deepstream-import-vision-model

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches model configuration and weights from HuggingFace (huggingface.co) and NVIDIA NGC (api.ngc.nvidia.com). These are established, trusted sources for AI model development. The downloads are performned using validated helper scripts (hf-download-config.sh, ngc-download.sh) that use HTTPS with TLSv1.2 and implement input validation to prevent path traversal.
  • [COMMAND_EXECUTION]: Executes shell commands via subprocess.run in reporting scripts (md-to-html-pdf.py, render-mermaid-for-pdf.py) to invoke standard system utilities like wkhtmltopdf and mmdc. These are used for generating benchmark reports and diagrams as described in the skill's primary purpose. The scripts use non-shell execution (shell=False) and implement timeouts and boundary checks.
  • [SAFE]: The skill implements security best practices, including the use of isolated virtual environments for dependencies, mandatory zero-initialization of DeepStream parser structures to prevent uninitialized memory issues, and strict encoder policies that avoid prohibited codecs in favor of hardware-accelerated (NVENC) or standard LGPL fallbacks.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 02:34 PM
Security Audit — agent-trust-hub — deepstream-import-vision-model