torch-geometric

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.70). The skill explicitly loads public, user-generated datasets (e.g., Reddit, Twitch, Flickr, GitHub, etc.) via torch_geometric.datasets as documented in references/datasets_reference.md and shown in scripts like benchmark_model.py and visualize_graph.py (which call Planetoid/Reddit/TUDataset), so the agent ingests and processes open third‑party content that could contain untrusted text.