control-to-video

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads AI model weights and configuration files from Hugging Face, which is a recognized and standard service for machine learning assets.
  • Evidence: The OF_MossSFX node in assets/motionfield_nodes/nodes_audio.py utilizes huggingface_hub.snapshot_download and implements a hardcoded repository allow-list (MOSS_REPO_ALLOWLIST) to prevent unauthorized downloads.
  • [COMMAND_EXECUTION]: Local shell commands are used for specialized tasks such as audio processing and executing ML models in specific Python environments.
  • Evidence: assets/motionfield_nodes/nodes_audio.py uses subprocess.run to execute ffmpeg for audio normalization and a local runner script (moss_cuda_runner.py) to manage specific sound generation library requirements.
  • [DYNAMIC_EXECUTION]: The skill employs standard plugin discovery mechanisms to register its nodes with the ComfyUI framework.
  • Evidence: assets/motionfield_nodes/__init__.py uses importlib.import_module to dynamically load internal node modules, and nodes_audio.py dynamically imports optional dependencies for extended functionality.
  • [INDIRECT_PROMPT_INJECTION]: The workflows provided by the skill are designed to interpolate user-provided natural language prompts into generative execution paths.
  • Evidence: Template strings like {{PROMPT}} are used throughout the workflows in the workflows/ directory to allow users to specify generation parameters. This is a primary intended function of the skill.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 08:11 PM