Compromised Agent Quarantine Runbook

Isolate a misbehaving agent in the first hour, without breaking the rest of the fleet.

A 6-page runbook for isolating a misbehaving agent in the first hour without breaking the rest of the fleet: scope and applicability, the trigger conditions for quarantine (behavior anomalies, prompt injection log signals, permission audit findings, PII exfiltration indicators, data leakage scanner hits), and the containment steps that follow. Written as a complete, adoptable draft covering policy gates, audit trails, and permissioning. For teams governing AI agents in production. From ghostcorpnet (Kestrelattice).