tao-train-metric-learning-recognition

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill's documentation and deployment references guide the user to run official NVIDIA container images from nvcr.io (e.g., nvcr.io/nvidia/tao/tao-toolkit). These are well-known technology service domains and are consistent with the vendor identity.
  • [COMMAND_EXECUTION]: The skill defines tool actions (train, evaluate, inference, export, gen_trt_engine) that execute the ml_recog command-line utility. These operations occur within the context of the TAO Toolkit container environment and are aligned with the skill's purpose for model lifecycle management.
  • [DATA_EXPOSURE]: The skill includes placeholders and examples for data paths (e.g., s3://bucket/data/train). No sensitive credentials or private data are hardcoded within the skill files.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied configuration files and dataset paths at runtime. While this creates a potential surface for indirect injection via malicious data content, the skill uses structured YAML/JSON schemas to validate parameters, which provides a layer of protection against malformed input.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 12:09 PM
Security Audit — agent-trust-hub — tao-train-metric-learning-recognition