Claude Code Routines — scheduled cloud automation in research preview
Source: Anthropic, Introducing Routines in Claude Code (claude.com blog, 2026-04-14). Press coverage: 9to5Mac, SiliconANGLE, VentureBeat.
Key takeaways
- Routines are saved Claude Code configurations — a prompt, one or more repos, and a connector set — that run on Anthropic’s web infrastructure without your machine online.
- Three trigger types: scheduled (cron-like), API (HTTP endpoint + token), GitHub webhook (repo events).
- Research preview. Requires Claude Code on the web enabled. Daily limits: Pro 5 · Max 15 · Team/Enterprise 25. Extra usage available beyond limits.
- Create via
/schedulecommand in the CLI or the web UI at claude.ai/code. - Shipped same day as a redesigned Claude Code desktop app (side-by-side sessions, integrated terminal, file editing, HTML/PDF preview).
What a Routine is
“A Claude Code automation you configure once — including a prompt, repo, and connectors — and then run on a schedule, from an API call, or in response to an event.” — Anthropic
Runs in an isolated cloud environment against existing repo connections. No local terminal required.
The three trigger types
| Trigger | Mechanism | Example |
|---|---|---|
| Scheduled | Cron-like cadence (hourly / nightly / weekly) | “Every night at 2am: pull top bug from Linear, attempt fix, open draft PR” |
| API | POST to dedicated endpoint with auth token; returns session URL | Datadog alert → Routine triages → posts summary to Slack |
| GitHub webhook | Repo event (PR opened, push, etc.) | Flag PRs touching auth module; post summary to #security |
Daily run limits
| Plan | Runs/day |
|---|---|
| Pro | 5 |
| Max | 15 |
| Team | 25 |
| Enterprise | 25 |
Extra usage available beyond these limits (mechanism not specified in announcement).
How this fits the Playbook
- Overlap with
/badand ralph-style loops./badruns overnight sprints locally via git worktrees. Routines move the same pattern to Anthropic-hosted infra — no local machine needed, but no local OMC orchestration either. Pick based on whether you need OMC/BMAD tooling in the loop. - Complements Managed Agents. Managed Agents are API-first orchestration primitives. Routines are the scheduled-trigger + webhook layer on top.
- Regulated-vertical caveat. Cloud-hosted, scheduled, autonomous writes to repos. For SR 11-7 / EU AI Act Article 12 verticals, audit-log requirements apply — confirm per-run attribution before wiring into production pipelines. See Regulated AI.
- GitHub webhook trigger is the interesting one. Matches the “autonomous code-review bot” blueprint but with Anthropic hosting the loop. Lower ops burden than the self-hosted blueprint.
Example use cases (from the announcement)
- Nightly bug triage — pull top Linear issue, attempt fix, open draft PR
- CI failure analysis overnight
- Weekly dependency audits
- Cross-language porting between branches
- PR-touching-sensitive-module flagging via webhook
What to actually do
- Pilot on one low-risk scheduled job. Nightly dependency audit or changelog generation — reversible, no production impact.
- Measure the daily-limit burn. 5 runs/day on Pro is tight once you add API and webhook triggers. Team/Enterprise headroom matters if you plan fan-out.
- Don’t auto-merge. Keep draft PRs + human review until you’ve observed a few dozen runs.
- Compare against
/badbefore committing. If your team already runs a BMAD autonomous pipeline, Routines duplicate the scheduler layer without the dependency-graph parallelism. - Audit logging gap first. Before any regulated use, confirm Routines emit per-run logs compatible with your EU AI Act Article 12 / SR 11-7 obligations.
Related Playbook pages
- /bad: BMad Autonomous Development — local worktree-based overnight execution
- Anthropic Managed Agents launch — the orchestration primitive underneath
- Autonomous code review bot blueprint — self-hosted webhook pattern
- Regulated AI — attribution and audit-log requirements
- GitHub Actions guide — the CI/CD side of Claude Code automation