deepchem

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly ingests untrusted third‑party content — e.g., arbitrary user CSV/SDF/FASTA files via CSVLoader/ SDFLoader/FASTALoader in scripts like train_on_custom_data and train_solubility_model, public MoleculeNet datasets via dc.molnet.load_* in multiple workflows, and pretrained models from external hubs via dc.models.HuggingFaceModel/GroverModel in transfer_learning.py — and the agent is expected to read and process that content as part of its training/prediction workflows.