NVIDIA/skills

Agent Skills for NVIDIA products — install into Claude Code, Codex, and other coding agents to run Physical AI, robotics, simulation, CUDA, and RAG workflows end to end.

NVIDIA Agent Skills

What it is – A catalog of agent skills – small, portable instruction sets that teach AI coding agents (Claude Code, Codex, Cursor, etc.) how to use NVIDIA software correctly. Each skill wraps the best‑practice workflow for a specific NVIDIA product (cuDF, cuOpt, DeepStream, Jetson, NeMo, DOCA, etc.) and can be installed into an agent with a single CLI command.

Why it matters – As AI agents become capable of writing and running code, they need reliable, up‑to‑date guidance for complex, GPU‑accelerated toolchains. NVIDIA Agent Skills provide that governance layer, letting agents automatically:

  • Install the right libraries (e.g., cuOpt, cuDF, cuPyNumeric).
  • Follow NVIDIA‑approved setup, deployment, and debugging steps.
  • Access product‑specific APIs without the user having to search documentation.

How to get it – The repository itself is a mirror; the actual skill files live in the individual product repos and are synced daily. To add skills to an agent you use the open‑source skills CLI (maintained by Vercel Labs):

# Interactive install – pick a skill from a menu
npx skills add nvidia/skills

# Non‑interactive install of a known skill
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --yes

# Target a specific agent (Claude Code, Codex, Cursor, etc.)
npx skills add nvidia/skills \
  --skill cuopt-numerical-optimization-api \
  --agent claude-code

The CLI also supports bulk installs, updates (npx skills update), listing (npx skills list), and troubleshooting.

What’s inside – The Skill Catalog (auto‑generated table) lists every product and the associated skill directories. A few highlights:

  • AIQ – Deploy AI‑Q services and run research workflows.
  • cuDF – Accelerated DataFrame operations, multi‑GPU joins, I/O, etc.
  • cuOpt – GPU‑accelerated linear/quadratic programming and routing APIs.
  • DeepStream – End‑to‑end video analytics pipelines, model import, calibration.
  • Jetson BSP / Device – Build, flash, and benchmark Jetson boards, including LLM serving and speculative decoding.
  • NeMo AutoModel / Fabric / MBridge – Distributed training recipes for LLMs/VLMs, checkpoint conversion, performance tuning.
  • DOCA – Programming guide and deployment steps for NVIDIA BlueField DPUs and ConnectX NICs.
  • Medical AI – MONAI‑based DICOM handling, segmentation, and evidence export.
  • …and many more covering robotics, quantum (CUDA‑Q), health, digital health, and more.

Typical workflow

  1. Pick a skill – e.g., cuopt-numerical-optimization-api.
  2. Install – npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent claude-code.
  3. Reload – In Claude Code run /reload-skills (or restart the agent).
  4. Ask the agent – “Solve a linear programming problem with cuOpt.” The agent follows the skill, handling installation, API usage, and result interpretation automatically.

Who can contribute – The repo is open‑source; contributions are welcomed via pull requests. The roadmap (linked in the README) outlines upcoming product integrations and improvements to the sync pipeline.

Documentation & resources

License – Dual‑licensed under Apache‑2.0 and CC‑BY‑4.0 (see the LICENSE badge).

TL;DR – NVIDIA Agent Skills is a real, NVIDIA‑maintained collection of ready‑to‑install “how‑to” modules that let AI coding agents reliably work with NVIDIA’s GPU‑accelerated software stack. Install via the skills CLI, pick the product you need, and the agent will automatically follow NVIDIA‑approved best practices.

Related

  • Project
    amd/skills

    A catalog of standardized skills for AI coding agents that provide the specialized knowledge and scripts needed to deploy and optimize AI workloads on AMD hardware.

  • Project
    KKKKhazix/khazix-skills

    A collection of standardized AI agent skills (instruction sets) for goal definition, disk cleanup, deep research, and project documentation maintenance.

  • Project
    onmax/nuxt-skills

    A collection of specialized knowledge modules (skills) for AI coding assistants to improve their proficiency in Vue, Nuxt, and the NuxtHub ecosystem.

  • Project
    MoizIbnYousaf/ai-agent-skills

    AI Agent Skills is an npm‑based CLI/TUI that ships a curated library of ~115 AI‑agent “skills” (prompt‑code bundles) organized into shelves (frontend, backend, mobile, workflow, agent‑engineering, marketing). It lets you browse, search, install, and manage both local “house‑copy” skills and upstream metadata‑only skills. You can also bootstrap a personal library with `init-library`, add or catalog new skills, vendor copies, sync updates, and generate documentation. Works with any Agent‑Skills‑compatible agent (Claude, Cursor, Gemini, etc.) and supports multiple install targets. Ideal for developers or teams who want a tidy, searchable toolbox of reusable agent capabilities.

  • Project
    tiangolo/library-skills

    A tool that installs and manages official AI skills provided by libraries, ensuring AI agents use the most current and correct code patterns instead of outdated training data.