A Minimalist Prompt for Zero-Shot Policy Learning