deepchem

Pass

Audited by Gen Agent Trust Hub on Aug 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill integrates with Hugging Face to download and fine-tune pretrained molecular models such as ChemBERTa and MolFormer. Hugging Face is a well-known service for hosting AI models.
  • [EXTERNAL_DOWNLOADS]: The skill utilizes the MoleculeNet benchmark suite to download standardized datasets (e.g., Tox21, BBBP, Delaney) for training and evaluation. These datasets are hosted on official library repositories.
  • [PROMPT_INJECTION]: The skill processes external molecular data, representing an indirect ingestion surface.
  • Ingestion points: dc.data.CSVLoader, dc.data.SDFLoader, and dc.data.FASTALoader in SKILL.md and included scripts.
  • Boundary markers: Not applicable, as chemical featurizers parse structural strings (SMILES) into numerical tensors rather than direct natural language instructions for the LLM.
  • Capability inventory: The skill has access to shell execution via the Bash tool and file manipulation via Read/Write/Edit tools for model training and data processing.
  • Sanitization: Data is validated and converted into mathematical representations by specialized featurizers like RDKit and DeepChem's internal loaders.
  • [SAFE]: No malicious obfuscation, persistence mechanisms, hardcoded credentials, or privilege escalation attempts were detected. The skill follows established best practices for scientific computing and machine learning research.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 17, 2026, 02:13 AM