NVIDIA Agent Skills

Official, NVIDIA-verified skills for AI agents.

---

Skills are portable instruction sets that teach AI agents how to use NVIDIA CUDA-X libraries, AI Blueprints, and platform tools correctly. From solving vehicle routing problems with cuOpt, to deploying RAG pipelines, to onboarding models into TensorRT-LLM, every skill listed here is published and verified by NVIDIA.

This repository is a catalog — skills are maintained in their respective product repos and mirrored here daily via an automated sync pipeline. It follows the open Agent Skills specification, making skills compatible with any AI agent or framework that supports the standard.

We are building this in the open. The catalog, sync pipeline, and roadmap are public so the community can see what we ship, what we plan to ship, and how we secure skills for use. As we add signing, scanning, and evaluation, each improvement lands here incrementally.

---

Quickstart

Use the skills CLI to install NVIDIA skills into the AI agent you use. The CLI runs through npx, so you do not need to clone this repo or copy skill folders by hand.

# Browse the NVIDIA catalog without installing anything
npx skills add nvidia/skills --list

# Install from the NVIDIA catalog (interactive)
npx skills add nvidia/skills

# Install one skill into the current project and skip prompts
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api-python --yes

That's it — the skill activates automatically the next time your agent encounters a relevant task. For example, ask your agent to "solve a linear programming problem with cuOpt" and the skill guides it through the cuOpt Python API.

The CLI can prompt you to choose where to install the skill, or you can target agents explicitly using a --agent option, e.g. --agent claude-code --agent codex --agent cursor.

For non-interactive installs, global installs, agent-specific installs, updates, removals, and fallback manual copying, see Advanced installation.

---

Available Skills

Product	Description	Skills	Catalog	Source	Version
CUDA-Q	CUDA Quantum — onboarding guide for installation, test programs, GPU simulation, QPU hardware, and quantum applications.	1	skills/CUDA-Q/	Source	e47b30b · 2026-05-08
cuOpt	GPU-accelerated optimization — vehicle routing, linear programming, quadratic programming, installation, server deployment, and developer tools.	15	skills/cuopt/	Source	f3ab099 · 2026-05-08
Megatron-Bridge	Bridge between NeMo and Megatron — data processing, model conversion, and training utilities.	27	skills/Megatron-Bridge/	Source	7c0d655 · 2026-05-08
Megatron-Core	Large-scale distributed training — model parallelism, pipeline parallelism, and mixed precision.	11	skills/Megatron-Core/	Source	dfd4a1a · 2026-05-08
Model-Optimizer	Model optimization — quantization, sparsity, and distillation for efficient inference.	8	skills/Model-Optimizer/	Source	e2d29c8 · 2026-05-08
NeMo Evaluator	LLM evaluation — launch evaluations, access MLflow results, NeMo Evaluator Launcher assistant, and bring-your-own benchmarks.	4	skills/NeMo-Evaluator-Launcher/	Source	7a44667 · 2026-05-08
NeMo Gym	RL training environments — add benchmarks, resources servers, agent wiring, and reward profiling.	2	skills/NeMo-Gym/	Source	ed190cd · 2026-05-08
NemoClaw	Secure agent sandboxing — run OpenClaw inside NVIDIA OpenShell with managed inference, policy management, remote deployment, sandbox monitoring, and contributor/maintainer workflows.	24	skills/NemoClaw/	Source	97f8a9a · 2026-05-08
Nemotron Voice Agent	Real-time conversational AI — deploy speech-to-speech voice agents on Workstation, Jetson Thor, or Cloud NIMs.	1	skills/nemotron-voice-agent/	Source	a87826c · 2026-04-22
RAG Blueprint	RAG pipeline — deploy, configure, troubleshoot, and manage retrieval augmented generation with Docker Compose or Helm.	1	skills/rag/	Source	56d3c61 · 2026-05-05
TensorRT-LLM	LLM inference optimization — model onboarding, performance analysis and optimization, kernel writing, CI diagnostics, code contribution, and codebase exploration.	24	skills/TensorRT-LLM/	Source	c87f002 · 2026-05-08
TileGym	Tile-based GPU programming — adding new kernels, cross-framework conversion, and performance optimization.	7	skills/TileGym/	Source	ed6f7cd · 2026-05-08
Video Search and Summarization	VSS Blueprint — deploy profiles, search and summarize video, generate analysis reports, manage alerts and incidents, query VIOS sensors, and use the RTVI VLM microservice.	10	skills/video-search-and-summarization/	Source	1fc4513 · 2026-05-07

