tao-train-image-classification
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines an interface for processing external datasets and labels, which is an inherent requirement for training models. This functional ingestion surface is mediated by structured configuration templates rather than direct execution of untrusted content.\n
- Ingestion points: Dataset directories and class label file paths are defined in
references/skill_info.yaml.\n - Boundary markers: The skill uses spec templates and JSON schemas to isolate user-provided data paths from instructional logic.\n
- Capability inventory: Model actions are executed via
docker runusing theBashtool to trigger the containerized TAO Toolkit.\n - Sanitization: Path validation and parameter filtering are handled by the TAO Toolkit's internal schema-based configuration parser.\n- [REMOTE_CODE_EXECUTION]: The skill executes model training and deployment logic by pulling official container images from NVIDIA's registry (
nvcr.io). These downloads are recognized as trusted, vendor-managed resources.
Audit Metadata