tao-run-automl-deft-pipeline
Pass
Audited by Gen Agent Trust Hub on Jun 30, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
Bashtool to perform file operations (copying checkpoints, merging configuration files) and manage Docker containers (docker login,docker pull,docker image inspect). These operations are necessary for the ML training workflow. - [EXTERNAL_DOWNLOADS]: Downloads container images from the official NVIDIA Container Registry (
nvcr.io). This is a well-known service provided by the vendor. - [DATA_EXFILTRATION]: No sensitive data is transmitted to untrusted domains. Network access is scoped to official NVIDIA infrastructure for workflow execution.
- [CREDENTIALS_UNSAFE]: The skill correctly instructs the agent to check for the presence of credentials (like NGC or HuggingFace tokens) in environment variables rather than hardcoding them or prompting for them mid-execution.
- [SAFE]: The skill implements a "Consolidated Pre-Flight" gate, which is a security best practice. It performs read-only checks (file existence, image resolution, data leakage) before requesting user approval for the entire automated run.
Audit Metadata