torch-geometric

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

• Third-party content exposure detected (high risk: 0.80). The skill demonstrates loading public third‑party datasets and files (e.g., torch_geometric.datasets like Planetoid, Reddit, TUDataset/QM9, the InMemoryDataset.download hook, and CSV loading examples) which fetch or ingest public/user‑generated data (notably Reddit) that the agent is expected to read and interpret as part of its workflow, exposing it to untrusted third‑party content.