pubchem-compound-search

Pass

Audited by Gen Agent Trust Hub on Jul 10, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill processes data from the external PubChem PUG-REST API, creating a surface for indirect prompt injection.
  • Ingestion points: Data from pubchem.ncbi.nlm.nih.gov (JSON, SDF, PNG) is fetched and processed in Python scripts.
  • Boundary markers: None. The skill does not instruct the agent to use delimiters or treat the API output as untrusted content.
  • Capability inventory: Python code includes file writing (saving SDF and PNG files) and network requests.
  • Sanitization: External data is handled directly without escaping or validation.
  • [COMMAND_EXECUTION]: Instructions provide standard commands to install well-known Python packages (requests and pandas).
  • [EXTERNAL_DOWNLOADS]: Retrieves chemical datasets, molecular properties, and structure files from the authoritative PubChem service.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 10, 2026, 03:35 PM
Security Audit — agent-trust-hub — pubchem-compound-search