Formal Modeling and Performance Evaluation of Explainable Agentic AI for Trust-Critical Autonomous Engineering