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.
license: Apache-2.0
name: doca-dpa
description: >
Use this skill when the user is doing hands-on DOCA DPA host-side
work on a BlueField — creating the `doca_dpa` Core context, loading
a DPACC-compiled DPA app image (`doca_dpa_app`), creating DPA
threads, launching kernels via `doca_dpa_kernel_launch_update_*`,
draining `doca_dpa_completion`, running `doca_dpa_cap_*` discovery,
choosing between the DPA comm component (inter-DPA messaging) and the
DPA verbs component (in-kernel RDMA), or debugging `DOCA_ERROR_*`
from `doca_dpa_*`. Trigger even without "DOCA DPA" or "Data-Path
Accelerator": "run compute on the DPA from my host", "DPA kernel
hangs, no completion", "DOCA_ERROR_DRIVER on launch", "DOCA/DPACC
version skew", or "does this BlueField expose a DPA". Route elsewhere
for DPA-side kernel programming itself, DPACC compiler internals,
host↔DPU messaging (doca-comch), host-side RDMA (doca-rdma), and
GPU-initiated networking (doca-gpunetio).
metadata:
kind: library
compatibility: >
Requires DOCA SDK installed at /opt/mellanox/doca on Linux
(Ubuntu 22.04/24.04 or RHEL/SLES) with a BlueField DPU whose
generation exposes the DPA processor to the host, plus the
matching DPACC compiler at a version listed compatible by
the DOCA Compatibility Policy. Reads `pkg-config --modversion
doca-dpa` and the installed `dpacc` version, and inspects
/opt/mellanox/doca/{lib,include,samples/doca_dpa,applications}.Where to start: This skill assumes DOCA is already installed,
the user's BlueField has a DPA processor and the host can see it
through DOCA, and the user is doing hands-on DPA work from the
host side — i.e. using doca-dpa to load a DPA application
image, launch DPA kernels, and exchange data with the DPA
processor. Open TASKS.md if the user wants to do
something (configure / build / modify / run / test / debug); open
CAPABILITIES.md when the question is what
can the host-side DPA API express on this version + this
BlueField generation. If the user has not installed DOCA yet,
route to doca-setup first; if the
user is asking how to write the DPA-side kernel itself (the
code that runs on the DPA processor, compiled by dpacc), that
is a different scope — route via
doca-public-knowledge-map
to the public DOCA DPA / DPACC / DPA-Comms / DPA-Verbs guides
(this skill does not redefine those DPA-side surfaces).
The CLASSES of DPA questions this skill is built to answer, each with one worked example. The agent should treat the class as the load-bearing piece — the worked example is a single instance.
CAPABILITIES.md ## Capabilities and modes
TASKS.md ## configure.doca_dpa_cap_* against the
active doca_devinfo plus the DOCA-install axis via
pkg-config --modversion doca-dpa) in
CAPABILITIES.md ## Capabilities and modes
TASKS.md ## configure.DOCA_ERROR_NOT_SUPPORTED even though DOCA Core looks
healthy?" — worked example: "the BlueField generation in
this host predates the DPA feature my code uses". Answered by
the env-precondition matrix in
CAPABILITIES.md ## Safety policy
TASKS.md ## configure step 1.CAPABILITIES.md ## Capabilities and modes
TASKS.md ## run.CAPABILITIES.md ## Version compatibility
which cross-links the canonical detection chain in
doca-version and adds the
DPA-specific DOCA must match DPACC overlay.DOCA_ERROR_* from a doca_dpa_* call mean
and which layer caused it?" — worked example: "DOCA_ERROR_DRIVER
on a host-side launch call — is it DOCA, the DPACC-produced
image, or the DPA processor itself?". Answered by the DPA
overlay on the cross-library taxonomy in
CAPABILITIES.md ## Error taxonomy
TASKS.md ## debug that escalates to
doca-debug.doca_dpa_app; thread A
sends a counter value to thread B over a DPA-side comms
endpoint and thread B signals the host through
doca_dpa_completion". Answered by the DPA-Comms routing
rule, primitive families, host-side capability-budget rule,
and DPA-Comms error overlay in
CAPABILITIES.md ## comms plus the
configure / build / modify / run / test / debug overlay in
TASKS.md ## comms. Disambiguates the DPA
device-side comm component from host-side doca-comch and
host-side doca-rdma.CAPABILITIES.md ## verbs plus the
workflow overlay in TASKS.md ## verbs.
