skills/nvidia/skills/deepstream-dev/Gen Agent Trust Hub

deepstream-dev

Pass

Audited by Gen Agent Trust Hub on May 18, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill instructions and Python code examples follow standard NVIDIA DeepStream development practices for constructing hardware-accelerated video analytics pipelines.
  • [EXTERNAL_DOWNLOADS]: The documentation guides the user to fetch pre-trained models from the official NVIDIA NGC registry (api.ngc.nvidia.com) and adds repositories for well-known services like Confluent for Kafka protocol support.
  • [COMMAND_EXECUTION]: Includes standard setup procedures such as sudo apt-get install for system-level SDK dependencies (e.g., librdkafka-dev, libmosquitto1) and pip install for required Python libraries.
  • [DATA_EXFILTRATION]: Provides implementation patterns for transmitting inference metadata to message brokers (Kafka/MQTT) and external REST endpoints, which is a core functional requirement of the DeepStream SDK for real-time analytics.
  • [PROMPT_INJECTION]: No patterns of prompt injection, behavior overrides, or instructions to bypass safety guidelines were detected in the skill content or metadata.
Audit Metadata
Risk Level
SAFE
Analyzed
May 18, 2026, 06:15 AM
Security Audit — agent-trust-hub — deepstream-dev