Navigation
ELOWEN · Documentation

Last updated: October 27, 2026

Browse documentation · Run a check before a scheduled prompt
Autonomous work and permissions

Run a check before a scheduled prompt

A check keeps a frequent scheduled job quiet until something actually changed, and it saves the model turn when nothing did.

What a check does

Chetty runs the command before the job's prompt and looks at the result:

  • if the command fails or prints nothing, the job is skipped without a model turn;
  • if it prints output, that output travels with the prompt.

The command is the cheap filter. The prompt should still say what to do with what the command found.

Add a check

  1. Open Settings → Automation and create or edit a recurring job.
  2. In the job's settings, add the command as its check.
  3. Keep the command itself quiet: print something only when there is news.
  4. Choose Run now and read History to confirm that a run with no output was skipped and a run with output reached the model.

A check that reports the same thing every time makes every run do a model turn, which is the cost you wanted to avoid.

check: "new-emails --since-last-run"

Permissions and limits

A check runs with the job's own execution permissions, so use only commands you trust. Keep the output short, because it becomes part of the prompt.

A check on the host needs an instance job or a personal job whose owner is an administrator, and that is checked before every run. A check inside a managed Project runs in that Project instead, and a managed-Project schedule must be personal.

The default time limit for a check is 60 seconds and an administrator can change it.

Tip: If a job keeps skipping, test the command in the same place the job runs it. A command that works in your terminal may fail in the job's environment.