alterlab-torch-geometric

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

• Third-party content exposure detected (high risk: 0.80). The skill's SKILL.md, references/datasets_reference.md, and included scripts (e.g., benchmark_model.py, visualize_graph.py) explicitly load public datasets via torch_geometric.datasets (e.g., Reddit, Twitch, Facebook, Planetoid), which are untrusted/user-generated content fetched from the open web and are ingested and used by the agent for training/analysis — allowing third-party content to materially influence model behavior or downstream actions.