What does this skill do, and when should you use it?
Omniverse Realtime Viewer is a top-level routing skill for USD viewer application requests. It selects local, streaming, or other focused reference paths based on delivery and feature requirements, then guides rendering, input, camera, selection, UI, and validation work. All USD and 3D rendering must use ovrtx; browser applications display an ovstream WebRTC video stream plus UI. It suits developers building NVIDIA Omniverse USD viewers, but the required runtime depends on the selected path.
Classifies viewer requests and reads the required routing, convention, recipe, capability, and validation references. It guides ovrtx-based USD scene loading and rendering, ovstream-based browser streaming, and implementation of cameras, input, selection, hierarchy, properties, render settings, UI, packaging, and deployment. It requires viewer-owned state to remain in session layers, composite layers, or application state, preserves user USD files, and captures build, run, and interaction validation evidence before handoff.
- A frontend developer building a browser-based USD viewer that uses WebRTC streaming and UI instead of a client-side 3D renderer.
- A C++, Python, Tauri, or Electron developer building a local workstation viewer with the appropriate local delivery path.
- An application team adding camera controls, object picking, hierarchy browsing, property panels, or render settings to a USD viewer.
- An engineer deploying a viewer on cloud GPUs and choosing a streaming architecture and deployment path.
How do you install this skill?
- Shell and Write are declared without per-action confirmation, least-privilege, rollback, or external-side-effect controls; example credentials in cloud-asset references must not be used in real environments.
- The BENCHMARK reports an overall FAIL, unavailable Tier 3 evaluation, duplicate content, and PII/GPS findings; findings should be remediated and the evaluation rerun before publication.
- Core paths depend on NVIDIA GPUs, ovrtx, and related runtimes; Chinese-language support and mainland-China network reachability are not documented and require separate validation.
- Shell / CLI
- Network access
- Local filesystem
NVIDIA GPU or compatible cloud GPU deploymentovrtxovstreamOpenUSDPythonNode.js/ReactTauri, Electron, or C++ depending on viewer path
Use the repository README's supported Skills CLI command:
npx skills add nvidia/skills --skill omniverse-realtime-viewer --yes
The README says the skill becomes available the next time the agent loads skills and encounters a relevant task; the CLI determines the installation destination.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/omniverse-realtime-viewer" ~/.claude/skills/
rm -rf "$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:
- Build a browser-based Omniverse USD viewer with camera controls, object selection, and a properties panel.
Give the agent a specific viewer request, for example: "Build a browser-based Omniverse USD viewer with camera controls, object selection, and a properties panel." The agent should classify the delivery path, read the relevant references, and proceed through rendering, input, camera, selection, UI, and validation. The supplied SKILL.md does not document a complete standalone run, build, or deployment command; those details are in the selected references.
What are this skill's strengths and limitations?
- Enforces ovrtx for USD and 3D rendering, keeping implementations aligned with the stated architecture.
- Routes across browser streaming, local viewer, Tauri, Electron, C++, and multiple UI paths.
- Defines ownership and safety rules for rendering, stage mutation, user USD files, and validation.
- The repository provides CLI installation and ongoing synchronization for NVIDIA skills.
- The SKILL.md is a routing entry point rather than a complete implementation recipe for every path.
- GPU, Omniverse runtime, OpenUSD, and language or framework dependencies vary by viewer path.
- The supplied material does not provide a skill-specific runnable example, test result, or fixed deployment configuration.
- Browser paths explicitly cannot use WebGL, Three.js, or other client-side USD 3D renderers.
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 |
|---|---|---|---|---|
| Omniverse Realtime Viewer this page ✓ NVIDIA · Official | 55 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Omniverse USD Performance Tuning ✓ NVIDIA · Official | 52 · Use with care | ★ 3.8k | today | Apache-2.0 |
| Code Clarity Refactor | 57 · Use with care | ★ 111k | 15d ago | MIT |
| Codex CLI Builder ✓ OpenAI · Official | 55 · Use with care | ★ 28k | today | — |
| MCP Server Builder Guide | 50 · Use with care | ★ 3.1k | 8mo ago | MIT |
Compared with browser-side WebGL, Three.js, Babylon.js, PlayCanvas, A-Frame, model-viewer, react-three-fiber, or glTF viewer approaches, this skill requires browsers to display an ovstream WebRTC video stream rather than render USD geometry client-side.
How did FollowSkills review this skill?
The document explicitly requires ovrtx, preserving user USD files, separating session state, and avoiding browser-side renderers; it also provides some error fallbacks and state constraints, supporting partial data-flow transparency and least privilege. Points are deducted because Shell and Write are declared without explicit user confirmation, permission boundaries, rollback procedures, or sensitive-credential handling. The cloud-asset references also show credentialed and external object-storage operations without stronger isolation or authorization guidance.
Routing, read order, non-negotiable rules, and extensive abnormal-input guidance make the main happy paths reasonably intelligible, with failure feedback for AOV, input, and camera flows. Points are deducted because this is a static review and cannot establish dependency availability, key-path reproduction, or executable correctness; the BENCHMARK reports an overall FAIL, no available Tier 3 evaluation, and many duplicate-content findings plus 32 Tier 1 findings, so the static ceiling applies.
The skill clearly targets Omniverse Realtime Viewer work and defines delivery paths, input, camera, AOV, UI, and automation scenarios; the eval cases cover browser, desktop, Electron, Tauri, and headless requests. Points are deducted because GPU, ovrtx, ovstream, and runtime requirements are substantial, non-fit boundaries and compatibility conditions are distributed across downstream references, and there is no Chinese-language or mainland-China network adaptation guidance. Some cloud-asset workflows also depend on external S3 services.
SKILL.md provides clear routing, read order, reference families, workflow, and completion checklists; metadata includes version, Apache-2.0 licensing, author, and tags, while the repository README describes synchronization and update mechanisms. Points are deducted because maintenance ownership is directed generally to product repositories without a specific maintainer, changelog, or stable-version policy for this skill; the BENCHMARK reports repeated content that increases maintenance and progressive-disclosure burden.
The objective, architecture constraints, reference selection, and representative eval expectations are concrete enough to plausibly help an agent build an RTX USD viewer, while explicitly prohibiting browser-side USD rendering and source-USD mutation. Points are deducted because the supplied files contain no verifiable representative generated outputs or third-party execution evidence; the BENCHMARK has no Tier 3 results and an overall FAIL, so only the static ceiling for core-task plausibility is justified.
The fixed revision, committed evals, validation reference, and BENCHMARK provide limited audit trails; the report also discloses unavailable Tier 3 evidence, static findings, and deduplication findings. Points are deducted because there are no execution logs, complete key-path test suites, CI coverage, or independent corroboration, and important evaluation signals are missing, preventing a higher static verifiability score.
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 →