datamol
Pass
Audited by Gen Agent Trust Hub on Mar 13, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted external data from formats such as SDF, CSV, and Excel, as well as from remote URLs (S3, GCS, HTTP). This creates a vulnerability to indirect prompt injection where malicious text could be hidden in molecular data or metadata fields.
- Ingestion points: Data loading functions including
dm.read_sdf,dm.read_csv, anddm.open_dfinSKILL.md. - Boundary markers: No specific delimiters or instructions to disregard embedded text are included in the documentation.
- Capability inventory: The skill includes functions for file writing (
dm.to_sdf) and remote data access viafsspec. - Sanitization: Molecular sanitization is mentioned for chemical validity, but no sanitization for natural language instructions is present.
- [EXTERNAL_DOWNLOADS]: Fetches the
datamolpackage from the standard Python Package Index. - [COMMAND_EXECUTION]: Installs the required library via the shell and uses parallel processing features that utilize subprocesses.
Audit Metadata