tao-train-mask-auto-encoder
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill's documented actions, such as 'train', 'evaluate', and 'gen_trt_engine', are consistent with the intended purpose of managing machine learning model lifecycles within the TAO Toolkit framework.
- [EXTERNAL_DOWNLOADS]: The skill references container images from NVIDIA's official registry (
nvcr.io/nvidia/tao/). These references are necessary for the skill's operation and target a well-known, trusted technology service. - [COMMAND_EXECUTION]: Documentation includes standard shell commands for invoking the
maeCLI and Docker containers. These commands are transparently described and are essential for performing the model pretraining and deployment tasks. - [DATA_EXFILTRATION]: While the skill includes configuration for Weights & Biases (
wandb) for experiment logging, this is a standard industry practice. No unauthorized or hidden data exfiltration attempts were found. - [PROMPT_INJECTION]: The 'AutoML policy' triggers mentioned in the documentation are used for legitimate functional control of the training pipeline and do not attempt to manipulate agent behavior or bypass safety guardrails.
Audit Metadata