Deepstream Dev
Stand up NVIDIA DeepStream 9.0 video analytics pipelines with Python pyservicemaker, GStreamer, and TensorRT for solo builders shipping real-time detection and tracking features.
Overview
deepstream-dev is an agent skill for the Build phase that guides NVIDIA DeepStream SDK 9.0 video analytics pipelines with pyservicemaker, GStreamer, TensorRT inference, and Kafka integration.
Install
npx skills add https://github.com/nvidia/skills --skill deepstream-devWhat is this skill?
- Targets DeepStream SDK 9.0 with the Python pyservicemaker API for custom GStreamer-based video graphs
- Covers TensorRT inference hooks for object detection and multi-object tracking in live streams
- Documents Kafka and message-broker integration patterns for downstream analytics consumers
- Includes NVSkills-Eval benchmark notes: 7 evaluation tasks with 2 attempts per task on external/local profile
- Explicit triggers: video analytics pipelines, GStreamer processing, and broker-backed event export
- 7 evaluation tasks in bundled NVSkills-Eval summary
- 2 attempts per task on documented external profile
Adoption & trust: 1 installs on skills.sh; 1.1k GitHub stars; 2/3 security scanners passed (skills.sh audits).
What problem does it solve?
You need production-style video analytics on NVIDIA stacks but lack a repeatable agent workflow for DeepStream graphs, TensorRT models, and stream export.
Who is it for?
Indie builders shipping on-prem or cloud GPU video features who already tolerate GStreamer and TensorRT complexity.
Skip if: Teams needing only browser webcam demos, non-NVIDIA hardware, or turnkey SaaS video APIs without custom pipelines.
When should I use this skill?
Building video analytics pipelines, GStreamer-based video processing, TensorRT inference integration, object detection/tracking, or Kafka/message broker integration.
What do I get? / Deliverables
After the skill runs, you have an actionable DeepStream pipeline plan with Python service hooks, inference integration points, and broker wiring ready to implement and test on your hardware.
- DeepStream pipeline structure with pyservicemaker service boundaries
- TensorRT inference and tracking integration notes
- Kafka or broker egress configuration outline
Recommended Skills
Journey fit
Pipeline construction and broker hooks land in Build because you are wiring inference, streams, and integrations before production hardening. DeepStream work is primarily third-party SDK and message-broker integration (Kafka, GStreamer plugins), which maps to the integrations subphase rather than pure application UI.
How it compares
Use for NVIDIA-native pipeline authoring instead of generic OpenCV scripts or unrelated MCP media servers.
Common Questions / FAQ
Who is deepstream-dev for?
Solo builders and small teams building real-time video analytics on NVIDIA DeepStream who want an agent-guided path through pyservicemaker, GStreamer, and TensorRT rather than piecing docs together manually.
When should I use deepstream-dev?
Use it in Build → integrations when defining video analytics pipelines, integrating TensorRT models for detection/tracking, or connecting DeepStream outputs to Kafka; revisit in Ship when hardening stream performance before launch.
Is deepstream-dev safe to install?
Review the Security Audits panel on this Prism page and your org’s GPU container policies before granting shell, network, or API access to an agent running this skill.
SKILL.md
READMESKILL.md - Deepstream Dev
{ "name": "deepstream-dev", "description": "NVIDIA DeepStream SDK 9.0 development with Python pyservicemaker API. Use when building video analytics pipelines, GStreamer-based video processing, TensorRT inference integration, object detection/tracking, or Kafka/message broker integration.", "author": "NVIDIA CORPORATION", "skills": "./" } # Evaluation Report Evaluation of the `deepstream-dev` skill before publication through NVSkills-Eval. This benchmark summarizes 3-Tier Evaluation from NVSkills-Eval results for the skill. The goal is to document whether the skill is safe, discoverable, effective, and useful for agents before it is published for broader workflow use. ## Evaluation Summary - Skill: `deepstream-dev` - Evaluation date: 2026-05-28 - NVSkills-Eval profile: `external` - Environment: `local` - Dataset: 7 evaluation tasks - Attempts per task: 2 - Pass threshold: 50% - Overall verdict: FAIL ## Agents Used - `claude-code` - `codex` ## Metrics Used Reported benchmark dimensions: - Security: checks whether skill-assisted execution avoids unsafe behavior such as secret leakage, destructive commands, or unauthorized access. - Correctness: checks whether the agent follows the expected workflow and produces the correct final output. - Discoverability: checks whether the agent loads the skill when relevant and avoids using it when irrelevant. - Effectiveness: checks whether the agent performs measurably better with the skill than without it. - Efficiency: checks whether the agent uses fewer tokens and avoids redundant work. Underlying evaluation signals used in this run: - `skill_execution` (Skill Execution): verifies that the agent loaded the expected skill and workflow. - `skill_efficiency` (Efficiency): checks routing quality, decoy avoidance, and redundant tool usage. - `accuracy` (Accuracy): grades final-answer correctness against the reference answer. - `goal_accuracy` (Goal Accuracy): checks whether the overall user task completed successfully. - `behavior_check` (Behavior Check): verifies expected behavior steps, including safety expectations. - `token_efficiency` (Token Efficiency): compares token usage with and without the skill. ## Test Tasks The benchmark dataset contained 7 evaluation tasks: - Positive tasks: 5 tasks where the skill was expected to activate. - Negative tasks: 2 tasks where no skill was expected. - Unlabeled tasks: 0 tasks where positive/negative intent could not be inferred. Task composition is derived from the evaluation dataset when possible. Entries with `expected_skill` set are treated as positive skill-activation cases, while entries with `expected_skill: null` are treated as negative activation cases. ## Results | Dimension | Num | `claude-code` | `codex` | |---|---:|---:|---:| | Security | 8 | 74% (+9%) | 57% (-2%) | | Correctness | 8 | 94% (+6%) | 88% (+9%) | | Discoverability | 8 | 86% (+11%) | 76% (+9%) | | Effectiveness | 8 | 81% (+6%) | 78% (+9%) | | Efficiency | 8 | 72% (+12%) | 64% (+9%) | Score values show skill-assisted performance. Values in parentheses show uplift versus the no-skill baseline when baseline data is available. ## Tier 1: Static Validation Summary Tier 1 validation passed with observations. NVSkills-Eval ran 9 checks and found 34 total findings. Top findings: - MEDIUM PII/gps_coordinates: GPS coordinates (location information) (`references/service_maker_api.md:804`) - MEDIUM PII/gps_coordinates: GPS coordinates (location information) (`references/service_maker_api.md:827`) - MEDIUM PII/gps_coordinates: GPS coordinates (location information) (`references/service_maker_api.md:829`) - MEDIUM PII/gps_coordinates: GPS coordinates (location information) (`references/service_maker_api.md:1279`) - MEDIUM PII/gps_coordinates: GPS coordinates (location information) (`references/use_cases_pipelines.md:842`) ## Tier 2: Deduplication Summary Tier 2 validation reported findings. NVSkills-Eval ran 2 checks and found 34 total findings. Top finding