Tested by AI Setups That Work. Ran on Grok Bot, Oct 30, 2026, without the template: made-up 3-Bot fleet audited, all 3 planted slips caught with both wordings. Bot interviews were simulated.
What it does
Keep a fleet of bots honest about what they were told, with a weekly check that flags instructions that slipped or conflict.
Who it is for
For anyone running more than a couple of bots who has noticed an old instruction quietly changing how a bot behaves.
Tools
The Librarian Bot template by Brian Roemmele, added from the public directory
A list of the bots you actually run, so the audit covers all of them
One place to keep the audit notes, so last week can be compared with this week
Steps
Add the Librarian Bot template and point it at the bots you run.
Run the weekly interview: for each bot, ask what standing instructions it is following, in its own words.
Compare the answers with the instructions you actually gave. Flag anything that slipped, changed, or conflicts with another instruction.
Fix the flagged instructions at the source, in the bot itself, not only in the audit note.
Keep the audit notes week to week, so a slow drift shows up as a trend instead of a surprise.
The prompt
Interview each bot I run, one at a time, and list the standing instructions it is following, in its own words. Compare them with the instructions I gave, and flag any instruction that slipped, changed, or conflicts with another instruction. Do not change any bot instruction yourself. Report the flags and the exact wording on both sides, so I can fix them at the source.
Tips
Run the audit on a schedule, not when something already broke.
Fix at the source. An audit note alone does not change a bot.
Watch for conflicts, not just drift. Two instructions can both be remembered and still disagree.
Sources
Brian Roemmele, GrokBot.Rocks directory: https://grokbot.rocks