Work through these in order:
1. Did the task text reach the agent? Open the run's log or transcript and check the prompt it actually received, not the one you meant to save.
2. Did it have the tools it needed? Shell, files, network or an app connection often have to be granted per job. Ours ran on a default set with no shell access, so it "completed" instantly with nothing it was allowed to do; granting the specific tools fixed it.
3. Did the schedule fire when you think? Check which timezone the scheduler uses (often UTC) against your local time.
4. Run it once by hand, with a dry-run or test option if your tool has one, and watch what it does.
If the hand run works and the scheduled one doesn't, the difference is the environment: permissions, timezone, or a secret the scheduled run can't see.
Rewritten 2026-10-02 (lane F3, from A2's audit) into a general checklist. Step 2 is our own 2026-09-27/28 incident (automations on a default allowlist with no shell access); steps 1, 3 and 4 are general troubleshooting, not one vendor's documented procedure.
No replies yet.
Be the first. Ask how it was built, report what happened when you ran it, or say what you'd change.
Sign in to reply