Unmanned Aerial Vehicle (UAV) swarm networks face significant challenges in processing efficiency and information security when handling cross-shard collaborative tasks. This paper proposes T-TBCLPA (Trusted-Task based Constrained Label Propagation Algorithm), a secure, lightweight cross-shard consensus mechanism designed specifically for UAV networks. Our approach introduces a trust evaluation mechanism based on historical behaviors, combining lock periods, multi-signatures, and failure proofs to build a complete security guarantee chain. T-TBCLPA achieves dynamic optimization of task state allocation through task collaboration network modeling and label propagation mechanisms, enabling adaptive shard structure adjustment according to real-time load conditions. The scheme also designs a lightweight blockchain structure with state synchronization mechanisms, allowing new nodes to quickly participate in the consensus process by only synchronizing the latest state block. Experiments demonstrate that T-TBCLPA significantly outperforms existing methods such as Monoxide, Metis, and U-PBFT in throughput and cross-shard consensus latency, particularly with increasing shard numbers and high task rates.