tao-train-codetr
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructions involve downloading pre-trained weights from HuggingFace and using images from the NVIDIA Container Registry. These are recognized as well-known services and official vendor repositories.- [COMMAND_EXECUTION]: The skill provides shell commands using
docker runto execute thecodetrtool for model operations.- [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests Hydra specs, image datasets, and classmap files. The capability inventory includes thecodetrtool's file processing and inference operations. The documentation does not specify boundary markers or sanitization for these external inputs, though the risk is localized to the model environment.
Audit Metadata