← Back to library

Give every agent routine a clear point where it stops

Michael Fenech (@Michael_Fenech_) shared an audit for Grok Bot teams that checks whether each agent and routine knows when its job is finished, when to hand back to a person and when to give up. We explain how to run it on any agent setup and show his prompt word for word with credit.

  • OutcomeHave your lead agent review every recurring job, monitor and follow-up your agents run, check each against four stop rules (what counts as done, what should hand control back to you, how many retries or days are too many, and how a stop gets recorded), then propose fixes ranked by risk, changing nothing until you approve.
  • SetupUnder an hour
  • LevelIntermediate
  • CreditMichael Fenech (@Michael_Fenech_)
View original
Works with
Grok BotClaude CoworkChatGPT agentCustom agent

Published Nov 2, 2026Updated Nov 2, 2026AI dry run on sample input: passed Nov 2, 2026Source verified Nov 2, 2026

When should you use it?

  • Agents keep nudging people who have already answered
  • Monitoring or research routines carry on long after the question is settled
  • You find usage climbing with no clear new work behind it

Not for: One-off tasks that end when the chat ends; Setups where you cannot see your agents' routines or saved instructions; Jobs that are meant to run indefinitely, such as an uptime check, unless you also want a review of their alerts

What do you need, and what do you get?

You need

  • A lead agent that can read the instructions and run history of your other agents
  • A list of active routines, follow-ups and monitors, or permission to find them
  • Your current approval rules, so the audit keeps them intact

You get

  • A table of every recurring job with the stop rules it already has and the ones it lacks
  • A list of jobs that could loop, repeat work or keep acting after the goal is met
  • Suggested stop wording for each job, ordered by risk and waste
  • A single report that waits for your approval before anything changes

How does it work, step by step?

  1. Run it read-onlyGive the prompt to your lead agent and confirm it should only report, not edit.
  2. Check the inventoryCompare its list of jobs with what you know is running and add anything it missed.
  3. Start with chasersReview follow-up and outreach jobs first, since a wrong stop rule there reaches real people.
  4. Approve in batchesAccept the suggested stop wording for a few high-risk jobs, then watch a week of runs.
  5. Read the stop notesEach stopped job should now leave a short record of why it ended and what is still open.

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, Claude Cowork, ChatGPT agent and others. Values in lime are placeholders.

Original prompt by Michael Fenech (@Michael_Fenech_)

Word for wordby Michael Fenech (@Michael_Fenech_)
Run an EXIT audit across my Grok Bot team.

Find every active Bot and routine that performs recurring tasks, monitors something, follows up with people or works towards an outcome.

For each one, evaluate these four EXIT rules:

E — END CONDITION

Define exactly what success looks like. Identify the event or result that should mark the job as complete and stop further work.

X — EXCEPTIONS

Identify situations that should immediately stop or pause the Bot and hand control back to me, including unexpected replies, changed circumstances, errors or decisions requiring my judgement.

I — INACTIVITY LIMITS

Set appropriate limits on retries, follow-ups, elapsed time and repeated attempts that aren't making progress. Identify when a job should stop rather than keep consuming resources.

T — TERMINATION CONFIRMATION

Define how the Bot records why it stopped, what it completed, what remains unresolved and whether anything needs my attention.

Now:

1: Review each Bot and routine against all four EXIT rules. Show me which protections already exist and which are missing.

2: Flag jobs that could run indefinitely, repeat unnecessary work or continue acting after their original purpose has been fulfilled.

3: Recommend specific EXIT instructions for each Bot and routine, prioritised by risk and wasted effort.

4: Preserve all existing safeguards and approval requirements. Do not interrupt or modify any active job without my approval.

5: Present your findings and proposed changes in one report. Wait for my approval before implementing anything.

Exactly as published by Michael Fenech (@Michael_Fenech_) in X post: a four-part EXIT audit for when Grok Bot routines should stop.

Our adapted version (optional)

Fill in the blanks (optional; the copied prompt uses your values)

Adapted prompt2 variables
Review every recurring job my agents run on {platform}: routines, monitors, follow-ups and anything working toward a goal.

For each job, tell me: (1) the result that means it is finished, (2) the situations where it should pause and ask me instead of continuing, (3) how many attempts, reminders or days it may use before giving up, and (4) what it writes down when it stops.
Mark which of these each job already has and which are missing, and list any job that could loop, redo finished work or act after its goal is met.
Suggest wording for the missing rules, ordered by risk to {people_or_systems} first and wasted usage second.
Keep every current approval step. Report only; do not pause, edit or restart any job until I approve.

