What does this skill do, and when should you use it?
This skill guides pip-based installation of the Holoscan SDK Python bindings using holoscan-cu12 or holoscan-cu13 selected for the host CUDA version. It requires Linux x86_64, an NVIDIA GPU and driver, the CUDA Toolkit, and Python 3.10–3.13. The workflow creates or activates a virtual environment, installs the wheel, and verifies the installation with hello_world and headless video_replayer examples. It is limited to Python installation and does not cover native C++, apt, or Conda installation.
It consults NVIDIA’s official installation documentation and can run nvidia-smi to determine the CUDA variant; checks for or creates ~/holoscan/venv, activates it, and installs holoscan-cu12 or holoscan-cu13 with pip; imports holoscan and prints its version; downloads official hello_world and video_replayer examples from the GitHub tag matching the installed version; edits the video_replayer configuration for 10 frames, non-repeating non-realtime execution, and headless mode; then runs the examples and checks for Hello World! and Graph execution finished. outputs.
- A Python developer needs Holoscan SDK bindings on a Linux x86_64 workstation with an NVIDIA GPU.
- A user has CUDA 12 or CUDA 13 and needs the matching Holoscan wheel without creating system-wide pip conflicts.
- An engineer wants to confirm that the Holoscan Python package imports and runs an official hello_world example.
- A user is validating video_replayer over SSH or on a headless server and can provide the required racerx data directory.
How do you install this skill?
- Before pip installation, verify wheel versions, hashes, and dependency sources; the instructions provide no lock or integrity-check procedure.
- Do not execute curl-fetched remote examples without reviewing them; separate download, review, and execution, and verify temporary-file provenance.
- Reachability and current requirements for PyPI, docs.nvidia.com, and GitHub were not verified statically; mainland-China users should prepare mirror or offline options.
- The optional sudo data-download command has system-level effects and requires explicit user approval and path verification.
- Shell / CLI
- Network access
- Local filesystem
Linux x86_64NVIDIA GPU and driverCUDA Toolkit 12 or 13Python 3.10–3.13PyPIdocs.nvidia.com
Install the selected skill from the NVIDIA skills collection:
npx skills add nvidia/skills --skill holoscan-install-wheel --yes
Then allow a compatible Agent Skills client to load it. The skill requires Linux x86_64, an NVIDIA GPU and driver, CUDA Toolkit 12 or 13, Python 3.10–3.13, and network access to PyPI and docs.nvidia.com.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/holoscan-install-wheel" ~/.claude/skills/
rm -rf "$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 the Holoscan SDK Python wheel in a virtual environment and verify it with hello_world.
A suitable trigger prompt is: "Install the Holoscan SDK Python wheel in a virtual environment and verify it with hello_world." The skill checks ~/holoscan/venv, creates it with python3 -m venv ~/holoscan/venv when missing, activates it, and installs the matching wheel. Before downloading and executing remote examples, the agent should tell the user that scripts will be fetched from the specified GitHub URL; if the user declines or GitHub is unreachable, the execution step is skipped. Each new shell session requires source ~/holoscan/venv/bin/activate and ulimit -s 32768.
What are this skill's strengths and limitations?
- Provides explicit selection rules for CUDA 12 and CUDA 13 wheel packages.
- Requires a virtual environment, reducing system Python package conflicts, especially on Ubuntu 24.04.
- Includes import, hello_world, and headless video_replayer verification steps.
- Documents common CUDA mismatch, stack-size warning, and missing racerx data failures.
- It only covers Python wheel installation and does not install C++ headers or libraries.
- video_replayer requires racerx data; the supplied alternative retrieval method still depends on an installed Debian package script.
- It requires Linux x86_64, an NVIDIA GPU, the CUDA Toolkit, and network access; other platforms are not covered.
- The source does not document a separate automated test suite; verification primarily relies on running examples.
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 Python Wheel Installer this page ✓ NVIDIA · Official | 50 · 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 |
| Holoscan Native Ubuntu Installer ✓ NVIDIA · Official | 52 · 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 |
Compared with the /holoscan-install-debian skill named in the source, this skill installs only the Python wheel for Python workflows; users needing C++ headers and libraries should use the Debian installation skill.
How did FollowSkills review this skill?
The skill scopes use to Linux, NVIDIA GPU, virtual environments, and Python, and requires user confirmation before downloading and executing GitHub examples. However, pip versions and hashes are not pinned, downloaded scripts are executed directly, and an optional sudo command is included without complete data-flow, dependency-integrity, pre-install confirmation, or rollback guidance, so points are deducted.
The sequence, CUDA variant mapping, verification commands, and troubleshooting are reasonably clear. Static review cannot reproduce execution; there are no attached tests covering the skill's key paths, remote documentation may change, and failure feedback for abnormal inputs and installation errors remains incomplete, so the score is constrained and reduced.
The audience, Python-only scope, exclusions for C++/apt and Conda, and hardware/CUDA prerequisites are explicit. However, it supports only Linux x86_64 systems with NVIDIA GPUs, does not describe Chinese-language interaction support, and depends on PyPI, NVIDIA documentation, and GitHub whose mainland-China reachability is unverified, so points are deducted.
The material provides name, version, license, author, tags, purpose, prerequisites, limitations, ordered instructions, verification, and troubleshooting. It lacks the recommended Instructions and Examples sections, a changelog, explicit maintenance ownership and update path, and stable local version details for externally changing requirements, so points are deducted.
The workflow covers venv creation, wheel selection, installation, and basic verification, and the supplied evaluation files report positive and negative routing tasks with strong correctness results. Those results were not executed or independently verified here; video_replayer also requires extra data and a Debian package, and users must adapt commands to their environment, so full marks are not justified.
Official documentation and repository references, expected outputs, evaluation tasks, and a static evaluation report provide limited auditability. There is no committed CI or test suite tied to the selected skill's key paths, and the report alone does not establish reproducibility for this revision, so points are deducted.
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 →