Track 222 · 20 prompts

Agent Rescue & Oversight

Rescue a misbehaving AI agent and put it under proper governance — contain, diagnose, fix, prove with tests, then supervise and monitor. Agents get installed quickly and supervised rarely. When one invents facts, ignores rules, takes wrong actions or leaks data, this track is the rescue procedure: a pause-or-continue decision and containment, evidence and root-cause diagnosis, rewritten instructions and restrictions, knowledge clean-up, least-privilege permissions with approval gates and a kill switch, prompt-injection hardening, a regression pack, a proportionate supervision model, monitoring and alerts, incident communications, vendor accountability, a rebuild-or-retire decision, an Agent Oversight Scorecard and a 30-Day Rescue Plan. Seventeen prompts; no promised outcomes.

First prompts in this track