<p>In the field of complex networks, link prediction uses known information about network nodes and structures to anticipate unknown or future connections. While most existing methods focus on undirected networks, there is less research on directed networks. Current link prediction techniques often did not fully exploit the structural information and node attributes of directed networks, frequently neglected the directional structural heterogeneity of common neighbor configurations (i.e., the diversity of topological patterns formed by different edge-direction combinations) and the role of structural stability in link formation. To address these challenges, we propose a novel link prediction method for directed networks that integrates local information and topological stability. Our approach begins by defining generalized common neighbors at the node attribute level in directed networks. By analyzing the heterogeneity of directed neighbors, we introduce the concept of edge connection probability to assess how different network structures contribute to potential connections. Building on this, we refine existing local similarity measures and further investigate the topological stability between node pairs. By calculating the similarity of stable node pairs, we can more accurately predict the likelihood of new links. This approach not only considers node degrees and neighbor relationships but also deeply analyzes the potential structural stability between node pairs. Finally, we combine a link prediction method based on optimized local information with one based on topological stability, leveraging both node attributes and global network characteristics. Experiments on various real-world networks demonstrate that our method significantly enhances link prediction performance.</p>