molfeat

Pass

Audited by Gen Agent Trust Hub on Nov 6, 2026

Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill implements local caching of molecular embeddings using the pickle module as seen in SKILL.md and references/examples.md. While standard for machine learning performance, pickle is susceptible to unsafe deserialization if local cache files are replaced with malicious ones.- [INDIRECT_PROMPT_INJECTION]: The skill processes SMILES strings and molecular structures, which are untrusted external inputs.
  • Ingestion points: MoleculeTransformer and FPCalculator take SMILES strings as input across documentation and examples.
  • Boundary markers: No explicit instruction delimiters or 'ignore embedded instructions' warnings are provided for the molecular data inputs.
  • Capability inventory: The skill primarily performs numerical transformations and lacks dangerous capabilities such as arbitrary shell execution or broad file system access.
  • Sanitization: Input molecules are validated and parsed by the underlying RDKit/datamol libraries, which typically filters out malformed chemical strings.- [EXTERNAL_DOWNLOADS]: The skill downloads pre-trained molecular models and embeddings from well-known services including HuggingFace, Microsoft (Graphormer), and DGL-LifeSci. These operations are core to the skill's functionality and target reputable repositories.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 6, 2026, 12:15 PM