OpenClaw Automation Guide

Use automation only after the same task succeeds interactively. Background execution removes the opportunity to correct a wrong model, missing credential, or unsafe tool request in the middle of a run.

Select the trigger

Use a cron job for work tied to a schedule. Use a heartbeat when an existing agent should periodically evaluate whether action is needed. Use a webhook for an authenticated external event.

Avoid running the same responsibility through several trigger types. Duplicate schedules can send repeated messages, overwrite the same file, or consume provider quota unexpectedly.

Inspect the installed CLI surface before creating a schedule:

openclaw cron --help
openclaw cron list
openclaw status --all

Bound the work

Define one owner agent, one destination, an explicit time zone, a maximum useful run duration, and a clear failure notification. Give the job only the tools and secrets it needs. Treat webhook payloads as untrusted input and authenticate the ingress before any agent turn starts.

For destructive or external actions, separate planning from execution and retain an approval step. A scheduled job should not silently broaden its own permissions.

Verify one run

Trigger or wait for a single execution, then inspect its session, output destination, logs, and next scheduled time. Confirm that a retry cannot create duplicate external effects.

Expected result

• Exactly one trigger owns the task.
• The intended agent and tool policy are used.
• The run appears in history with a clear result.
• Failure is visible instead of silently retried forever.
• The next run time and time zone are correct.

Review schedules after provider, channel, or credential changes. Background work that was safe under an old configuration may fail or behave differently after an update.