Turn an AI coding agent's reconnaissance into a reviewable engineering plan by separating evidence from inference, naming decisions and risks, and defining validation before implementation begins.