torchdrug

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

• Third-party content exposure detected (high risk: 0.70). The SKILL.md workflows explicitly instruct loading public datasets (e.g., datasets.BBBP(), datasets.Hetionet(), datasets.USPTO50k(), AlphaFoldDB/ChEMBL/ZINC references) and note "automatic downloading", so the skill will fetch and ingest external, public third‑party data that the agent reads and that can materially influence model decisions and downstream actions.