This brings PyTorch Geometric's full graph neural network toolkit to Claude: 60+ GNN layers (GCN, GAT, GraphSAGE, GIN), heterogeneous graph support, and mini-batch training for node, link, and graph-level tasks. The skill covers PyG 2.7+ with clear guidance on the notorious extension wheel installation (pyg-lib, torch-scatter) that trips up newcomers. It explains the critical edge_index COO format and lazy initialization patterns. Use this when you're building GNNs on citation networks, molecular graphs, or point clouds. The MessagePassing API coverage means you can implement custom layers, not just stack built-ins. One thing: PyG's design choices (no built-in activations, manual bidirectional edges) are footguns if you're coming from vanilla PyTorch, and this skill walks through them.
npx skills add https://github.com/k-dense-ai/scientific-agent-skills --skill torch-geometric