← Back to library

Compare the last three runs of a recurring agent job

Michael Fenech (@Michael_Fenech_) shared a workflow audit for Grok Bot that compares the last three runs of a job you hand it regularly. It looks for repeated fixes, delays, conflicting instructions and recurring failures, then recommends a single change with evidence and a test, waiting for your approval. We show his prompt word for word with credit.

  • OutcomeHave your agent look back over the last three times it ran one recurring job, list the corrections you repeated, the points where it waited on you, the rules that clashed or went stale, the problems it fixed twice and the run that went best, then propose one evidence-backed change and a way to check it, without touching anything until you approve.
  • SetupMinutes
  • LevelStarter
  • CreditMichael Fenech (@Michael_Fenech_)
View original
Works with
Grok BotClaude CoworkChatGPTClaudeCustom agent

Published Nov 16, 2026Updated Nov 16, 2026AI dry run on sample input: passed Nov 16, 2026Source verified Nov 16, 2026

When should you use it?

  • A weekly or daily job gets done, but you keep nudging it the same way each time
  • You suspect an old instruction is making the agent wait for something it could find itself
  • You want one small, testable improvement rather than a full rebuild of the routine

Not for: Jobs that have run fewer than three times, since there is nothing to compare yet; Agents that cannot read their own past messages, outputs or instructions; Broad reviews of a whole agent team; this looks at one job at a time

What do you need, and what do you get?

You need

  • One recurring job, named the way you usually ask for it
  • An agent with access to the instructions, messages and outputs from at least three past runs
  • A few minutes to read its findings and approve or reject the proposed change

You get

  • How many past runs it could inspect and where its view was incomplete
  • Findings under five headings: corrections, delays, instructions, failures and better ways
  • One recommended change with the evidence behind it and the exact new wording
  • A plan for checking whether the change helped without hurting quality

How does it work, step by step?

  1. Choose one routinePick a job the agent repeats on a schedule, such as a Friday report or a Monday inbox sweep, and use the name you normally give it.
  2. Paste the auditGive the agent Fenech's prompt with the job filled in. It reads the last three runs and tells you which ones it could actually see.
  3. Check the evidenceFor each finding, open the message or rule it quotes. Treat findings without evidence as guesses.
  4. Approve one changeAccept or edit the single recommendation. Leave the other findings for a later round so you can tell what made the difference.
  5. Check the next runAfter the next run, confirm that the output was still right and that the friction it targeted is gone.

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 and others. Values in lime are placeholders.

Original prompt by Michael Fenech (@Michael_Fenech_)

Word for wordby Michael Fenech (@Michael_Fenech_)
Review the last 3 times you handled [job]. Compare what happened on each run using the instructions, messages and outputs you can access. Tell me how many runs you could inspect and flag any gaps.

1 - Corrections. What did I have to fix or explain more than once?

2 - Delays. Where did the job stop and wait? What input, decision or approval was missing?

3 - Instructions. Did an outdated, duplicated or conflicting rule cause extra work? Show me the wording if you can find it.

4 - Failures. Did you fix the same problem twice? What caused it, and what evidence supports that explanation?

5 - Better ways. Did one run go more smoothly than the others? Which step made the difference, and would it work next time?

Pick the one change worth making first. Show me the evidence, write the exact change you recommend and tell me how we’ll check whether it helped without lowering the quality of the result. Wait for my approval before changing anything.

Exactly as published by Michael Fenech (@Michael_Fenech_) in X post: a workflow audit that compares the last three runs of a Grok Bot job.

Our adapted version (optional)

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

Adapted prompt1 variable
Look at the three most recent runs of {recurring_job} and list anything I had to correct more than once, anywhere the job paused waiting on me, and any rule that caused extra work.
Say how many runs you could open.
Then suggest a single improvement, quote the evidence, write the new instruction and explain how we will know on the next run whether it worked and the result stayed as good.
Change nothing until I say yes.

What can go wrong?

