tao-train-pose-classification
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: No prompt injection or instructions to bypass safety guidelines were detected.
- [DATA_EXFILTRATION]: The skill does not access sensitive files or hardcode credentials. Network usage is restricted to official NVIDIA container registries and example S3 datasets.
- [OBFUSCATION]: No obfuscation or multi-layer encoding was found in the documentation or schemas.
- [REMOTE_CODE_EXECUTION]: External dependencies are limited to official NVIDIA TAO Toolkit Docker images. No untrusted remote code execution patterns were identified.
- [INDIRECT_PROMPT_INJECTION]: The skill defines an ingestion surface for skeletal pose data (JSON, NumPy, Pickle). This is assessed as safe because the data is processed by specialized ML tools within a container rather than the agent's prompt context. 1. Ingestion points: DeepStream BodyPose JSON and associated skeletal data files. 2. Boundary markers: Absent, as data is handled by CLI tools. 3. Capability inventory: Execution of containerized training and inference via Bash. 4. Sanitization: Handled by the TAO Toolkit internal validation logic.
Audit Metadata