Essay writing is a crucial component of middle school Chinese education. The NLPCC 2025 Evaluation of Essay On-Topic Graded Comments (EOTGC) task centers on evaluating primary and secondary school essays. It includes two subtasks: Relevance scoring of essays (Track1) and Generate relevance comments (Track2). This paper proposes a Progressive Augmented-Prompt Fusion framework (PAPF) based on Large Language Models (LLMs), which integrates multiple prompt enhancement techniques including role-playing, task decomposition, Chain-of-Thought, and In-Context Learning into structured prompt templates. The framework progressively guides the model to conduct more accurate essay writing evaluation through a hierarchical, step-by-step approach. Based on this framework, we carefully designed detailed prompt words for Track1 and Track2 respectively according to the actual evaluation requirements, and significantly improved the model’s performance without fine-tuning with training data. Our approach achieves the highest overall score across all leaderboards. Extensive comparative and ablation experiments have confirmed the effectiveness and universality of our method.