复制安装命令
用 Codex 或 Claude 安装复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它先审查 Skill 页面再帮你安装。
复制前请先查看来源、License 和安全提示。
Official, NVIDIA-verified Agent Skills for Claude Code, Codex, and other coding agents.
用 Codex 或 Claude 安装复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它先审查 Skill 页面再帮你安装。
复制前请先查看来源、License 和安全提示。
来源文件:README.md
Official, NVIDIA-verified Agent Skills for Claude Code, Codex, and other coding agents.
📖 Docs: docs.nvidia.com/skills · 📺 Livestream: From Vulnerable to Verified · 📝 Blog: NVIDIA Verified Agent Skills: Capability Governance for AI Agents
Skills are portable instruction sets that teach AI agents how to use NVIDIA software optimally: Physical AI and robotics workflows, simulation, CUDA-X libraries, RAG and AI Blueprints, and platform tools. This repository is a catalog: skills are maintained in their respective product repos, and mirrored here daily via an automated sync pipeline. Skills are being added continuously, so check back for updates. We are building this infrastructure in the open, and contributions are welcome. See the Roadmap for what is planned next.
Install NVIDIA skills with the default skills CLI flow:
npx skills add nvidia/skills
The CLI runs through npx and prompts you to choose a skill and install destination. You do not need to clone this repo or copy skill folders by hand.
Requires a current
skillsCLI (v1.5.16 or newer). Installing vianpx skills@latest add nvidia/skillsalways uses the latest. On older CLIs (v1.5.15 and earlier), skills may install but not appear in Claude Code — see Troubleshooting.
The skill is available the next time your agent loads skills and encounters a relevant task. For example, ask your agent to "solve a linear programming problem with cuOpt" and the skill guides it through the cuOpt Python API. In Claude Code, run /reload-skills to load newly installed skills in your current session.
Use this when you already know the skill name and want to skip prompts.
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --yes
Replace cuopt-numerical-optimization-api with any skill name from the Skill Catalog.
Use --agent to target a specific AI coding agent. Initially, we'll support common client targets, expanding the list over time. For the full list of clients supported by the spec, see the skills CLI Supported Agents table.
Claude Code
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent claude-code
Codex
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent codex
Snowflake CoCo
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent cortex
Cursor
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent cursor
Kiro
npx skills add nvidia/skills --skill cuopt-numerical-optimization-api --agent kiro-cli
Use --agent more than once to install the same skill into multiple agents.
npx skills add nvidia/skills \
--skill cuopt-numerical-optimization-api \
--agent claude-code \
--agent codex \
--agent cursor \
--agent kiro-cli
New skills land continuously, and existing ones are revised, renamed, or consolidated as the catalog evolves. Refresh what you have installed with:
npx skills update
Run it interactively and the CLI also flags skills that were removed or merged upstream (for example, when several skills are consolidated into one) and offers to remove the stale local copies. Use npx skills list to see what is installed and npx skills check to preview what is out of date first.
Use this when you want to see available NVIDIA skills before installing anything.
npx skills add nvidia/skills --list
For non-interactive installs, global installs, agent-specific installs, updates, removals, and fallback manual copying, see Advanced installation.
Where to file an issue depends on what's broken:
Per-product source repo links:
For issues with this catalog repo itself (README, structure, listing a new product): open an issue here.
Every published skill ships with a detached OMS signature (skill.oms.sig). The sync pipeline drops any skill missing the required artifacts before publishing, so every skill in the catalog carries:
SKILL.md — the skill instructions consumed by the agentskill-card.md — skill identity and governance cardskill.oms.sig — detached OMS signature (verifiable against nv-agent-root-cert.pem)evals/evals.json, evals/*.json, eval/*.json, or benchmark/evals.jsonBENCHMARK.md — generated benchmark report capturing verifiable uplift dataVerify a skill against the NVIDIA trust anchor nv-agent-root-cert.pem:
pip install model-signing
model_signing verify certificate SKILL_DIR \
--signature SKILL_DIR/skill.oms.sig \
--certificate_chain nv-agent-root-cert.pem \
--ignore_unsigned_files
A successful verification confirms that the skill contents have not been modified since signing by NVIDIA.