---

Getting Help & Contributing

For skill-related issues, feature requests, new skill ideas, discussions, and contributions — use the source repo for the relevant product:

Product	Issues	Discussions	Contributing	Security
CUDA-Q	Issues	Discussions	Contributing	Security
cuOpt	Issues	Discussions	Contributing	Security
Megatron-Bridge	Issues	Discussions	Contributing	Security
Megatron-Core	Issues	Discussions	Contributing	Security
Model-Optimizer	Issues	Discussions	Contributing	Security
NeMo Evaluator	Issues	Discussions	Contributing	Security
NeMo Gym	Issues	Discussions	Contributing	Security
NemoClaw	Issues	Discussions	Contributing	Security
Nemotron Voice Agent	Issues	Discussions	Contributing	Security
RAG Blueprint	Issues	Discussions	Contributing	Security
TensorRT-LLM	Issues	Discussions	Contributing	Security
TileGym	Issues	Discussions	Contributing	Security
Video Search and Summarization	Issues	Discussions	Contributing	Security

For issues with this catalog repo itself (README, structure, listing a new product): open an issue here.

---

Roadmap

• ✅ Public skills catalog with NVIDIA-verified skills across multiple products
• ✅ Automated sync pipeline with skills mirrored from product repos daily
• ✅ Security scanning for all published skills covering instruction safety and supply-chain integrity
• 🔲 Skills signing so every published skill carries a verifiable NVIDIA signature
• 🔲 Skills universal evaluation criteria and task-specific criteria
• 🔲 Skills Card with machine-readable metadata for identity, provenance, quality, and behavioral boundaries
• 🔲 Compliance gates before external publication
• 🔲 Syndication to external marketplaces and MCP hubs

---

Repository Structure

NVIDIA/skills/
├── skills/                  # All skills, mirrored daily from product repos
│   ├── README.md             # Install guidance for people browsing this folder directly
│   ├── CUDA-Q/               # CUDA-Q skills
│   ├── cuopt/                # cuOpt skills
│   ├── Megatron-Bridge/      # Megatron-Bridge skills
│   ├── Megatron-Core/        # Megatron-Core skills
│   ├── Model-Optimizer/      # Model-Optimizer skills
│   ├── NeMo-Evaluator/       # NeMo Evaluator skills
│   ├── NeMo-Evaluator-Launcher/
│   ├── NeMo-Gym/             # NeMo Gym skills 
│   ├── NemoClaw/             # NemoClaw skills 
│   ├── nemotron-voice-agent/ # Nemotron Voice Agent skills
│   ├── TensorRT-LLM/         # TensorRT-LLM skills 
│   └── ...
├── components.d/            # Product registry — one file per component, teams onboard here
│   ├── cuda-q.yml
│   ├── cuopt.yml
│   ├── megatron-bridge.yml
│   ├── ...
│   ├── tensorrt-llm.yml
│   └── README.md             # Schema and onboarding instructions
├── docs/
│   └── advanced-install.md   # Advanced skills CLI usage
├── .github/workflows/       # Automated sync pipeline
├── CONTRIBUTING.md          # Contribution guidelines
├── SECURITY.md              # Security reporting policy
├── CODE_OF_CONDUCT.md       # Community code of conduct
└── LICENSE                  # Apache 2.0

Skills are maintained in their respective product repos (see the Source column in Available Skills) and automatically synced to this repo daily.

---

Standards & Compatibility

This repository adheres to the Agent Skills specification:

• Skills are portable directories with a SKILL.md file at their root.
• Metadata uses YAML frontmatter with required name and description fields.
• Skills follow a progressive disclosure model — lightweight metadata loads at startup, full instructions load on activation.
• Validate your skill using the skills-ref reference library.

---

License

This project is dual-licensed under the Apache License 2.0 and Creative Commons Attribution 4.0 International (CC BY 4.0).