When should you use it?
- You keep writing long how-to lists and the agent still asks follow-up questions or takes an expensive path
- A recurring job needs the same limits every time, such as never buying anything or never sending mail on its own
- Several agents will share a job and each one needs the same picture of done and of what is off limits
Not for: A one-line favor where filling nine fields would take longer than doing the task yourself; Work that has no real success check, since the done field only helps if you can test it; Jobs whose limits change every run, which belong in a fresh note rather than a saved template
What do you need, and what do you get?
You need
- The result you want and the file, page or message that should exist when the job is finished
- The accounts, folders and tools the agent may read
- Your rules for acting alone, asking first and never crossing a line
You get
- A filled nine-field brief you can paste at the start of a chat or save as a reusable instruction
- A closing line that tells the agent to reuse what you already have and to question only missing facts that change the result
- A clearer handoff for anyone else who later runs the same job
How does it work, step by step?
- Start from the resultWrite the outcome and the finished item first. If you cannot name those, you are not ready to delegate.
- Name sources and timingList the places the agent should look and the deadline or start signal, including the time zone.
- Draw the three permission linesSeparate what it may do alone, what needs a yes from you, and what it must never do.
- Define done and stuckSay how you will check the work and what it should do if a source is missing or a step fails.
- Say how to hand it backName the format and the place the finished work should go, then add the line that favors a simple method.
The prompt
The original prompt by Michael Fenech (@Michael_Fenech_), word for word, with a link to where it was published. Below it is our adapted version with fill-in blanks, if you want it. Works in Grok Bot, ChatGPT agent, Claude Cowork and others. Values in lime are placeholders.
Original prompt by Michael Fenech (@Michael_Fenech_)
1: ACHIEVE: [The outcome you want and the finished deliverable] 2: USE: [The sources, accounts, tools and existing resources available] 3: TIMING: [The deadline, schedule or trigger, including time zone] 4: YOU MAY: [What Grok Bot can do without asking permission] 5: ASK ME BEFORE: [Actions requiring your approval] 6: DO NOT: [Anything it must avoid or never change] 7: DONE MEANS: [How to verify the job was completed successfully] 8: IF BLOCKED: [Whether to retry, stop or escalate, and when] 9: DELIVER: [The format and where to send the finished work] Then add this instruction at the end: "Use the simplest reliable approach. Reuse existing connections and workflows. Ask only about gaps that materially affect the result."
Exactly as published by Michael Fenech (@Michael_Fenech_) in X post: a nine-field brief for giving Grok Bot a job.
Our adapted version (optional)
Fill in the blanks (optional; the copied prompt uses your values)
Fill out a job brief for this task: {task}. Write short answers for: the result and finished item; sources you may use ({sources}); the deadline ({deadline}, {timezone}); what you may do without asking; what needs my approval; what you must never do ({forbidden}); how we will check that the work is finished; what you will do if a source is missing or a step fails; and the format and place for the finished work ({delivery}). Prefer a simple method and reuse tools I already have. Questions should cover missing facts that would change the result. Then do the job using that brief, and show the filled fields before you start.
What can go wrong?
- Leaving the approval and ban fields empty is worse than skipping the template. An agent that does not know its limits will guess.
- A vague done test such as 'make it good' cannot fail, so the agent cannot tell when to stop.
- Filling every field with a paragraph turns the brief back into the long how-to list this template was meant to replace.
- Saving the filled brief inside one chat hides it from the next agent. Keep a copy in the place your other instructions live.
Our take
Most weak agent jobs fail in the same place: the person describes the path and leaves the destination fuzzy, so the agent follows the path even when a shorter one exists. Turning that around, and stating the result before the method, is advice we already publish. What this template adds is the rest of the contract in one place you can scan: sources, a deadline, three permission levels, a test that can fail, and a plan for getting stuck. The closing line about picking a simple method matters more than it looks. Without it, agents often browse page by page when a connected app or a saved script would have been enough. We would keep each field to a sentence or two so the brief stays shorter than the how-to it replaces, and we would reuse a filled copy for recurring jobs rather than rewriting it from memory. The example in the original post, finding a few options under a budget without buying anything and flagging prices that cannot be checked, is a good test of whether the fields are doing real work. In our dry run for a fictional printer shop that needed three paper-stock quotes by 8 am local time, the agent compared current prices from named suppliers, refused to place an order, marked one catalog price as unchecked and delivered a one-page table instead of a long narrative.
Where the time goes back: A short filled brief replaces a long how-to and the extra round of questions that usually follows, and it stops the agent from taking a slower path you already have a simpler tool for.
FAQ
Do I have to fill in all nine fields every time?
How is this different from an outcome-first brief?
Does it only work with Grok Bot?
Source and credit
The original prompt on this page is by Michael Fenech (@Michael_Fenech_), shown word for word with a link to where it was published. The summary, notes and adapted version are ours.
- Original
- X post: a nine-field brief for giving Grok Bot a job
- Creator
- Michael Fenech (@Michael_Fenech_)
- Published on
- X
- Original date
- Nov 7, 2026
- License
- Not openly licensed; the original prompt is shown word for word with credit and a link to the original
- Originally for
- Grok Bot
- AI dry run
- Nov 8, 2026: passed. Dry-run tested by our AI test agent on sample input. All nine fields filled, no purchase attempted, unverified price flagged, comparison delivered as a table.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Nov 8, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
Are you the creator? Claim or correct this page.