molfeat

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

• Third-party content exposure detected (high risk: 0.80). The skill's ModelStore and PretrainedMolTransformer explicitly load pretrained models and model cards from public hubs (e.g., "HuggingFace Models" and ModelStore.load / model_card.usage in the documentation), which can ingest user-uploaded, untrusted third‑party model artifacts and usage text that the agent will read and display.