Detects and mitigates hallucinations in agent outputs by self-checking facts, verifying claims, and correcting unsupported or contradictory information.