See Verify Signed Agent Skills for signature layout, the trust pipeline, and policy options.
NVIDIA/skills/
├── skills/ # NVIDIA-verified skills (count grows continuously),
│ │ synced from upstream product repos
│ ├── README.md # Browser-facing install guidance
│ ├── <product-prefix>-*/ # Flat layout — one dir per skill, product-prefixed
│ │ # e.g. aiq-*, cuopt-*, cupynumeric-*,
│ │ # dali-*, deepstream-*, dicom-*, digital-health-*,
│ │ # dynamo-*, earth2studio-*, holoscan-*, hsb-*,
│ │ # jetson-*, launch-nemo-rl, mcore-*,
│ │ # nemo-automodel-*, nemo-data-designer-plugin,
│ │ # nemo-evaluator-plugin, nemo-mbridge-* (20 skills),
│ │ # nemo-retriever, nemo-rl-* (4 skills),
│ │ # nemoclaw-user-guide, nemotron-*, nemotron-speech,
│ │ # nv-* (medical AI), physicsnemo-*, rag-*,
│ │ # skill-card-generator, tao-*, tilegym-*,
│ │ # vss-* (15 skills), accelerated-computing-cudf,
│ │ # cudaq-guide, portfolio-optimization
│ ├── omniverse-*/ # Physical AI — manually staged (see manual-components.yml)
│ └── physical-ai-*/ # Physical AI — manually staged
├── components.d/ # Product registry — one file per component, teams onboard here
│ ├── README.md # Schema and onboarding instructions
│ └── <product>.yml # one file per registered product
├── plugins/ # Packaged plugin distributions
│ └── nvidia-skills/ # Curated NVIDIA skills bundle (Claude Code, Codex)
├── plugins.d/ # Plugin build registry — config for `build-plugins.py`
│ ├── README.md
│ ├── _defaults.yml
│ └── nvidia-skills.yml
├── .claude-plugin/ # Claude Code marketplace metadata
│ └── marketplace.json
├── .agents/plugins/ # Agent marketplace metadata (other clients)
│ └── marketplace.json
├── docs/ # Long-form documentation (published via Fern)
│ ├── README.md # How to build the docs locally
│ ├── index.mdx
│ ├── advanced-install.mdx
│ ├── agent-skill-trust-pipeline.mdx
│ ├── release-checklist.mdx
│ ├── scanning-agent-skills.mdx
│ ├── signing-agent-skills.mdx
│ └── skill-cards.mdx
├── fern/ # Fern docs site configuration
├── .github/
│ ├── workflows/ # Sync pipeline, plugin validation, DCO check, author verify
│ └── scripts/ # regenerate-readme.sh, build-plugins.py,
│ # manual-components.yml (temp Physical AI catalog
│ # exception, removed after Computex 2026),
│ # marketplace/metadata.json (skill metadata sidecar)
├── nv-agent-root-cert.pem # Trust anchor for OMS signature verification
├── skills.sh.json # Skills.sh marketplace grouping config
├── CHANGELOG.md
├── CONTRIBUTING.md # Contribution guidelines
├── SECURITY.md # Security reporting policy
├── CODE_OF_CONDUCT.md # Community code of conduct
├── LICENSE-APACHE # Apache 2.0 (source code)
└── LICENSE-CC-BY-4.0 # CC BY 4.0 (documentation/skills)
Skills are maintained in their respective product repos (see the Source column in the Skill Catalog) and synced to this repo daily. Products only appear under skills/ after the sync pipeline confirms each skill carries:
skill.oms.sig — detached OMS-format signature (verifiable against nv-agent-root-cert.pem)skill-card.md — skill identity and governance cardevals/evals.json, evals/*.json, eval/*.json, or benchmark/evals.jsonWhen evaluation runs produce a BENCHMARK.md, it ships alongside the skill so consumers can see verifiable benchmark uplift data.
