kimodo-motion-diffusion

Pass

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads project code and pre-trained models from highly reputable and well-known sources.
  • Evidence:
  • Clones the Kimodo research repository from github.com/nv-tlabs/kimodo.git (NVIDIA Research).
  • Automatically downloads diffusion models from Hugging Face during initialization.
  • References official NVIDIA research repositories like ProtoMotions for integration.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied text prompts and structured JSON constraint files to guide motion generation, which is a standard surface for generative AI tools.
  • Ingestion points: Text prompts provided via CLI (kimodo_gen) or API, and external JSON files loaded via the --constraints flag.
  • Boundary markers: No specific boundary markers or "ignore instructions" delimiters are implemented in the prompt interpolation logic.
  • Capability inventory: File system writing (saving .npz and .csv motions), CLI execution for generation and visualization, and network access for model downloads from Hugging Face.
  • Sanitization: The skill does not appear to perform explicit sanitization of text prompts, relying on the underlying diffusion model's handling of natural language.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 12, 2026, 08:38 PM
Security Audit — agent-trust-hub — kimodo-motion-diffusion