rfdiffusion-nim

Pass

Audited by Gen Agent Trust Hub on Aug 13, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides shell commands to set up a local Docker environment, including directory creation and permission management. Specifically, it uses chmod 777 on a local cache directory (LOCAL_NIM_CACHE) to facilitate container volume mounting, which is a common but overly permissive practice.
  • [EXTERNAL_DOWNLOADS]: The skill instructions facilitate downloading Docker images from the official NVIDIA Container Registry (nvcr.io). This is a legitimate operation for the skill's purpose and targets a well-known vendor repository.
  • [PROMPT_INJECTION]: The skill processes external protein data, creating a potential surface for indirect prompt injection.
  • Ingestion points: Reads protein structure files via Path("target.pdb").read_text() in SKILL.md.
  • Boundary markers: None; the PDB content is interpolated directly into the JSON payload.
  • Capability inventory: The skill performs network requests using requests.post to NVIDIA's API and writes files to the local system using Path.write_text.
  • Sanitization: None; the skill assumes the input PDB is well-formed and safe.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 13, 2026, 07:52 AM
Security Audit — agent-trust-hub — rfdiffusion-nim