schedule-bounded-agent-work

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# Schedule bounded agent work

## Inputs

- Provide one deterministic command or typed action, its working directory, and its success signal.
- Provide cadence, time zone, runtime platform, allowed writes, and the owner of irreversible
  decisions.
- Name required environment variables without copying secret values into the schedule definition.

## Workflow

1. Confirm that the work is recurring or future-dependent. Finish one-time work now instead of
   hiding it behind a schedule.
2. Make the command idempotent and non-interactive. Give it a timeout, an overlap policy, durable
   logs, and an explicit no-op result.
3. Write a schedule specification containing cadence, time zone, working directory, environment,
   timeout, overlap behavior, state path, success signal, failure route, write boundary, and owner.
4. Run the exact command once in the intended environment and capture its mechanism-level output.
5. Use a native scheduler already present on the host, such as Codex Automations, launchd, a systemd
   timer, cron, or GitHub Actions. Do not invent a second scheduler inside the workflow.
6. Install or modify the schedule only within the authority granted for the task.
7. Verify that the scheduler lists the job with the intended next fire time. After a real fire,
   verify the durable output rather than treating process exit as proof of success.
8. Document platform-specific setup, localization points, and the uninstall or disable path.

## Hard gates

- Never place credentials, tokens, personal data, or private document text in the schedule file or
  log.
- Use a named resource lock or skip policy when two runs could write the same state.
- Keep external messages, payments, submissions, and permission changes behind human approval.
- A scheduler entry without a successful same-environment dry run is not complete.

## Output

Return the schedule specification, installed adapter or exact platform mapping, dry-run evidence,
next-fire verification, and recovery or removal instructions.

From qilinxiaoxiang/agent-skills-and-cards, MIT.

More from qilinxiaoxiang/agent-skills-and-cards