Introduction

Incorporating coding agents into software development has become commonplace. Using NVIDIA-Verified Agent Skills (hereinafter referred to as NVIDIA Skills), which we will introduce here, you can teach coding agents how to correctly use NVIDIA CUDA-X libraries, NVIDIA Blueprints, NVIDIA Omniverse™ and physical AI workflows, NVIDIA NeMo™ training and inference tools, and other platform components. This article will demonstrate how to apply NVIDIA Skills to the installation of an AI agent called NVIDIA NemoClaw™.

Operating environment

The versions of the software used to create this article are as follows:

For hardware, we used an MSI EdgeXpert 4TB drive, which is NVIDIA® DGX Spark™ compatible and features an NVIDIA GB10 Grace Blackwell Superchip.

Agent Skills

Agent Skills is a standardized way to give AI agents new capabilities and expertise. Agent Skills transform generic agents into experts. Unlike including expertise in prompts, Agent Skills load on demand, eliminating the need to repeatedly provide the same guidance across multiple conversations. Agent Skills are essentially folders containing Markdown files named SKILL.md that provide detailed instructions for the AI agent, and can also include scripts, reference materials, templates, and more.

　
Agent Skills was originally developed by Anthropic, but is now released as an open standard and supported by numerous AI tools and agent clients. The Agent Skills specification is available on the Agent Skills website.

NVIDIA Skills

NVIDIA Agent Skills is a leading open-source collection of physical AI agent skills and tools for robotics, autonomous driving vehicles, vision AI, and industrial digital twins, spanning NVIDIA Omniverse, NVIDIA Cosmos™, NVIDIA Alpamayo, and NVIDIA Metropolis, compliant with Agent Skills standards. (Source: NVIDIA Announces Leading Collection of Open-Source Agent Tools and Skills for Physical AI - NVIDIA | Japan Blog)

NVIDIA Skills allows agents to leverage NVIDIA libraries, models, and frameworks to accelerate data generation, simulation, training, evaluation, and deployment pipelines in robotics, AV, factories, and labs. As of June 2026, NVIDIA Skills covers NVIDIA software products in the following areas:

• Robotics and Edge AI

• autonomous driving cars

• Real-time vision AI agent

• Industrial AI

• Healthcare

npx skills add nvidia/skills

◆  Select skills to install (space to toggle)
│  ...
│  ◻ nemo-mbridge-perf-hierarchical-context-parallel
│  ◻ nemo-mbridge-perf-megatron-fsdp
│  ◻ nemo-mbridge-perf-memory-tuning
│  ◻ nemo-mbridge-perf-moe-comm-overlap
│  ◻ nemo-mbridge-perf-moe-dispatcher-selection
│  ◻ nemo-mbridge-perf-moe-hardware-configs
│  ◻ nemo-mbridge-perf-moe-long-context
│  ◻ nemo-mbridge-perf-moe-optimization-workflow
│  ◻ nemo-mbridge-perf-moe-vlm-training
│  ◻ nemo-mbridge-perf-parallelism-strategies
│  ◻ nemo-mbridge-perf-sequence-packing
│  ◻ nemo-mbridge-perf-tp-dp-comm-overlap
│  ◻ nemo-mbridge-recipe-recommender
│  ◻ nemo-mbridge-resiliency
│  ◻ nemo-retriever
│  ◻ nemo-rl-auto-research
│  ◻ nemo-rl-brev-etiquette
│  ◻ nemo-rl-docs
│  ◻ nemo-rl-session-memory
│  ◼ nemoclaw-user-agent-skills (Describes the agent skills shipped with NemoClaw and how ...)
│  ◼ nemoclaw-user-configure-inference (Connects NemoClaw to a local inference server. Use when s...)
│  ◼ nemoclaw-user-configure-security (Presents a risk framework for every configurable security...)
│  ◼ nemoclaw-user-deploy-remote (Explains how to run NemoClaw on a remote GPU instance, in...)
│  ◼ nemoclaw-user-get-started (Installs NemoClaw, launches a sandbox, and runs the first...)
│  ◼ nemoclaw-user-manage-policy (Adds, removes, or modifies allowed endpoints in the sandb...)
│  ◼ nemoclaw-user-manage-sandboxes (Explains operational tasks after the quickstart: listing ...)
│  ◼ nemoclaw-user-monitor-sandbox (Inspects sandbox health, traces agent behavior, and diagn...)
│  ◼ nemoclaw-user-overview (Explains how OpenClaw, OpenShell, and NemoClaw form the e...)
│  ◼ nemoclaw-user-reference (Describes the NemoClaw plugin and blueprint architecture ...)
│  ◻ nemotron-customize
│  ◻ nemotron-policy-generator
│  ◻ nemotron-retrieval-recipes
│  ◻ nemotron-speech
│  ◻ nv-generate-ct-rflow
│  ◻ nv-generate-mr (Used for generating synthetic body MRI volumes with NV-Ge...)
│  ◻ nv-generate-mr-brain
│  ...
└

Install NemoClaw

With everything ready, I started installing NemoClaw using NVIDIA Skills. Since NemoClaw itself is an AI agent, this resulted in an interesting structure where one AI agent (GitHub Copilot) requests the installation of another AI agent (NemoClaw).

