Claude Code Routines — Put Claude on a Schedule, in the Cloud
Configure once: prompt + repo + connectors. Runs on Anthropic's web infra — your laptop doesn't need to be on. Schedule / API / GitHub triggers. Pro 5, Max 15, Team & Enterprise 25 routines per day.
What a Routine Is — Above Plugins and Skills
A Routine is a replayable task unit: a prompt + connected repo + connectors (Slack / Linear / GitHub etc.). The three-layer model: Routine answers 'when to run what', Plugin answers 'what the client can do', Skill answers 'how to do a task class'. Routines run on Anthropic's cloud web infra — no local machine required and no laptop-stays-open assumption. It's the step that turns Claude Code from a dev tool into a SaaS automation platform.
Three Trigger Modes
1) **Schedule trigger**: prompt + cadence (hourly / nightly / weekly). Good for daily reports or weekly repo-health audits. 2) **API trigger**: each routine gets its own endpoint + auth token. External systems (Zapier / GitHub Actions / your CI) POST to fire it. Good for embedding into deploy pipelines. 3) **GitHub event trigger**: subscribe to repo events (PR / issue / push / release) — Claude opens a new session per match. Good for first-pass PR review or three-way issue triage.
Quotas, Visibility, vs GitHub Actions
Quotas: Pro 5 routines/day, Max 15, Team/Enterprise 25. Each execution counts (failures included). The web UI at claude.ai/code shows run history, logs, and failure reasons. Versus GitHub Actions: Actions runs your code, Routines runs Claude reasoning + tool calls — complementary, not substitutes. A common combo: Actions calls a Routine API in CI for Claude-driven review or summary.
CLI `/schedule` + Access from China via QCode
Type `/schedule` in Claude Code CLI and follow the prompts to create a routine — local config syncs to cloud. Or use the claude.ai/code web UI directly. Routines run on Anthropic's cloud; China users access via QCode's transparent gateway and get the same create/manage/trigger experience as overseas users. Requires a Pro+ plan.
FAQ
Does a failed routine count against my quota?
Today (research preview): yes, failed runs count. Platform-side errors (Anthropic infra issues) can be appealed through support. Best practice: add retry logic or a graceful-degrade prompt to your routine.
How do I filter GitHub triggers to avoid running on every PR?
The routine config supports filters by label, branch, or path pattern. E.g., only run for PRs with `area/auth` label, or only on main-branch events. Set filters in the web UI.
Routines vs GitHub Actions — what's the core difference?
Actions runs containers/scripts; Routines runs Claude reasoning. Routines are right when you need LLM understanding, cross-file edits, or natural-language reports. Actions is right for compilation, tests, deploys — deterministic steps. Common combo: Actions POSTs to a Routine API for Claude-driven review/summary.
Use Claude Code Routines via QCode
Transparent access to Claude Pro/Max/Team from China — same routine quotas, API, and GitHub triggers as overseas.
Start Your QCode Plan