molmim-nim

Pass

Audited by Gen Agent Trust Hub on Sep 30, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user to pull the nvcr.io/nim/nvidia/molmim:1.0.0 Docker image. This is the official NVIDIA Container Registry for the MolMIM NIM microservice and is a trusted vendor resource.
  • [COMMAND_EXECUTION]: The skill provides shell commands for starting a Docker container, performing a docker login to the NVIDIA registry, and polling a local health endpoint (/v1/health/ready) to ensure the service is active before performing inference.
  • [CREDENTIALS_UNSAFE]: Instructions are provided to source an NGC_API_KEY from a local .env file and map it to NGC_CLI_API_KEY for Docker authentication. This aligns with standard developer practices for secret management and does not indicate credential exfiltration.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: SMILES (Simplified Molecular Input Line Entry System) strings are accepted as input sequences in SKILL.md and references/examples.md for molecular optimization.
  • Boundary markers: The instructions do not define specific delimiters for the SMILES data when interpolated into request bodies.
  • Capability inventory: The skill uses Bash for Docker operations, Write for saving molecular artifacts, and requests (Python) for network communication with hosted and local endpoints.
  • Sanitization: The skill recommends using RDKit for validation of generated SMILES strings to ensure chemical validity, uniqueness, and property ranges before further downstream use.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 30, 2026, 07:49 AM