docs/ai/ai-failure-modes.md

# AI Failure Modes

The canonical [failure-mode catalog](../../.ai/guides/failure-mode-catalog.md) documents symptoms, likely causes, observable signals, prevention, and recovery for:

- attention, goal, scope, and context drift;
- recursive decomposition and recursive investigation;
- compulsive and repetitive verification;
- analysis paralysis, retry loops, tool loops, test/fix loops, refactor loops, replan loops, critic loops, and context reload loops;
- architecture rabbit holes, premature optimization, strategy oscillation, and infinite refinement;
- deadlock, livelock, evidence-free exploration, hallucinated repository knowledge, and completion avoidance.

Choose the dominant observable mode and apply one controller. Do not assign every available label, and never convert a behavioral analogy into a psychiatric diagnosis.