One outcome, one charter, a first task, a hard stop line. Skill first, overnight second.
You want the post standing before you close the laptop. You do not want a scheduled job you never signed.
One job: short name, one result, a hard stop line. After the lid closes, it runs work you would sign.
Pick a job you already do every week. Short name, one result. The profile is the contract.
When the format is wrong, say: use this shape from now on. Only then save the skill.
It will still run after the lid closes. Send, publish, reimburse, and production changes stay behind your yes.
Bots on one account can see sessions the others left. Leave a session you should not keep, then get out.
What's good: the stop line is a field, not a later reminder. The catch: schedule overnight first and fix the rules later, and the failure happens while you are gone.
Stand up one overnight job. Skip general helper. Use this seven-part shape and fill it with a job I already do every week: 1. Name 2. One job, written as an outcome 3. Profile description (the standing contract for Edit Profile) 4. First task (outcome, sources, constraints, deliverable, review point) 5. Hard approval boundary — sending, publishing, filing, reimbursing, and production changes stay with me 6. Tools it needs to sign into 7. Optional routine — only after the output is something I would sign Walk me through creating the named bot, pasting the profile, and taking over the computer for password / passkey / 2FA / CAPTCHA (never in chat). Send the first task. I will correct the format out loud: "Use this shape from now on." Then save a skill. A routine comes last. Turn Auto Review on. Do not write "allow everything in the browser."
改写自 https://dev.to/promptway/jobs-i-would-let-a-grok-bot-run-after-i-close-the-laptop-139p
Write-ups follow public posts and official docs. No likes, views, or dashboard numbers. Open the source to check the original. About