NVIDIA DeepStream Development
Reliable Python pyservicemaker guidance for building NVIDIA DeepStream video analytics pipelines.
The document restricts optional components and includes some secure-configuration, input-validation, and secret-handling examples. However, Kafka, Docker, networking, and display examples create external effects without consistent user confirmation, least-privilege guidance, credential handling, data-flow disclosure, or rollback instructions. Repository license metadata is NOASSERTION, further weakening verifiable governance evidence, so points are deducted.
Core rules, quick error references, and failure symptoms are fairly concrete, but the references contain inconsistencies: for example, the core guidance prefers nvurisrcbin while best_practices examples still use filesrc. Some examples contain placeholders or incomplete-looking sections. No complete statically verifiable test coverage is provided, so the score is conservatively capped below the static limit.
The name, audience, and DeepStream/pyservicemaker scenarios are clear, and the evaluation file includes positive and negative trigger cases. However, non-fit boundaries, input/output contracts, compatibility matrices, and non-NVIDIA or headless-environment constraints are under-specified. Chinese support and mainland-China network reachability are not addressed.
The skill has metadata, version, owner, a reference-document index, an error quick reference, and a benchmark report. It lacks the recommended Instructions and Examples sections; maintenance ownership and change history are incomplete; the license conflicts with repository NOASSERTION metadata; and some references are broad or show maintenance-quality risks.
The core rules address common DeepStream pipelines, dynamic ONNX models, pyservicemaker, Kafka, and virtual-environment failures, while the evaluation file defines actionable targets. However, output formats and representative artifacts are not verified within the skill, and examples still require adaptation to the user’s hardware, model, dependencies, and paths. Static review therefore cannot justify more than 7.
The materials provide version information, paths, failure symptoms, reference files, and a BENCHMARK.md containing reported evaluation results. They do not provide an independently reproducible test suite, CI workflow, or corroborating third-party evidence for key paths. The benchmark report alone is insufficient for independent verification, so the score remains below the static ceiling.
- Do not treat the documented versions, plugin properties, or API examples as execution-verified; test them on the target DeepStream release and hardware.
- Before using Docker, Kafka, RTSP, display servers, or external networks, explicitly confirm permissions, credentials, data destinations, and rollback; avoid adopting broad examples such as xhost + or host networking without review.
- Resolve conflicts between core rules and reference examples, and add CI, reproducible tests, Examples, non-fit boundaries, version change history, and clear license documentation.
What does this skill do, and when should you use it?
This skill supports developers building video analytics applications with the NVIDIA DeepStream SDK and its Python pyservicemaker API. It covers GStreamer-based processing, TensorRT inference, object detection and tracking, and Kafka or message-broker integration. It requires the agent to consult relevant reference documents and highlights exact rules for pipeline components, metadata iteration, request pads, buffer cloning, and configuration formats. It is a strong fit for DeepStream projects that need precise API guidance and troubleshooting, provided the required NVIDIA software stack is available.
Guides agents in creating and troubleshooting DeepStream Python applications, including pipelines using nvurisrcbin, nvstreammux, nvinfer or nvinferserver, nvosdbin, and platform-specific sinks; iterating frame_items and object_items metadata; configuring TensorRT, trackers, message brokers, dynamic sources, and ONNX models; handling NVMM zero-copy buffers, asynchronous processing, queue types, and pyservicemaker installation inside virtual environments; and diagnosing configuration, pad, inference-dimension, probe, and state-transition errors.
- Developers processing local video, HTTP, HLS, MPEG-DASH, or RTSP sources with DeepStream and Python.
- Engineers integrating TensorRT or dynamic ONNX models into video inference pipelines.
- Teams implementing object detection, explicitly requested multi-object tracking, or object metadata extraction.
- Application developers sending video analytics results to Kafka or another message broker.
- Developers troubleshooting DeepStream pipelines on Ubuntu 24.04 x86_64 or Jetson ARM64 systems.
What are this skill's strengths and limitations?
- Covers practical DeepStream development across GStreamer pipelines, TensorRT, metadata, dynamic sources, and Kafka integration.
- Provides explicit version requirements, installation paths, common errors, and troubleshooting directions.
- Imposes clear boundaries on optional components, supporting minimal working pipelines.
- Addresses dynamic ONNX models, differing YOLO output formats, and pyservicemaker virtual-environment issues.
- Depends on a specific environment including GStreamer 1.24.2, NVIDIA Driver 590+, CUDA 13.1, TensorRT 10.14.1.48, and Ubuntu 24.04.
- Requires several reference documents whose contents are not included here; adopters should verify they are present after installation.
- The supplied material does not show a dedicated test suite, benchmark results, or runtime validation for this individual skill.
- Trackers, analytics components, and message brokers are not enabled by default and must be explicitly requested.
How do you install this skill?
Install it with the Skills CLI command documented in the repository README: npx skills add nvidia/skills --skill deepstream-dev --yes. To target an agent explicitly, use for example: npx skills add nvidia/skills --skill deepstream-dev --agent codex. The skill becomes available the next time the agent loads installed skills.
How do you use this skill?
After installation, give the agent a concrete DeepStream task, such as: “Create a Python pyservicemaker DeepStream pipeline that reads an RTSP camera and runs TensorRT object detection.” The skill requires relevant reference documents to be read before code is generated. If a virtual environment is used, install the specified pyservicemaker wheel and pyyaml inside it; the actual reference-document contents and project layout are not included in the supplied material.