Aspect sentiment triple extraction (ASTE) task is an important and practically valuable subtask in sentiment analysis, aimed at predicting the sentiment polarity according to extract entities and opinions from text. In existing research, the use of prompt templates was overly broad and lacked specificity, while also neglecting the interrelationships and complexity between different linguistic features. Therefore, we propose a model that uses Task-specific Prompt and Multi-level Linguistic Feature Enhancement (TP-MLFE) for aspect sentiment triple extraction. Specifically, to obtain the task-specific input, we designed prompt templates tailored to the characteristics of the ASTE task and input them along with the original text to reconstruct the linguistic feature graph, guiding the extraction of triples. Then, we constructed a multi-level linguistic feature enhancement module to strengthen various linguistic features at both local and global levels, allowing for better capture of both details and the macro context. Meanwhile, the multi-head attention mechanism combines global and local linguistic features, improving the model’s sensitivity to details and its precise understanding of the global context, which in turn boosts the overall performance of the model. Finally, biaffine attention and GCN are employed to integrate linguistic features. Experimental results show that TP-MLFE outperforms existing models in several benchmark tests, demonstrating its potential to effectively handle the complexity of ASTE tasks.