deepchem
Pass
Audited by Gen Agent Trust Hub on Mar 31, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads molecular datasets and pretrained model weights from well-known services. Evidence: Functions like dc.molnet.load_tox21() in SKILL.md fetch benchmark data from DeepChem official infrastructure. Evidence: HuggingFaceModel in scripts/transfer_learning.py downloads pretrained weights from Hugging Face for models like ChemBERTa and MolFormer.
- [COMMAND_EXECUTION]: The skill provides Python scripts to execute machine learning tasks via a command-line interface. Evidence: scripts/graph_neural_network.py and scripts/predict_solubility.py allow users to train and evaluate models on local or benchmark data.
- [SAFE]: The skill follows security best practices and uses trusted repositories for its operations. Evidence: The scripts use restricted choices for dataset and model selection, preventing arbitrary function execution. Evidence: Data processing is limited to specific chemical formats (SMILES, SDF, FASTA) which are unlikely to contain executable malicious code.
Audit Metadata