Overview
Agent Skills are portable instruction sets that teach an AI agent how to use a piece of software properly. NVIDIA Agent Skills is NVIDIA's official, verified catalog of them: skills for Physical AI and robotics workflows, simulation, CUDA-X libraries, RAG and AI Blueprints, and platform tools. Instead of hoping a model remembers the right cuOpt or BioNeMo API, you install the skill and the agent follows NVIDIA's own guidance for that product.
The repository is a catalog rather than the source of truth: each skill is maintained in its respective NVIDIA product repo and mirrored here daily by an automated sync pipeline. That matters for staleness — skills are added and revised continuously, and the catalog reflects the product teams' current guidance rather than a snapshot someone wrote once. The project is developed in the open, with a published roadmap and contributions welcome.
Installation goes through the standard `skills` CLI, so the same command works across agents: Claude Code, Codex, Cursor, Kiro, Snowflake CoCo and the other clients the Agent Skills specification supports. You can browse the catalog, install one skill non-interactively by name, target several agents at once, and refresh everything later — the interactive update flow also flags skills that were removed or consolidated upstream and offers to clean up the stale local copies.
What it does
- Official, NVIDIA-verified skills — CUDA-X libraries, Physical AI and robotics, simulation, RAG and AI Blueprints, platform tools
- Catalog is mirrored daily from each product's own repository by an automated sync pipeline
- Installs through the standard `skills` CLI, so the same command works across supported coding agents
- Per-agent installs (`--agent claude-code`, `codex`, `cursor`, `kiro-cli`, `cortex`), repeatable to install into several at once
- Non-interactive install by skill name for scripted setups (`--skill <name> --yes`)
- `npx skills update` flags removed or consolidated skills and cleans up stale local copies
- Licensed Apache-2.0 for code and CC-BY-4.0 for documentation
Getting started
Installation uses the skills CLI through npx — you do not clone this repository or copy skill folders by hand. A current CLI (v1.5.16 or newer) is required; on older versions skills may install but not appear in Claude Code.
Install a skill interactively
The CLI prompts you to choose a skill and an install destination.
npx skills add nvidia/skillsBrowse the catalog first
List what is available before installing anything.
npx skills add nvidia/skills --listInstall one skill, no prompts
Use this when you already know the skill name — for example the cuOpt optimization skill — and want it in specific agents.
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --yes
npx skills add nvidia/skills \
--skill cuopt-numerical-optimization-api \
--agent claude-code \
--agent codex \
--agent cursorUse it, then keep it fresh
The skill loads the next time your agent picks up skills — in Claude Code, /reload-skills loads it into the current session. Ask for the task in natural language ("solve a linear programming problem with cuOpt") and the skill guides the agent through the API. Check for and apply updates as the catalog evolves.
npx skills list
npx skills check
npx skills updateCommands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Reach for it when an agent has to write CUDA-X, cuOpt or BioNeMo code and you want NVIDIA's own API guidance in context
- Reach for it for robotics and simulation workflows where the agent must follow a vendor-specific end-to-end procedure
- Reach for it to standardise a team's agent setup — the same skills installed the same way across Claude Code, Codex and Cursor
- Reach for it when you would otherwise paste NVIDIA documentation into the prompt every session
How NVIDIA Agent Skills compares
NVIDIA Agent Skills alongside other open-source agent skills & plugins tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| Superpowers | ★ 297k | A composable skills plugin that installs a spec-first, TDD, subagent-driven development methodology into Claude Code, Codex, Cursor, Gemini CLI and other coding harnesses. |
| Skills for Real Engineers | ★ 282k | Matt Pocock's everyday agent skills for coding agents, covering alignment grilling, planning, code review and research — small, composable and meant to be edited. |
| ECC | ★ 276k | An installable plugin that adds 68 agents, 286 skills, hooks, rules, memory and an agent-config security scanner to Claude Code, Codex and other coding harnesses. |
| Karpathy Coding Guidelines | ★ 218k | A Claude Code plugin and CLAUDE.md rule set built from Andrej Karpathy's observations on LLM coding pitfalls: think before coding, keep it simple, make surgical changes, and work to verifiable success criteria. |
| Ponytail | ★ 159k | A skill and plugin that makes a coding agent take the laziest correct route — a seven-rung ladder from "does this need to exist?" to "the minimum that works" — while keeping validation, security and accessibility off the chopping block. |
| Caveman | ★ 111k | A skill plus local proxy that shortens an AI coding agent's prose while leaving code, commands and paths untouched — the rule file cuts what the agent writes, the proxy compresses what it reads. |
| Addy’s Agent Skills | ★ 104k | A pack of production engineering skills for AI coding agents, with nine lifecycle slash commands — /spec, /plan, /build, /test, /review, /ship — installable into 70+ agents. |
| NVIDIA Agent Skills | ★ 3.5k | NVIDIA-verified skills that teach coding agents to drive CUDA-X, robotics and RAG workflows |