What does this skill do, and when should you use it?
This skill helps users install the NVIDIA Holoscan SDK C++ runtime and headers natively on Ubuntu. It inspects the operating system, NVIDIA driver, CUDA version, and existing packages to select the appropriate Holoscan apt package. It then validates the installation with hello_world, tensor_interop, and headless video_replayer examples. Its scope is limited to Ubuntu-based C++ apt installation and does not include Python bindings.
Reads the Ubuntu release and nvidia-smi output; checks cuda-keyring, installed Holoscan packages, and apt visibility; selects holoscan-cuda-12, holoscan-cuda-13, or holoscan according to the driver and platform; downloads and installs cuda-keyring when needed; runs apt-get update and apt-get install; sets LD_LIBRARY_PATH, HOLOSCAN_INPUT_PATH, and the stack limit; runs three C++ examples and downloads racerx data when required; finally provides a reusable shell environment snippet.
- A developer needs to install the Holoscan C++ SDK on an Ubuntu 22.04 or 24.04 NVIDIA GPU host.
- A Jetson or IGX user needs to determine the correct Holoscan apt package variant.
- A team needs post-install validation of CUDA compatibility, shared libraries, and example data.
- An operator needs to troubleshoot driver-version errors, missing packages, missing libraries, or example crashes.
How do you install this skill?
- Obtain confirmation before using sudo, modifying APT sources, downloading external artifacts, downloading example data, or removing a CUDA package.
- The cuda-keyring URL, package versions, and architecture are not pinned or verified; Ubuntu 22.04, ARM64, IGX, and x86_64 paths require per-host validation against the official documentation.
- Core dependencies are NVIDIA documentation and download services; no mainland-China mirror, proxy, or offline procedure is provided.
- Verification writes /tmp/vr.yaml and may download the racerx dataset, but cleanup, rollback, and recovery steps after partial failure are not specified.
- Shell / CLI
- Network access
- Local filesystem
Ubuntu 22.04 or 24.04NVIDIA GPU with working driversudoapt/dpkgwgetPython 3
Install this skill from the NVIDIA skills collection:
npx skills add nvidia/skills --skill holoscan-install-debian --yes
The skill becomes available to the agent for relevant tasks. The README does not document a manual destination for copying the skill folder.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/holoscan-install-debian" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Install and verify the Holoscan SDK C++ environment from apt on this Ubuntu NVIDIA GPU host.
Ask an agent: “Install and verify the Holoscan SDK C++ environment from apt on this Ubuntu NVIDIA GPU host.” The agent should consult NVIDIA’s official installation documentation first, run lsb_release and nvidia-smi, inspect cuda-keyring and package state, install the correct variant, and run hello_world, tensor_interop, and headless video_replayer. For Python support, use or pair it with /holoscan-install-wheel.
What are this skill's strengths and limitations?
- Covers environment detection, package selection, installation, and example-based verification.
- Distinguishes CUDA 12, CUDA 13, IGX, and nvgpu cases.
- Includes concrete troubleshooting for shared libraries, stack size, example data, and headless video execution.
- Supports only Ubuntu 22.04/24.04; other distributions require a container or wheel installation.
- The apt installation does not provide Python bindings.
- Requires an NVIDIA GPU, working driver, sudo, network access, and a local shell.
- The source provides no separate test suite or evidence beyond the documented platform scope.
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Holoscan Native Ubuntu Installer this page ✓ NVIDIA · Official | 52 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Holoscan Python Wheel Installer ✓ NVIDIA · Official | 50 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Holoscan SDK Source Builder ✓ NVIDIA · Official | 49 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Holoscan Conda Installer ✓ NVIDIA · Official | 52 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Holoscan SDK Setup Guide ✓ NVIDIA · Official | 55 · Use with care | ★ 3.8k | today | Apache-2.0 |
The source explicitly identifies container and wheel installation as alternative paths and recommends /holoscan-install-wheel when Python is needed; it does not provide a comparison with other products.
How did FollowSkills review this skill?
The workflow transparently discloses network access, sudo, APT installation, example-data download, and environment changes, but it does not require confirmation before privileged or external operations. The keyring, APT packages, and data are not version-pinned or checksum-verified, and the CUDA-variant removal path lacks a complete rollback plan, so points are deducted.
The steps, prerequisite checks, verification commands, and troubleshooting guidance are broadly consistent, with a CUDA-variant selection rule. However, coverage of Ubuntu 22.04/24.04 and x86_64/ARM64 is incomplete because the download example is hard-coded to Ubuntu 24.04 x86_64; IGX detection is not reproducible; and abnormal-input handling and failure feedback remain thin. Static calibration therefore keeps this at or below 10.
The audience and scenario are clear: Ubuntu systems with NVIDIA GPUs needing a C++ APT installation. Python, other distributions, and CUDA-variant non-fit boundaries are stated, and the negative Python trigger is explicit. Chinese-language interaction is not addressed, and reachability of the NVIDIA documentation, repository, and download host from mainland China is unspecified, so points are deducted.
The skill has readable frontmatter, version, Apache-2.0 license, author, tags, purpose, limitations, installation stages, and troubleshooting, with some maintenance-source signals. It lacks separate recommended Instructions/Examples sections, a changelog, explicit maintenance ownership and update procedure, and its skill-card licensing metadata is not fully aligned with the repository metadata, so points are deducted.
The stated goal is concrete: install the C++ runtime and headers and run bundled examples. Commands and expected outputs are directly usable, and the skill clearly excludes Python APT installation. Static review cannot confirm that current package names, paths, example configuration, and commands work; the evaluation report is limited author-supplied evidence rather than independent reproduction, so points are deducted.
Official installation and product-repository references, version information, concrete checks, and expected outputs provide some auditability. However, there is no committed test suite or CI execution evidence covering this skill's key paths, and BENCHMARK.md is a report rather than independently reproducible evidence for this review. Static calibration therefore keeps this at or below 5.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →