molfeat

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructions and documentation in SKILL.md and references/available_featurizers.md reference the installation of the MAP4 featurizer from an external GitHub repository (github.com/reymond-group/map4) which is not included in the list of trusted organizations.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process molecular representation strings (SMILES), which are external data inputs that can be manipulated to influence agent behavior, particularly given the skill's access to system tools.
  • Ingestion points: The skill processes SMILES strings through the MoleculeTransformer and FPCalculator classes as demonstrated in references/examples.md and references/choosing_a_featurizer.md.
  • Boundary markers: There are no explicit instructions provided for the agent to use delimiters or boundary markers to isolate molecular data strings from its operational instructions during processing.
  • Capability inventory: The skill is granted Bash, Read, Write, and Edit tools. The library utilizes parallel processing (n_jobs) and performs file system operations for model caching and configuration state management.
  • Sanitization: While the skill recommends molecular standardization via datamol.standardize_mol in references/examples.md, this process validates chemical structure rather than filtering potential natural language injection patterns within the strings.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 01:46 AM
Security Audit — agent-trust-hub — molfeat