This repository adheres to the Agent Skills specification:
SKILL.md file at their root.name and description fields.skills-ref reference library.Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
This code is dual-licensed with documentation/skills under the CC-BY-4.0 AND source code under Apache-2.0 license terms. The full license texts can be found in LICENSE-APACHE and LICENSE-CC-BY-4.0 respectively.
name: omniverse-cad-to-simready
description: "Coordinate the end-to-end CAD/source-asset to SimReady workflow. Use for broad requests such as CAD to SimReady, source asset to simulation-ready USD, or prop packaging that require conversion, material/physics assignment, SimReady conformance, validation, and optional package creation; deploy or verify Content Agents services first when property assignment is enabled; route single-stage work through nested references."
version: "0.1.0"
license: Apache-2.0
tools:
- Read
- Shell
compatibility: >
Orchestrator skill. Managed Content Agents deployment requires NVIDIA_API_KEY
(build.nvidia.com), Docker + NVIDIA Container Toolkit + GPU, Python 3.12, and
an upstream checkout of nvidia-omniverse/content-agents on branch
main. Reused/provided endpoints may instead use explicit endpoint and
usage-token environment variables. Linux/macOS only.
metadata:
author: Omniverse
tags:
- physical-ai
- simready
- workflow
- cad
- conversion
domain: ai-ml
languages:
- pythonUse this workflow skill when the user wants an end-to-end pipeline from a source asset to a SimReady asset or package. This skill coordinates existing conversion, authoring, validation, conformance, rendering, and packaging references directly. Do not replace the workflow with a single monolithic runner command.
This skill is documentation-driven and does not ship scripts/run.py. It
should not depend on a repository checkout. When a stage needs deterministic
execution, run the portable script from that stage reference's installed
directory. Shell is declared because this workflow invokes installed stage
reference scripts directly; it still must not grow a monolithic runner.
preflight reference first for deterministic setup. It
installs or verifies local upstream checkouts, writes a
cad-to-simready-preflight.json manifest, and exports
PHYSICAL_AI_PREFLIGHT_MANIFEST plus PHYSICAL_AI_REQUIRE_PREFLIGHT=1 for
downstream references.uv (per repo README.md).https://build.nvidia.com when local Content Agents
deployment will run. Already-running endpoints may instead use explicit
endpoint variables plus usage tokens such as NGC_API_KEY, NVCF_API_KEY,
or CONTENT_AGENTS_*_TOKEN.${PHYSICAL_AI_SKILL_HUB_UPSTREAM_ROOT:-$HOME/.physical-ai-skill-hub/upstreams}
when a downstream stage needs upstream scripts or specs.Conversion-only is a valid workflow request. When the user asks only to convert
or smoke-test source asset conversion, set property_assignment_intent=skip,
do not deploy Content Agents, run convert-to-usd, then run
validate-usd-minimum on the generated USD if conversion succeeds.
Do not imply that uv sync installs every source converter runtime. URDF,
MuJoCo/MJCF, and the repo Python dependencies are handled by the project
environment, but NVIDIA-backed source conversion requires an installed and
validated NVIDIA-Omniverse/usd-convert-cad checkout. If that runtime is
missing or does not support the source, preserve the blocked conversion report
and its install_hint instead of attempting an unrequested local build or
substituting another converter.
For any broad CAD/source-asset to SimReady request, assume
property_assignment_intent=run unless the user explicitly asks for
conversion-only, validation-only, or no material/physics assignment.
Before invoking converter, validation, Content Agents, OVRTX, packaging, or FET
helper scripts, run the preflight reference or verify an existing
PHYSICAL_AI_PREFLIGHT_MANIFEST. Treat preflight as the mandatory dependency
bootstrap step, not as workflow routing. If the user explicitly asks not to
deploy services or asks for conversion-only/validation-only, use
--skip-content-agents.
When PHYSICAL_AI_REQUIRE_PREFLIGHT=1 is set and a required component is not
ready in the manifest, downstream references must block with the preflight
guardrail instead of rediscovering upstreams or services directly.
When property_assignment_intent=run, the first operational action after
confirming the source path and resolving intent is to verify or deploy Content
Agents services. Do this before asset-context inspection, converter dependency
checks, conversion, validation, conformance, rendering, packaging, or upstream
source builds.
Use healthy existing endpoints when available. If OVRTX, Material, or Physics
endpoints are missing or unhealthy, run deploy-content-agents
first and do not continue until the shared standalone OVRTX renderer plus
independent Material and Physics service containers are healthy and exported
through CONTENT_AGENTS_*_BASE_URL. Deploy the Texture Agent too when texture
generation is requested.
