tooluniverse-small-molecule-discovery

Pass

Audited by Gen Agent Trust Hub on Mar 30, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted user input (compound names, SMILES strings) and uses it to query external tools.
  • Ingestion points: compound_name and smiles parameters in tool calls across all phases.
  • Boundary markers: Absent; there are no explicit delimiters or warnings to ignore instructions embedded within the chemical data.
  • Capability inventory: The skill can perform network operations via ChEMBL, PubChem, and ADMET-AI APIs, and generate search URLs for commercial vendors.
  • Sanitization: No sanitization or validation of the input strings is described in the skill instructions.
  • [EXTERNAL_DOWNLOADS]: The skill specifies a dependency on the tooluniverse[ml] Python package for advanced ADMET prediction features. This package aligns with the vendor's naming convention and the skill's primary scientific purpose.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 30, 2026, 11:09 AM