Therapeutic peptides have shown great potential in pharmaceutical applications, but AI-assisted peptide drug discovery remains underexplored. To address this gap, the paper proposes PPFLOW, a target-aware peptide design method based on conditional flow matching on torus manifolds. PPFLOW models the internal geometries of torsion angles to design peptide structures. The authors also establish PPBench2024, a large-scale protein-peptide binding dataset, to support deep learning models for structure-based peptide drug design. Extensive experiments demonstrate that PPFLOW outperforms baseline models in peptide drug generation and optimization tasks, showing superior performance in binding affinity, stability, validity, and novelty. Additionally, PPFLOW is effective in flexible re-docking and side-chain packing tasks. The contributions of the paper include a new task, a novel model, and the establishment of a comprehensive dataset.Therapeutic peptides have shown great potential in pharmaceutical applications, but AI-assisted peptide drug discovery remains underexplored. To address this gap, the paper proposes PPFLOW, a target-aware peptide design method based on conditional flow matching on torus manifolds. PPFLOW models the internal geometries of torsion angles to design peptide structures. The authors also establish PPBench2024, a large-scale protein-peptide binding dataset, to support deep learning models for structure-based peptide drug design. Extensive experiments demonstrate that PPFLOW outperforms baseline models in peptide drug generation and optimization tasks, showing superior performance in binding affinity, stability, validity, and novelty. Additionally, PPFLOW is effective in flexible re-docking and side-chain packing tasks. The contributions of the paper include a new task, a novel model, and the establishment of a comprehensive dataset.