Given that human skeletal data can be naturally represented by graph structures, Graph Convolutional Networks (GCNs) have shown considerable effectiveness in skeleton-based action recognition. To capture intricate associations and long-distance connections within skeletal data, researchers have introduced hypergraph learning into action recognition. However, existing research overlooks the inherent kinetic laws governing human motion in dynamic hypergraph generations, and fails to further explore and utilize the high-order semantic information embedded with hyperedges. To this end, we propose a novel Dual-Attention Dynamic Hypergraph Contrastive network (DADHCN) for skeleton-based action recognition. Recognizing the kinematic correlation of adjacent joints and the coordination of symmetric joints during the motion, we propose a dynamic hypergraph construction method based on neighbor-coordination constraints to capture local and global hypergraph structures. Furthermore, hypergraph contrastive decoupling is proposed to ensure the uniqueness of the contextual information in each hyperedge, thereby augmenting the informativeness of hypergraph structures. Upon constructed hypergraphs, we integrate a Dual-Attention Hypergraph Network to facilitate effective spatio-temporal modeling for the action features. Specifically, the dual-attention hypergraph module is designed to accomplish effective aggregation and correlation among a diverse range of informative hyperedges. The experimental results have verified the effectiveness of the proposed DADHCN, which achieves the state-of-the-art performance compared with previous hypergraph-based works on NTU-RGB + D 60, NTU-RGB + D 120, and N-UCLA.