Skills
Navigate
skills/nvidia/skills/tao-train-fast-foundation-stereo/Gen Agent Trust Hub

tao-train-fast-foundation-stereo

Pass

Audited by Gen Agent Trust Hub on Sep 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides instructions for the agent to use the Bash tool to run docker run and find commands, which are standard for the TAO Toolkit workflow.\n
  • The docker run command mounts host directories and environment variables to facilitate containerized model training and deployment.\n
  • A development tip in references/setup-and-run.md suggests using find ... -exec rm -rf {} + to clear __pycache__ directories in bind-mounted source paths.\n- [INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface where external dataset annotation files are ingested by the agent or its tools.\n
  • Ingestion points: The data_file parameter in references/skill_info.yaml points to annotations.txt files provided by the user or dataset registry.\n
  • Boundary markers: No explicit instruction delimiters or sanitation steps for the content of these files are mentioned in the skill.\n
  • Capability inventory: The skill involves file system access and execution of the depth_net tool via Docker.\n
  • Sanitization: The skill does not describe specific sanitization for the annotation file contents before they are processed by the TAO entrypoints.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 2, 2026, 03:31 PM
Security Audit — agent-trust-hub — tao-train-fast-foundation-stereo