Built a meta-agent to audit 100+ $GROK bots because they kept dropping standing instructions over time.
The Librarian Bot crawls every conversation thread, extracts all "from now on" and "never do that" directives, then audits whether each bot actually persists those rules across sessions. First run: 381 rules kept, 72 partially kept, 90 completely forgotten. One instruction repeated across 47 bots but saved in zero of them.
It also maps overlapping directives across the bot fleet, identifies redundant work, flags missing error alerts, and outputs a heat map + PDF report showing which instructions are honored and which drift.
This is agent orchestration at scale: you need a system that tracks what you told each agent and enforces memory persistence. Without it, your fleet degrades into entropy.
The Librarian Bot crawls every conversation thread, extracts all "from now on" and "never do that" directives, then audits whether each bot actually persists those rules across sessions. First run: 381 rules kept, 72 partially kept, 90 completely forgotten. One instruction repeated across 47 bots but saved in zero of them.
It also maps overlapping directives across the bot fleet, identifies redundant work, flags missing error alerts, and outputs a heat map + PDF report showing which instructions are honored and which drift.
This is agent orchestration at scale: you need a system that tracks what you told each agent and enforces memory persistence. Without it, your fleet degrades into entropy.
