run in the cloud on a schedule, from an API call, or from a GitHub event. The company says routines are available on Pro, Max, Team, and Enterprise plans with Claude Code on the web enabled (announcement, docs).

The docs add the important mechanical detail that a routine is a saved package of four things: prompt, repositories, environment, and connectors. Those runs are fully autonomous cloud sessions, so there is no permission picker or approval step once a run starts, and any GitHub commits, PRs, or connector actions happen through the user's own linked accounts (docs).

The docs break the trigger model into three parts:

• Scheduled: hourly, daily, weekday, or weekly runs, with custom cron available through /schedule update
• API: a per-routine HTTP endpoint secured by a bearer token
• GitHub: repository events such as pull requests, pushes, issues, and workflow runs

Anthropic also lets one routine combine multiple triggers, so the same configuration can run nightly, react to PRs, and be started on demand from a deploy script (docs).

The API docs make that endpoint more concrete. According to Trigger a routine via API, a POST to /fire starts a new session for an existing routine and returns a session ID and URL, with an optional text field for per-run context like an alert body or failing log snippet. Anthropic labels the API experimental and says dated beta headers will carry breaking changes.

Anthropic rolled out a rebuilt Claude Code desktop app alongside routines, pitching it as a redesign from the ground up. The strongest product clue is what is missing: the UI is no longer organized around an editor pane.

According to Gupta's walkthrough, the sidebar is a task list, the main pane is a live preview, and multiple jobs can run in parallel on separate git worktrees. That is a sharp shift from "one prompt, one terminal" toward reviewing, dispatching, and triaging several agent sessions at once.

Anthropic's docs also draw a clean line between local and remote automation inside the app. In the Desktop app, "New remote task" creates a cloud routine tied to your Claude account, while "New local task" creates a machine-bound scheduled task that still depends on your own desktop being available (docs).

Anthropic's examples lean heavily toward software chores with a clear trigger and a clear output. The docs list use cases like:

• Backlog maintenance against an issue tracker
• Alert triage that opens a draft PR
• Bespoke PR review on pull_request.opened
• Deploy verification after production release
• Documentation drift checks after merged changes
• Cross-SDK library porting after a merged PR

That list reads like a catalog of jobs teams were already duct-taping together with cron, webhooks, and custom scripts. Greg Isenberg's reaction pushed the same idea into business language, but the official examples are already enough to show where Anthropic thinks the product lands first: repetitive engineering work that starts from an event and ends in a reviewable artifact.

The same release that brought routines and the rebuilt desktop app also changed the core product underneath. In v2.1.108 on GitHub, Anthropic added a recap feature for returning to a session, exposed 1-hour prompt-cache controls through new environment variables, and let the model discover built-in slash commands such as /init, /review, and /security-review through the Skill tool.

That release also tightened a few practical edges: /undo now aliases /rewind, /resume defaults to sessions from the current directory, and error messages now distinguish plan limits from server-side rate limits. It is a useful reminder that the April 14 launch was not just a new cloud automation surface, it was also a fairly dense Claude Code release.