AURA: Behavioral matrices and validation tooling to detect manipulation, social engineering, and grey-zone threats in LLM interactions.