If required deployment authentication is missing, ask the user for
NVIDIA_API_KEY and wait. If a provided endpoint requires usage auth, ask for
the appropriate usage token instead. If deployment cannot produce healthy
services, report Content Agents readiness as blocked instead of proceeding to
conversion.
output_root, and classify
the request as end-to-end, conversion-only, validation-only, or packaging.property_assignment_intent before running any asset inspection,
converter probe, conversion, validation, conformance, rendering, or
packaging step.preflight for the selected workflow targets, unless a ready
PHYSICAL_AI_PREFLIGHT_MANIFEST is already configured. Source the generated
env file before running downstream scripts. Treat preflight as dependency
setup only: it may use a provided --source-asset, --source-format, or
--conversion-tools value to scope dependency checks, but convert-to-usd
and the upstream converter references still decide actual conversion support.property_assignment_intent=run; block on missing authentication or
unhealthy services instead of continuing.references/workflow.md and references/commands.md, then run only
the stage references needed for the current request.identify-asset-context on the original source asset when web search is
available or property assignment will run.convert-to-usd, or skip conversion for existing
USD input and treat the source path as the current USD path.validate-usd-minimum before expensive downstream work. Treat this as a
viability gate only: record unit/profile issues such as metersPerUnit != 1.0, but do not run simready-conform-profile, FET001, or any other FET
repair before Content Agents assignment when property assignment will run.simready-conform-profile on the latest simulation USD path after
property assignment and preserve every selected FET repair report.omni-asset-validate,
omni-asset-validate-geometry, omni-asset-validate-physics, and
simready-validate.simready-conform-profile when simready-validate reports a
repairable requirement, then rerun profile validation on the newest authored
USD.ovrtx-render-service when preview, thumbnail, or inspection images
are requested. When package outputs are requested, run
assemble-package-source next to create the clean deliverable/ package
source from the final USD and thumbnail, then run nv-core-package-sample
and nv-core-package-sample-validation on that deliverable folder only.Use the simready-conform-profile reference only after property assignment
when property_assignment_intent=run. It routes feature repair to upstream
SimReady Foundation FET skills such
as simready-foundation-conform-fet-000-core,
simready-foundation-conform-fet-001-minimal,
simready-foundation-conform-fet-004-simulate-multi-body-physics, and
simready-foundation-conform-fet-005-simulate-grasp-physics from branch
main.
If simready-validate reports a repairable requirement after the first
conformance pass, feed the structured requirement IDs back into
the simready-conform-profile reference before writing the final result. In
particular, GSP.001 is owned by upstream
simready-foundation-conform-fet-005-simulate-grasp-physics; run that skill when a
vision-capable agent can inspect visual evidence or explicit grasp points were
provided, otherwise record the FET005 step as blocked by missing vision/points
instead of treating it as an optional preview task.
For RB.MB.001, route the failure to
upstream simready-foundation-conform-fet-004-simulate-multi-body-physics. Do not assume
multiple visual prims are multiple rigid bodies; inspect
UsdPhysics.RigidBodyAPI applications. When the Physics Agent report shows
composed topology optimization or the USD has existing component colliders/part
roots and the profile validator reports FET004/RB.MB.001, FET004 should promote
those existing components into rigid bodies without creating geometry. Do not
mark the gate not applicable until after confirming there are fewer than two
reusable body candidates.
Emit a consolidated workflow report in Markdown, and include JSON when the workflow writes structured artifacts. The report must include:
passed, blocked, failed, or needs_rerun.Read only the references needed for the current request:
references/preflight/README.md: deterministic local setup, manifest/env
contract, Linux and Windows wrappers, Content Agents deployment opt-out, and
guardrail behavior.references/workflow.md: inputs, source routing, detailed workflow,
validation policy, output report fields, approval points, and next steps.references/commands.md: concrete portable script command patterns for each
stage.references/assemble-package-source/README.md: two-zone package source
assembly, canonical root USD naming, thumbnail placement, and self-contained
deliverable checks.Use skills/omniverse-cad-to-simready/ as the source of truth for this product
repo's skill. The .agents/skills symlink is a compatibility alias for local
agentskills.io-style discovery, and .codex/skills and .claude/skills are
agent-specific compatibility aliases.
