tao-train-dino
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references container images from the NVIDIA Container Registry (nvcr.io) and pull example datasets from author-controlled S3 buckets (e.g., s3://nvcf-storage-handling/).
- [COMMAND_EXECUTION]: The skill provides instructions to execute
dinoCLI tools (train, evaluate, etc.) within a TAO Toolkit container using theBashtool to perform machine learning tasks. - [INDIRECT_PROMPT_INJECTION]: The skill has a theoretical attack surface due to the ingestion of external data and configurations.
- Ingestion points: Data source paths (S3 URIs) and YAML configuration templates are passed to model actions (SKILL.md, references/skill_info.yaml).
- Boundary markers: No explicit markers are defined for dataset content; configuration files use standard YAML delimiters.
- Capability inventory: Includes shell command execution and local file access via
BashandReadtools. - Sanitization: No custom content filtering or validation is specified for the remote dataset content by the skill instructions.
Audit Metadata