drug-discovery-pipeline

Pass

Audited by Gen Agent Trust Hub on Oct 5, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill requires the installation of the safe-mol Python package. This library is used for encoding molecule scaffolds into SAFE notation, which is the required format for the GenMol generation service.
  • [DATA_EXFILTRATION]: Molecule data and protein structures are sent to NVIDIA-hosted API endpoints at health.api.nvidia.com. The skill uses an Authorization header with a Bearer token sourced from the NGC_API_KEY environment variable. This is standard functional behavior for interacting with the specified BioNeMo NIM microservices.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests data from local protein target files (e.g., target.pdb) to perform molecular docking and affinity prediction.
  • Ingestion points: The script reads the full text of target.pdb from the workspace.
  • Boundary markers: There are no explicit delimiters or safety instructions provided to the agent to ignore potentially malicious content within the protein data file.
  • Capability inventory: The skill utilizes Bash, Write, and Read tools, and executes Python code capable of network communication.
  • Sanitization: The implementation includes a filtering step that extracts only lines starting with ATOM, which limits the content sent to the API to structural records and excludes other metadata fields.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 5, 2026, 06:43 PM