Run recurring work

What you enable first

• Enable any tool or network integration the run depends on, in Home → Integrations. A schedule cannot widen its own access when it fires.
• Decide where the output lands: a file in agent-home, an App, or a summary you read in the run history.

See Schedules for how script and model runs differ and how each firing executes.

The prompt to send

Ask for the work to be built and tested once before it is scheduled. A schedule that has never run successfully by hand tends to fail quietly at 3am.

Write a script in agent-home that collects yesterday's numbers from our
analytics export and writes them to reports/daily-2026-01-01.md, using the
run date in the filename. Run it once now and show me the result.

Once it works, schedule it daily at 07:15 Europe/London.
Rerunning it for the same day must replace that day's file, not add another one.

What you check afterwards

• Open Schedules and confirm the first firing produced the artifact you named.
• Review failed firings there; the failure is usually a missing dependency or an instruction that assumed context from the thread where you created it.
• Re-read one output a week later to confirm the run still says something useful.

The daily report schedule failed on its last two firings. Read the run history,
fix the underlying cause, run it once by hand to prove it works, and tell me
what was actually wrong.