datamol
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill requires the installation of the 'datamol' Python package from a standard repository.
- [DATA_EXFILTRATION]: The skill features built-in support for remote file systems via the 'fsspec' library. Functions such as 'dm.read_sdf', 'dm.to_sdf', 'dm.read_csv', 'dm.open_df', and 'dm.viz.to_image' can perform network operations to read from or write data to S3, Google Cloud Storage, Azure, and arbitrary HTTP/HTTPS URLs. This provides a mechanism for data exfiltration if the agent is directed to interact with an external, attacker-controlled endpoint.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests complex molecular data from external files (SDF, CSV, Excel, Parquet), which serves as a potential vector for indirect prompt injection.
- Ingestion points: Data is loaded from untrusted files using various reader functions in the 'dm.io' module.
- Boundary markers: The skill does not prescribe the use of delimiters or 'ignore' instructions when processing data from these external sources.
- Capability inventory: The skill possesses extensive capabilities, including network file access, writing to the file system, and parallel processing using all available CPU cores ('n_jobs=-1').
- Sanitization: While 'datamol' provides chemical structure sanitization, it does not include mechanisms to sanitize or escape text-based metadata or SMILES strings that could be used to influence the agent's behavior.
Audit Metadata