Gait recognition provides a long-range, contactless biometric solution by leveraging individuals’ unique walking patterns. However, unimodal methods based on either silhouettes or skeletons often struggle under challenging conditions such as clothing variations, occlusions, and viewpoint changes. Silhouette-based approaches offer rich appearance cues but are susceptible to visual noise, while skeleton-based methods exhibit structural robustness yet lack detailed spatial context. To address these limitations, we propose the Gait Collaborative Assessment Network (GCAN), a multi-modal framework that jointly exploits silhouette and skeleton information to learn more comprehensive and discriminative identity features. GCAN comprises two complementary representation learning branches: the silhouette branch introduces a Gait Spatiotemporal Heterogeneous Unit (GSHU) and a Semantic Assessment Module to capture fine-grained temporal appearance features and assess semantic quality at the silhouette level; the skeleton branch incorporates an Augmented Spatiotemporal Graph Convolutional Network (AST-GCN) to model joint-level motion patterns and spatiotemporal dependencies. These two modalities are integrated through a collaborative feature fusion module that enables mutual enhancement and robust identity representation. Experiments conducted on the CASIA-B and Gait3D benchmarks demonstrate that GCAN consistently outperforms state-of-the-art methods under various challenging scenarios.