tao-finetune-cosmos-embed

Pass

Audited by Gen Agent Trust Hub on Jun 30, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local helper script to resolve the correct Docker image for the model and action. It provides command-line instructions for training, evaluation, and inference using docker run with performance-optimizing flags such as --network=host and --ipc=host which are standard for high-performance machine learning workflows.
  • [CREDENTIALS_UNSAFE]: The skill requires a HuggingFace API token for model access but correctly implements safe secret management by instructing the user to pass the token via environment variables or persistent cache mounts. It explicitly warns against writing the token into specification files, logs, or reports.
  • [EXTERNAL_DOWNLOADS]: The skill fetches Docker images and pretrained weights from official NVIDIA repositories on HuggingFace and installs a specific version of the protobuf library from PyPI to address environment-specific compatibility issues within the container.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 30, 2026, 05:48 PM
Security Audit — agent-trust-hub — tao-finetune-cosmos-embed