Routines and /loop scheduling

Claude Code offers local-session scheduling and hosted scheduling:

#/loop

Each trigger is an ordinary prompt turn, inheriting the session's context, permissions, MCP, model and Passion8 configuration.

/loop 5m check whether the deployment finished

/loop check CI and address any review comments

Units include s, m, h and d. Cron rounds seconds up to minutes and adjusts intervals that do not divide evenly to an approximate cadence.

#loop.md

Customize the prompt for bare /loop:

Check the current PR. If CI failed, retrieve the failure logs and fix the issue.
Address and reply to review comments. If everything is quiet, output one status line.

Keep it short. Overlong content may be truncated, and adding it every turn increases fixed context cost.

#Cron tools

Ask for session-scoped scheduled tasks:

in 45 minutes, check whether the integration tests passed

remind me at 3pm to push the release branch

#Cache strategy

1. Each /loop trigger is a normal turn and can refresh the cache.
2. Waiting beyond TTL requires reading history again.
3. Dynamic loops choosing long waits should not assume a continuously warm five-minute cache.
4. Background sessions remain the same session, but paused time does not refresh TTL.
5. Cloud routines use separate sessions and do not share local Passion8 caches.

#Routines

A routine saves its prompt, repository, environment, connectors and triggers. It runs in Anthropic's cloud, even when your computer is off.

#/schedule

Create/manage routines conversationally from the CLI:

/schedule daily PR review at 9am

/schedule tomorrow at 9am, summarize yesterday's merged PRs

CLI creation primarily uses schedule triggers. Configure API/GitHub triggers at claude.ai/code/routines.

#Routine permissions and networking

Cloud routines run autonomously rather than requesting approval for every tool call. Limit their access:

MCP connector traffic uses Anthropic-hosted connections; this does not guarantee the cloud VM can access the same domain. Use local /loop or Desktop tasks for local databases, LAN services or files.

#API triggers

An API trigger provides a dedicated URL and bearer token. POST to /fire starts a new session and returns its ID/URL. An optional text field carries alert, log or deployment details.

curl -X POST https://api.anthropic.com/v1/claude_code/routines/trig_xxx/fire \
  -H "Authorization: Bearer sk-ant-oat01-xxxxx" \
  -H "anthropic-beta: experimental-cc-routine-2026-04-01" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{"text": "prod deploy failed, check alert SEN-4521"}'

This research-preview API may change headers, limits and response shape. Keep tokens out of repositories.

#Choosing with Passion8

#Troubleshooting

#Official references

• Scheduled tasks
• Routines
• Claude Code on the web
• Commands
• Prompt caching