The Agent Skills Directory

[SAFE]: The skill instructions and metadata contain no malicious patterns, obfuscation, or unauthorized access attempts across the 18 analyzed files.- [COMMAND_EXECUTION]: The skill defines the execution of nvdinov2 CLI actions within official NVIDIA TAO Toolkit Docker containers. This behavior is consistent with the skill's stated purpose for model training and deployment.- [EXTERNAL_DOWNLOADS]: The skill references official container images from the NVIDIA registry (nvcr.io), which is a trusted and well-known service.- [DATA_EXFILTRATION]: The skill supports logging to Weights & Biases (wandb), a standard and well-known service for machine learning experimentation. This is documented for telemetry purposes and is not a security concern in this context.

tao-train-nvdinov2