What can go wrong?

  • A stop rule that is too strict can end a follow-up before the person had a fair chance to reply. Set limits in days, not just attempts.
  • Agents may only see their own routines. Ask each one for its list if your lead agent cannot read the rest.
  • Stopping a job silently is its own failure. Insist that every stop leaves a note you can find.
  • Do not let the audit rewrite approval rules while adding stop rules; keep both visible in the report.

Our take

Most agent instructions explain how to begin a task and say nothing about how to finish it, which is why so many routines quietly keep going. Fenech's audit is useful because it treats stopping as a design decision with four separate parts, and the split matters: knowing what success looks like is different from knowing when a surprise should pull you back in, and both are different from a plain retry limit. The fourth part, a written record of why a job ended, is the one most setups skip, and it is what lets you trust that a quiet routine has actually finished rather than failed. We also like that the prompt explicitly protects existing approval rules, so tightening stop conditions cannot loosen anything else. Our caution is about follow-ups that reach customers or candidates: pick limits in calendar days as well as attempts, so a busy week does not end a conversation too early. In our dry run on a fictional agency with three agents, the audit found a payment reminder that kept sending after the invoice was paid and a news monitor with no end date, and proposed a paid-status check plus a 30-day review, all left for approval.

Where the time goes back: Jobs end when their purpose is met, so you stop paying for repeated searches and stop apologizing for follow-ups sent to people who already replied.

FAQ

What do the four letters stand for?
In Fenech's prompt they are the end condition, exceptions that hand control back to you, limits on inactivity and retries, and a confirmation of why and how the job stopped.
Will it change my routines?
Not on its own. The prompt asks for findings and proposed wording in one report and tells the agent to wait for your approval first.
Does this work outside Grok Bot?
Yes. Any lead agent that can see your other agents' routines, such as Claude Cowork or ChatGPT agent with scheduled tasks, can run the same review.

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 four-part EXIT audit for when Grok Bot routines should stop
Creator
Michael Fenech (@Michael_Fenech_)
Published on
X
Original date
Oct 30, 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 2, 2026: passed. Dry-run tested by our AI test agent on sample input. Found both seeded runaway jobs, kept approval rules, ranked by risk, proposed limits in attempts and days, no changes before approval.
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 2, 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.

More in Operations & Admin

All 203 Operations & Admin playbooks →

Related playbooks

Operations & AdminPrompt✓ AI dry run

Audit how you manage your agent team, not only the agents

Have your chief-of-staff agent review the last 30 days of how you delegated, clarified, interrupted, corrected, assigned, approved and followed up across your agent team, score eight management areas with evidence, list your three biggest weaknesses, and draft a seven-day improvement plan, without changing any bot, routine or instructions until you approve.

Credit: Michael Fenech (@Michael_Fenech_)

Under an hour
Operations & AdminPrompt✓ AI dry run

Brief an agent with nine fields, not a how-to

Before you hand a job to an agent, fill in nine short fields that name the result, the sources it may use, the deadline, what it may do alone, what needs your say-so, what it must never do, how to prove the work is finished, what to do when it gets stuck, and where to put the deliverable, then add a line that tells it to pick the simplest reliable method.

Credit: Michael Fenech (@Michael_Fenech_)

Minutes
Operations & AdminPrompt✓ AI dry run

Brief your agent on the outcome, not every step

Hand a new job to your lead agent by describing the finished result instead of the steps, and have it confirm what success looks like, ask only for context it cannot find itself, pick the simplest dependable approach, state what it will do alone and what needs your approval, and check the result against your goal before calling it done.

Credit: Michael Fenech (@Michael_Fenech_)

Minutes
Operations & AdminWorkflow✓ AI dry run

Drive desktop apps when an integration is blocked

Get work done through apps you cannot connect by API, by letting an agent control your desktop and operate them the way you would, with you approving sends.

Credit: Prasid

Under an hour
Operations & AdminRoutine✓ AI dry run

Get a weekly performance report on your rental listings

Have an agent review your short-term rental pricing and bookings every week and send a short report with suggested price changes you approve.

Credit: madaxe_again on Hacker News

Under an hour