deepchem

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

• Third-party content exposure detected (high risk: 0.70). The skill's workflows and scripts explicitly load public third‑party data and models (e.g., dc.molnet.load_* calls in SKILL.md and scripts/graph_neural_network.py / scripts/predict_solubility.py to fetch MoleculeNet benchmarks and dc.models.HuggingFaceModel with model id 'seyonec/ChemBERTa-zinc-base-v1' in scripts/transfer_learning.py), which are untrusted external resources that the agent ingests and uses to train/fine‑tune models and drive downstream predictions, so external content can materially influence agent behavior.