rdkit
Warn
Audited by Gen Agent Trust Hub on Mar 31, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The 'Best Practices' section of SKILL.md recommends using the 'pickle' module for fast loading of molecules. Pickle is inherently unsafe as it can execute arbitrary code during deserialization; loading molecules from untrusted .pkl files poses a significant risk.
- [PROMPT_INJECTION]: The skill's ingestion of molecular data formats like SMILES, SDF, and MOL creates a surface for indirect prompt injection. 1. Ingestion points: Molecule parsers and suppliers in SKILL.md and the provided scripts. 2. Boundary markers: Absent; there are no instructions to differentiate between chemical data and potentially malicious instructions in metadata. 3. Capability inventory: The skill includes scripts for file system write operations (CSV and SDF export). 4. Sanitization: While RDKit validates chemical structures, it does not sanitize text-based metadata fields in molecular files which an agent might process.
Audit Metadata