Use when you need to map how an AI feature can fail and what guards each failure — building a risk register for an LLM/agent feature before launch or review. Produces a register of failure modes with trigger, severity, likelihood, detection, mitigation, and residual risk, prioritized by exposure.
