arthai-marketplace

/schedule-routine

Walk through setting up a Claude Code cloud-hosted scheduled routine — cron, GitHub webhook, or API trigger.

Synopsis

/schedule-routine [--list] [--delete <name>]

When to use it

Automating recurring work on Anthropic-hosted infrastructure — daily PR digests, weekly issue triage — with no server, no polling, no local process
Reacting to GitHub events (PR opened/merged/closed, issue opened, release published) with a Claude Code routine
Exposing a routine as an HTTP endpoint you can trigger from any system
Not for: local event-driven watchers in your own session — see Monitors

Quickstart

/schedule-routine

What you’ll see: a 6-step guided setup — name, repository, prompt, trigger, connectors, validate & deploy. Progress is saved to .claude/.routine-draft.json, so you can stop and resume across sessions.

Examples

/schedule-routine                       # start the guided 6-step setup
/schedule-routine --list                # list existing routines (not scoped to the current repo)
/schedule-routine --delete pr-digest    # delete a named routine

Arguments & flags

Flag	Values	Default	What it does
`--list`	—	off	List existing routines (name, trigger, last run, status)
`--delete`	`<name>`	—	Delete a named routine

What it does

If a draft from a previous session exists, you’re first asked whether to resume it or start fresh (user-confirmation checkpoint). Then the guided setup runs:

Name — asks for a short kebab-case name (user-confirmation checkpoint); validates the format.
Repository — detects your repos via the gh CLI and asks you to pick one (user-confirmation checkpoint); falls back to manual org/repo entry.
Prompt — shows the rules for a good routine prompt (self-contained, clear success criteria, explicit “nothing to do” case), then asks you to write it (user-confirmation checkpoint). Vague or context-dependent prompts (“as discussed”, too short) are flagged for rewrite.
Trigger — asks you to choose cron schedule, GitHub event, or API call (user-confirmation checkpoint), then collects the schedule/event details. Plain-English schedules are converted to cron and confirmed back to you.
Connectors (optional) — asks which external services the routine needs: GitHub, Linear, Slack, Google Drive (user-confirmation checkpoint), with authorization notes for each.
Validate & deploy — shows the full configuration summary and a validation checklist. If warnings exist, you choose between fixing the prompt or deploying anyway (user-confirmation checkpoint). Requires the Claude Code CLI to be installed — if it’s missing, the skill shows install instructions and stops (your draft is preserved). If the CLI is installed but its routines create syntax differs from what the skill expects, it falls back to showing manual web-UI deployment steps instead.

After deployment you get a post-deployment checklist: test the routine, verify connector authorizations, confirm it appears in --list, and set up a /monitor ci watcher if you want to be notified of routine failures.

Output & artifacts

A deployed cloud-hosted routine running on Anthropic infrastructure
.claude/.routine-draft.json — in-progress setup state (deleted after successful deployment)
For API-triggered routines: an HTTP endpoint and key you can invoke from any system

Troubleshooting

Problem	Fix
`Claude Code CLI not found`	Install the Claude Code CLI and authenticate, then re-run — your draft is saved, nothing is lost
500 error on `--list` / `--delete`	Known API bug (claude-code issue #43440) — routines still execute correctly; only management operations are affected
`gh` CLI not authenticated in Step 2	Type the repo manually as `org/repo`
Prompt flagged as vague or context-dependent	Routines run with no conversation memory — rewrite the prompt to be fully self-contained
Invalid cron expression	The skill shows the format and common examples, then asks again

Monitors — local event-driven watchers; pair /monitor ci with a routine to catch failures
/onboard — session briefing that surfaces what’s worth automating