Frontmatter keeps version and tools at top level for agentskills.io runtime
compatibility. NVCARPS discoverability fields live under metadata.
The nested references/ tree is intentional. It keeps one public catalog skill
while retaining script-bearing atomic stage references, upstream handoff notes,
and router documentation under the workflow. Do not flatten those references or
promote nested README references to sibling SKILL.md files unless the repo's
publishing model changes.
simready-foundation-conform-fet-005-simulate-grasp-physics needs visual
review or explicit grasp points before it can author a meaningful grasp
vector.| Symptom | Cause | Fix |
|---|---|---|
| Downstream reference reports that cad-to-simready preflight has not prepared a component | PHYSICAL_AI_REQUIRE_PREFLIGHT=1 is set, but the manifest is missing or the required runtime/service is not ready | Run preflight/scripts/preflight.py, source the generated env file, or explicitly disable service deployment with --skip-content-agents only when Content Agents are out of scope. |
Workflow stops on GSP.001 and reports the failure as unclassified | Visual evidence or explicit grasp points were not provided to FET005 | Run upstream simready-foundation-conform-fet-005-simulate-grasp-physics only after a vision-capable agent has reviewed the asset, or pass explicit grasp points. Otherwise report the FET005 step as blocked, not failed. |
| Validation fails after a meaningful USD artifact already exists | Workflow stopped at the first validation finding | Continue remaining diagnostic gates and mark the result needs_rerun. Do not stop at validation findings once a USD artifact has been produced. |
| Property-assignment stage fails with a missing service endpoint | Content Agents service was not deployed before conversion | Run deploy-content-agents first. Do not start asset inspection, conversion, validation, conformance, rendering, or packaging before Content Agents readiness when property assignment will run. |
Material Agent reports that rendering produced 0 images after unit or profile repair | A FET repair, commonly FET001 unit normalization, was applied before Material Agent and changed the USD layering/scene state consumed by the service | Rerun assignment from the converted/minimum-valid USD: Material Agent first, then Physics Agent, then run simready-conform-profile and FET repairs on the latest service-authored USD. |
Material or Physics Agent local optimized path reports Permission denied: '/app/.build-resources/scene_optimizer_core/python' | Local Docker Scene Optimizer bundle permissions prevent the non-root service user from reading the packaged SO runtime | Repair the relevant local container with docker exec --user root content-material-agent-service chmod -R a+rX /app/.build-resources/scene_optimizer_core or docker exec --user root content-physics-agent-service chmod -R a+rX /app/.build-resources/scene_optimizer_core, then rerun the same optimized agent command. Do not treat the no-optimizer fallback as the root cause for instanced/prototype assets. |
RB.MB.001 fails even though the asset has many prims | The profile counts UsdPhysics.RigidBodyAPI prims, not visual or collider prims; Physics Agent may author one root rigid body | Route to upstream simready-foundation-conform-fet-004-simulate-multi-body-physics. First ensure Physics Agent used composed-topology optimization when applicable, then promote existing component colliders/part roots when the active profile reports FET004/RB.MB.001 and no geometry must be invented. |
PHYSICAL_AI_REQUIRE_PREFLIGHT=1 is set, do not bypass the
manifest with direct upstream discovery.omniverse-cad-to-simready runner command.convert-to-usd reference; do not
substitute another converter for CAD or mesh formats.simready-conform-profile or
any FET helper before Content Agents. Validate minimum USD first, then run
Content Agents on that converted/minimum-valid USD, then apply FET repairs to
the latest service-authored USD.simready-validate or any
SimReady profile validation before Content Agents. The only validation gate
allowed before service calls is validate-usd-minimum, which is a basic USD
viability check.needs_rerun.GSP.001 profile failure as an unclassified final finding.
Route it to upstream simready-foundation-conform-fet-005-simulate-grasp-physics; if
the current agent cannot inspect renders or no explicit grasp points are
available, report a blocked FET005 repair with the visual evidence path or
missing input reason.
评论 (0)
暂无评论,成为第一个评论者吧!