rtvi-cv-scaffold-vss-service — Claude Code skill
rtvi-cv-scaffold-vss-service generates a standalone DeepStream perception microservice that feeds detection payloads into VSS Search and Alerts via Kafka mdx-raw.
LicenseRef-Proprietary ★ 4,106 updated 2026-10-23
npx skills add NVIDIA/skills --skill rtvi-cv-scaffold-vss-service -a claude-codeWhat rtvi-cv-scaffold-vss-service does
The skill generates a runnable perception microservice repository built on DeepStream 9.1 with a YOLO26 reference detector. The scaffold handles primary inference plus tracking, formats object metadata to match the VSS protobuf schema, and routes it directly to the mdx-raw Kafka topic. From there, services downstream like Alert Verification, Search Workflow, and Behavior Analytics consume the stream using standard VSS Compose profile flags.
The generated structure includes pgie-yolo26-config.txt, parser bindings for NvDsInferParseYolo26, and Compose setups. Without the customer ONNX file, matching class labels, and compiled parser library, the live DeepStream pipeline will refuse to run. The skill will safely stop at scaffold generation and host-level unit tests when those model assets are absent.
Generating the code requires Python 3.10+. Running the resulting service requires an NVIDIA GPU host running Linux, Docker with NVIDIA Container Runtime, the nvcr.io/nvidia/deepstream:9.1-triton-multiarch image, and an accessible VSS deployment at version 3.2.1.
What it helps with
- generate a standalone DeepStream vision microservice repo
- stream YOLO26 object detections into VSS Search and Alerts via Kafka mdx-raw
- validate the protobuf-2 integration contract before GPU deployment
- scaffold perception pipeline placeholders while model weights are being prepared
Example prompts for the agent
- Scaffold an RTVI perception microservice running YOLO26 in DeepStream to feed VSS through Kafka
- Validate the generated RTVI VSS service structure locally and prepare host unit tests
- Can the live VSS integration run if the YOLO26 ONNX weights and parser library are not ready?
How to install rtvi-cv-scaffold-vss-service
1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill rtvi-cv-scaffold-vss-service -a claude-code2Manually: put the files in place
~/.claude/skills/rtvi-cv-scaffold-vss-service/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/rtvi-cv-scaffold-vss-service/
1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill rtvi-cv-scaffold-vss-service -a codex2Manually: put the files in place
.agents/skills/rtvi-cv-scaffold-vss-service/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill rtvi-cv-scaffold-vss-service -a cursor2Manually: put the files in place
.cursor/skills/rtvi-cv-scaffold-vss-service/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install NVIDIA/skills skills/rtvi-cv-scaffold-vss-service2Manually: put the files in place
.github/skills/rtvi-cv-scaffold-vss-service/SKILL.md1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill rtvi-cv-scaffold-vss-service -a gemini-cli2Platform CLI
gemini skills install https://github.com/NVIDIA/skills --path skills/rtvi-cv-scaffold-vss-serviceManually: put the files in place
.devin/skills/rtvi-cv-scaffold-vss-service/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about rtvi-cv-scaffold-vss-service
How to install rtvi-cv-scaffold-vss-service in Claude Code?
Run npx skills add NVIDIA/skills --skill rtvi-cv-scaffold-vss-service -a claude-code. The skill file lands in ~/.claude/skills/rtvi-cv-scaffold-vss-service/SKILL.md.
What does rtvi-cv-scaffold-vss-service do?
It creates a deployable computer vision microservice repo that processes video with DeepStream and publishes detection metadata to a VSS Kafka broker.
Can the scaffold run live without model assets?
No. Execution of deepstream-app and kafka_smoketest.py requires an actual ONNX model, a matching labels file, and a custom shared library containing NvDsInferParseYolo26.