tao-train-pose-classification

Pass

Audited by Gen Agent Trust Hub on Aug 25, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes skeleton keypoint sequences from external data files, which is a standard operational requirement but creates an inherent surface for indirect prompt injection.
  • Ingestion points: Data enters the workflow through file paths for NumPy arrays (.npy), Pickle files (.pkl), and DeepStream BodyPose JSON files as defined in SKILL.md and references/skill_info.yaml.
  • Boundary markers: Absent; the instructions do not specify the use of delimiters or 'ignore' instructions for content embedded within these data formats.
  • Capability inventory: The skill uses the Bash tool to execute pose_classification CLI commands for training, evaluation, and inference, as configured in the references/ directory templates.
  • Sanitization: Not explicitly documented; the skill relies on the internal parsing logic of the TAO Toolkit to handle these data formats.
  • [COMMAND_EXECUTION]: The skill generates and executes shell commands for the pose_classification toolkit. This is the primary intended function of the skill for NVIDIA TAO workflows and does not involve unauthorized or suspicious command sequences.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 25, 2026, 02:38 PM
Security Audit — agent-trust-hub — tao-train-pose-classification