　
First, I checked whether GitHub Copilot recognized NVIDIA Agent Skills.

ワークスペースにインストールされているスキルをリストしてください。

The input was correctly recognized, and the following response was obtained. Note that the wording of the response may vary depending on the AI model used by GitHub Copilot. Even with the same AI model, the wording of the response should change slightly each time you try it.

　
--- (↓ Below is the output from GitHub Copilot) ---

--- (↑ Output from GitHub Copilot) ---

　
The installation method for NemoClaw is described in NVIDIA Skills, so I requested GitHub Copilot to install NemoClaw with the following simple instructions.

スキルを参照の上、NemoClawをインストールしてください。今まで、NemoClawを使ったことがないので、インストール中のオプション選択もサポートしてください。

It seems that explicitly instructing GitHub Copilot to use skills yields better results. Without this, GitHub Copilot sometimes had to search the internet for instructions on how to install NemoClaw. Furthermore, it was important to instruct it to support option selection, which allowed GitHub Copilot to explain the meaning of the options.

　
GitHub Copilot loaded the nemoclaw-user-get-started skill and first checked whether the system met the prerequisites.

　
--- (↓ Below is the output from GitHub Copilot) ---

前提条件を確認するため、詳細を読み込みます。 Read skill `nemoclaw-user-get-started`: [](file:///home/msispark/nemoclaw- projects/.agents/skills/nemoclaw-user-get-started/references/prerequisites.md) 前提条件を確認しました。Linuxをお使いなので、インストーラーがDockerとNode.jsを必要に応じてインストールします。 まず、現在のシステム状態を確認します： Ran terminal command: docker --version && node --version && npm --version && df -h / | tail -1 | awk '{print "Disk free:", $4}' && free -h | grep Mem | awk '{print "RAM:", $2}' システム要件を満たしています： - Docker: 29.2.1 ✓ - Node.js: v26.2.0 ✓ - npm: 11.13.0 ✓ - ディスク: 2.0TB空き ✓ - RAM: 121GB ✓

--- (↑ Output from GitHub Copilot) ---

　
Once the conditions were confirmed to be met, the installation process began.

--- (↓ Below is the output from GitHub Copilot) ---

NemoClawのインストールを開始します。インストーラーは対話式なので、各ステップで選択をサポートします： Ran terminal command: curl -fsSL https://www.nvidia.com/nemoclaw.sh | bash インストーラーが起動しました。サードパーティソフトウェアの注意事項への同意を求めています。 内容の要約： - NemoClawはApache 2.0ライセンス - サードパーティソフトウェア（OpenClaw、OpenShell等）を自動的に取得・使用する - それらのソフトウェアは別の利用規約に従う - 「AS IS」（現状のまま）で提供され、保証なし 回答を分析しています... ターミナルに `yes` を送信する

--- (↑ Output from GitHub Copilot) ---

　
I'll omit some of the details to keep it brief, but the installation was completed as follows:

　
--- (↓ Below is the output from GitHub Copilot) ---

--- (↑ Output from GitHub Copilot) ---

Started using NemoClaw

After the installation was complete, I chatted with NemoClaw to confirm that it was working correctly.

ブラウザでチャットしたいです

--- (↓ Below is the output from GitHub Copilot) ---

ブラウザー を開きました

--- (↑ Output from GitHub Copilot) ---

The dashboard launched in the browser, but a page requesting authentication information appeared. GitHub Copilot noticed that authentication was required for the chat with NemoClaw and, based on information from NVIDIA Agent Skills, suggested a NemoClaw command to display a URL containing the authentication token.

--- (↓ Below is the output from GitHub Copilot) ---

--- (↑ Output from GitHub Copilot) ---

I was able to successfully start a chat with NemoClaw.

Summary

How was it? Installing software for the first time can be daunting, as you may not understand the meaning of the options presented in the configuration settings. However, by using NVIDIA Skills, you can install NemoClaw with confidence while confirming the meaning of each option. It is also very important that the information is official information provided by NVIDIA, rather than unreliable information scattered across the internet. This time, we covered skills useful in the initial phase of using NemoClaw, such as installation and dashboard display, but there are also skills useful in the operational phase, such as configuring NemoClaw's security policy. Furthermore, many skills are offered, not limited to NemoClaw, but mainly for NVIDIA's generation AI-related software products. You can check the currently offered skills in the NVIDIA Skills GitHub repository. Please make use of them in your own software development.

Reference URL

Click here for inquiries