tao-train-nvpanoptix3d

Pass

Audited by Gen Agent Trust Hub on Jun 30, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: No malicious prompt injection patterns, such as instructions to override safety filters or extract system prompts, were found in the skill body or metadata.- [DATA_EXFILTRATION]: No unauthorized data access or exfiltration patterns were detected. The skill uses standard S3 path placeholders for data sources and does not access sensitive local credentials.- [REMOTE_CODE_EXECUTION]: The skill executes a specific Python module from the "nvidia_tao_pytorch" package. It does not download or run arbitrary code from the internet.- [COMMAND_EXECUTION]: Command execution is limited to the official TAO entrypoint. The skill provides clear guidance on the hardware and software environment (Docker) required for safe execution.- [CREDENTIALS_UNSAFE]: No hardcoded secrets, API keys, or private keys were found. The "encryption_key" parameter is documented as a user-provided input for model security.- [SAFE]: The skill processes structured dataset files (JSON/NPZ). While this represents a theoretical ingestion point for untrusted data (Indirect Prompt Injection surface), the use of formal JSON schemas for validation effectively mitigates this risk.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 30, 2026, 05:48 PM
Security Audit — agent-trust-hub — tao-train-nvpanoptix3d