Once you have a working agent, the next win is handing it a repetitive job and letting it run on a schedule.
Pick a job worth automating
Good candidates are checks you do on a cadence: a morning summary, a status poll, a cleanup pass. Start with something low-risk and read-only.
Describe the task clearly
Write the instruction as if briefing a teammate — objective, inputs, and what a good result looks like.
Every morning at 8am: check the deploy status,
summarize any failures, and post the summary to #ops.
Schedule it
Create a scheduled routine so the agent fires without you.
openclaw schedule create "0 8 * * *" "morning deploy summary"
⚠ Warning
Give scheduled tasks a tightly scoped prompt. An open-ended instruction that runs unattended can wander — be explicit about what's in and out of scope.
Verify and iterate
Watch the first few runs. Tighten the wording wherever the output drifts from what you wanted.
★ Tip
Have the task report even when there's nothing to do ("all green"). Silence is ambiguous — a heartbeat tells you it actually ran.