  • Approving several fixes at once. If the next run goes wrong, you will not know which edit caused it.
  • Running it on a job the agent cannot fully see. If it reports gaps in the history, the comparison is partial.
  • Measuring only speed. The prompt asks for a quality check too, so keep the original success test alongside the new one.
  • Letting it rewrite instructions during the audit. The prompt says to wait for approval; keep it that way.

Our take

Most agent routines do not fail outright; they leak time through the same small corrections, run after run. Fenech's audit is useful because it asks the agent to compare several runs instead of judging the last one, which is how you spot a pattern rather than a bad day. We especially like two details. First, the agent has to say how many runs it could inspect, so a thin history is visible instead of hidden behind confident findings. Second, it must pick only one change, show the evidence, write the exact wording and propose how to test it without lowering quality. That discipline keeps you from rewriting a working routine on a hunch. His Friday report example is the right kind of target: a job that is technically correct but still needs you every week. Use it on routines with a stable goal and at least three real runs. In our dry run on a fictional weekly sales summary, it found the same export being resent each week and a stale rule that caused the wait, and proposed one folder check with a test for next Friday.

Where the time goes back: Small repeated frictions, like resending the same file every week, cost minutes on every run and rarely get fixed because each one feels too minor. A short review across recent runs finds the one change that removes the most repeat work.

FAQ

Why only three runs?
Three is enough to tell a pattern from a one-off while keeping the review short. For a daily job you can widen it, but keep the single-change rule.
Will it change its own instructions?
It does not. It proposes the exact wording and waits for your go-ahead before editing anything.
Does this only work with Grok Bot?
Fenech wrote it for Grok Bot, but any assistant that can read its past runs, messages and instructions can follow it.

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 workflow audit that compares the last three runs of a Grok Bot job
Creator
Michael Fenech (@Michael_Fenech_)
Published on
X
Original date
Nov 15, 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 16, 2026: passed. Dry-run tested by our AI test agent on sample input. Run count and gaps stated, evidence quoted, one change with a check, approval awaited.
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 16, 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 & AdminWorkflow✓ AI dry run

Build a daily briefing from today's field-service jobs

From a day's job or call export, get an internal briefing that counts unique jobs, puts decisions and callbacks at the top, deduplicates repeat rows, flags incomplete or conflicting records instead of filling them in, and stays with named internal recipients. Nothing in the briefing is sent to customers.

Credit: Tyler Griffin (@TylerGriffin)

Minutes
Operations & AdminWorkflow✓ AI dry run

Give each office job its own assistant with written limits

Set up a small group of office assistants with a written role card each, so sales drafts, inbox drafts, people-admin drafts, review replies, operations notes, number summaries and social drafts stay in their own lanes, cross-boundary requests are routed or refused, hidden instructions to skip approval are flagged, and nothing is sent, posted, booked or paid until a named person takes the final step.

Credit: Tyler Griffin (@TylerGriffin)

Under an hour
Operations & AdminWorkflow✓ AI dry run

Pick the nearest eligible tech for a recall or callback

From a recall or callback record and a roster that includes skills, certifications, service areas, availability and drive-time estimates, get one recommended technician and one backup, with ineligible people excluded and reasons given, ranked by drive time rather than straight-line distance, plus a booking draft for the office to confirm. Nothing is booked and no one is contacted until a dispatch lead approves it.

Credit: Tyler Griffin (@TylerGriffin)

Under an hour
Operations & AdminRoutine✓ AI dry run

Tidy your agent team's memory and keep a decisions log

Have your assistant review everything it has saved about you subject by subject, propose which outdated, repeated or conflicting entries to merge or remove, move standing decisions into a dated decisions page, and run a weekly tidy-up that changes nothing until you approve it.

Credit: Michael Fenech (@Michael_Fenech_)

Minutes
Operations & AdminPrompt✓ AI dry run

Find the old instructions your agent keeps obeying

When you keep correcting an agent about the same thing, trace the correction to the conflicting instructions behind it, fix the source with your approval, and retest safely.

Credit: Michael Fenech (@Michael_Fenech_)

Minutes