torch-geometric

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly loads and processes open/public, user-generated datasets (e.g., torch_geometric.datasets like Reddit, Planetoid/TUDataset, OGB, MovieLens) as shown in references/datasets_reference.md and in scripts such as benchmark_model.py and visualize_graph.py, meaning the agent ingests third‑party web content (social posts, forum data, etc.) that is read and used to train/analyze models and thus could materially influence behavior.