tao-train-optical-inspection
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references official NVIDIA container images hosted on the NVIDIA Container Registry (nvcr.io). These are standard components for the TAO Toolkit workflow and originate from a trusted vendor.
- [COMMAND_EXECUTION]: The skill instructions and test suite involve executing shell commands, primarily through Docker or the provided validation script. These commands are necessary for the skill's primary function and are used in a controlled manner.
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided datasets. It mitigates potential injection risks by providing a dedicated validation utility (scripts/validate_dataset.py) that verifies the data schema and content before training or inference begins.
- [DYNAMIC_EXECUTION]: The documentation describes the use of an environment variable (TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD) that affects PyTorch checkpoint loading behavior. The skill explicitly warns against using this setting with untrusted models, demonstrating a security-conscious design.
Audit Metadata