GRP: Goal-Reversed Prompting for Zero-Shot Evaluation with LLMs