tao-train-optical-inspection

Pass

Audited by Gen Agent Trust Hub on Aug 25, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill processes manufacturing datasets (CSV and image archives) which establishes an indirect prompt injection attack surface.\n
  • Ingestion points: Data enters the context through dataset.csv and images.tar.gz as defined in SKILL.md and references/skill_info.yaml.\n
  • Boundary markers: No explicit delimiters or safety instructions are used to distinguish untrusted data from instruction context.\n
  • Capability inventory: The agent utilizes the Read Bash tool and generates complex docker run instructions.\n
  • Sanitization: The instructions lack explicit validation steps for external data content before processing.\n- [COMMAND_EXECUTION]: The skill generates shell commands for managing Docker containers and ML tasks. It also provides guidance on using the TORCH_FORCE_NO_WEIGHTS_ONLY_LOAD=1 environment variable to load trusted checkpoints, including appropriate safety warnings for untrusted artifacts.\n- [EXTERNAL_DOWNLOADS]: The skill references container images and repositories from nvcr.io and github.com/NVIDIA-TAO, which are official vendor resources and are treated as trusted sources.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 25, 2026, 02:39 PM
Security Audit — agent-trust-hub — tao-train-optical-inspection