Work
Correction log: turn 'that's not what I meant' into approved rules weekly
You correct your bots, and next week they make the same mistake.
What hurt
@Michael_Fenech_ shared a prompt for the Chief of Staff: a shared correction log every bot writes to, lessons tracked from new to applied, a Monday 8am review that groups repeats and proposes exact wording, and a check on whether corrections actually drop. Nothing becomes a rule until you approve it. Full prompt quoted. Not affiliated with xAI.
What to install or create
Log every correction → weekly grouped proposals → you approve → apply and measure.
Bots
Chief of Staff
Owns the log and the Monday review.
Plugins
None on day one. Follow the first-run steps.
Routines
Weekly correction review. Mondays 8am, your time zone (author's setup). Group last week's corrections and propose rules for approval.
The prompt
Correction log + weekly review (author-quoted)
Quoted from @Michael_Fenech_ — as published. Do not invent a second system prompt.
Help me set up a correction log so my Bots get better at working the way I want. Keep existing permissions for sending, buying and publishing unchanged. 1 - Create a shared correction log. Choose a location the relevant Bots can read and update. Use an existing shared workspace where possible, and confirm access. For each useful correction, record the date, Bot name, task, what it did, what I wanted instead, and a proposed rule that could prevent the same mistake. Track each lesson as: new (captured), proposed (ready for my review), approved (I’ve agreed), applied (the Bot’s instructions have been updated), or rejected (not becoming a rule). 2 - Roll out correction logging across my Bot team. Add this instruction to each Bot: “When I correct your work, record any reusable lesson in the shared correction log. Include what you did, what I wanted instead, and a proposed rule that could prevent the same mistake. Distinguish a lasting preference from a one-off request. If unsure, flag it for review. Link to the example where possible, then continue with the corrected work. Don’t make a proposed lesson a permanent rule until I approve it.” Confirm each Bot has the instruction and can write to the shared log. 3 - Review the patterns weekly. Prepare a Monday 8am review in my timezone, owned by you as Chief of Staff. Group similar corrections from the previous week. Propose a change when a correction repeats or I flag it as important. For each proposal, show the examples, affected Bot, exact wording and where it belongs: instructions, a skill or memory. Check for conflicts with existing rules. Send one short list I can approve or reject. Apply only approved changes through a supported method, and confirm which were actually saved. 4 - Keep examples of what good looks like. When I identify an output I love, save it with a short explanation of why it worked. Reference it in relevant proposals so the Bot has a concrete standard. 5 - Check whether it helps. Report corrections per Bot and repeat corrections after a change was applied. Include completed-task counts where available, so a quieter week isn’t mistaken for improvement. Before creating the routine or changing instructions, show me a sample log entry and weekly review. Clearly label illustrative examples. After approval, confirm the log location, participating Bots and saved schedule.
First run
- 1Read https://x.com/Michael_Fenech_/status/2112858937959604427 (credit @Michael_Fenech_).
- 2Paste the author-quoted prompt into your Chief of Staff.
- 3Approve the sample log entry and sample review it shows you first.
- 4Confirm every bot has the logging instruction and write access.
- 5Approve or reject each Monday's short list.
Approval fence
Existing send, buy, and publish permissions stay unchanged; rules apply only after your approval.
What to tweak
Steal the status ladder: new, proposed, approved, applied, rejected.
Source
Michael Fenech @Michael_Fenech_ · Original post
Not affiliated with xAI. We did not write this prompt. Third-party template terms still apply.
Pro pack
Want the full setup, not just the idea?
Pro packs give you the job text, prompts, schedules and pitfalls for popular use-cases.