Middleware for AI Agents that verifies grounding and prevents hallucinations. Returns structured retry suggestions for self-correction.