Agent Production Readiness

Assess whether an AI agent is safe to run in production — autonomy limits, tool-call blast radius, prompt-injection resistance, observability of agent behavior, and human-in-the-loop controls. SRE discipline applied to non-deterministic systems.