Gait recognition is a biometric identifier that distinguishes individuals based on their walking patterns. Unlike other biometrics, gait recognition operates from a distance, requires no active cooperation, and is challenging to disguise. Traditional approaches predominantly utilize silhouette sequences to extract gait features, yet these methods are vulnerable to occlusions and fail to preserve fine-grained spatial details. To address these limitations, model-based methods utilize pose-estimation techniques to effectively capture spatial and temporal joint information, establishing skeleton-based recognition as a powerful alternative. However, existing skeleton-based methods often neglect the importance of different body parts and overlook the dynamic motion characteristics inherent in gait sequences. To address these challenges, this study proposes an Attention-Driven Dynamic Motion-Aware Graph Neural Network (AD-GNN), which integrates an Attention-Driven Graph Convolutional Network (AGCN) and a Dynamic Motion Network (DMN) module. AGCN facilitates selective focus on critical joint relationships across spatial and temporal dimensions, while DMN captures intricate local and global motion dynamics, enabling a richer feature representation of gait sequences. By aggregating the strengths of AGCN and DMN, the proposed model extracts discriminative features that significantly enhance gait recognition performance. Evaluations of the CASIA-B dataset demonstrate that the AD-GNN model achieves superior accuracy compared to state-of-the-art skeleton-based approaches, establishing its effectiveness and robustness for real-world gait recognition tasks.