tao-train-dino
Pass
Audited by Gen Agent Trust Hub on Aug 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill instructions and associated metadata files were analyzed and no malicious patterns, hardcoded secrets, or persistence mechanisms were found.
- [EXTERNAL_DOWNLOADS]: The skill documentation references NVIDIA-controlled resources for datasets and container images. These include S3 paths (e.g., s3://nvcf-storage-handling/) and the official container registry (nvcr.io).
- [PROMPT_INJECTION]: Analysis identified a surface for indirect prompt injection via user-supplied dataset URIs and training configurations. Ingestion points: Training and validation dataset URIs are provided by the user in the configuration spec as described in SKILL.md. Boundary markers: Not explicitly present in the data-source arrays, though the skill relies on the underlying TAO Toolkit CLI for execution. Capability inventory: The skill uses the Bash tool to execute dino commands within a controlled container environment. Sanitization: The skill includes a robust set of JSON schemas in the schemas/ directory to validate the structure and content of user-provided configuration files before processing.
Audit Metadata