Includes the climb-back rule for when the latency-tuning
premise stops holding.This skill serves external developers building applications
that consume the DOCA DPA library from the host side — i.e.,
users whose code calls doca_dpa_* from host C / C++ to stand
up the per-DPA-instance context, load a DPA application image
that dpacc produced from their DPA-side source, create one or
more DPA threads, launch DPA kernels with arguments, and drain
completions. It is not for NVIDIA developers contributing to
DOCA DPA itself, nor is it the place to learn how to write
the DPA-side kernel code (that path goes through the public
DOCA DPA, DPACC, DPA-Comms, and DPA-Verbs guides via
doca-public-knowledge-map).
Language scope. DOCA DPA ships as a host-side C library
with pkg-config module name doca-dpa. The host-side API is
C; the DPA-side kernel is a separate translation unit written
in the language the DPACC compiler accepts and compiled by
dpacc into a binary that the host packages into the
executable as the DPA application image. The shipped samples
under /opt/mellanox/doca/samples/doca_dpa/ are written in C
plus DPA-side source (NVIDIA's choice). Other-language
consumers are limited in practice — the DPA-side kernel has no
FFI escape hatch because it must be a translation unit dpacc
accepts — but a Rust / Go / Python host-side wrapper that drives
doca_dpa_* setup and launches a DPA kernel image built
separately is still useful, and the skill keeps the lifecycle,
capability-discovery, env-precondition, and error-taxonomy
guidance language-neutral.
Load this skill when the user is doing hands-on DOCA DPA work
from the host side, in any host language plus a DPA-side
translation unit built by dpacc. Concretely:
doca_dpa against a doca_dev that maps to a
BlueField with a DPA processor visible to the host.doca_dpa_app) that
dpacc produced from the user's DPA-side source, into the
doca_dpa context.doca_dpa_thread)
so DPA kernels have somewhere to run on the DPA processor.doca_dpa_kernel_launch_update_* family, and
reasoning about which argument shape is supported on this
install.doca_dpa_completion to observe when async DPA
work finishes, and draining it from the host side.doca_devinfo via the doca_dpa_cap_* family — BlueField
generations differ in DPA hardware support.DOCA_ERROR_* returned from a doca_dpa_* call
— in particular disambiguating DPA not present on this
BlueField from DPA feature too new for this hardware
generation from DPACC-produced image mismatched against
the host-side DOCA install from DPA driver layer
reporting failure.dpacc —
the env-precondition and capability-discovery rules in this
skill still apply.libdoca_dpa_dev_comm.a (header
doca_dpa_dev_comch_msgq.h) — for inter-DPA-thread
messaging or coordination signals between DPA threads on the
same doca_dpa_app. The DPA-Comms routing rule, primitive
families, capability rule (there is no per-primitive host
cap-query family — host-side DPA discovery is only
doca_dpa_cap_is_supported /
doca_dpa_cap_get_max_kernel_time_alive_supported), error
overlay (_AGAIN → kernel must yield; _BAD_STATE
disambiguation from the parent's host-side _BAD_STATE), and
the configure / build / modify / run / test / debug overlay
live in CAPABILITIES.md ## comms and
TASKS.md ## comms under this same skill.libdoca_dpa_dev_verbs.a (header
doca_dpa_dev_verbs.h) — for RDMA from inside
the DPA kernel to a remote peer when the host round-trip is
the measured latency bottleneck. The 4-way RDMA matrix, the
host-configures-QP / DPA-uses-QP coupling rule, the
capability rule (no per-verb host cap-query family exists;
verb availability follows the BlueField generation + matched
DOCA/DPACC install, read from doca_dpa_dev_verbs.h and the
shipped sample), the IO_FAILED → CQE-inspection
overlay, and the climb-back rule live in
CAPABILITIES.md ## verbs and
TASKS.md ## verbs under this same skill.Do not load this skill for general DOCA orientation,
install of DOCA or the DPACC compiler, the DPA-side
programming model itself (how to write a DPA kernel; the
DPA device-side comm and verbs components
(libdoca_dpa_dev_comm.a / libdoca_dpa_dev_verbs.a) that
run inside the DPA kernel), or non-DPA library questions.
For those, route through
doca-public-knowledge-map
to the matching upstream guide.
This is a thin loader. The body keeps only the orientation needed to pick the right next file. The substantive DPA-specific material lives in two companion files:
CAPABILITIES.md — what the host-side DPA API can express on
this version + this BlueField generation: the per-DPA-instance
doca_dpa context, the loaded doca_dpa_app image produced
by dpacc, the doca_dpa_thread execution context, the
host-initiated kernel launch surface
(doca_dpa_kernel_launch_update_*), the doca_dpa_completion
mechanism, the capability-query surface (doca_dpa_cap_*),
the DPA error taxonomy mapped onto the cross-library
DOCA_ERROR_* set, the observability surface (host-side
completions plus the public DPA developer tools surface
reachable via doca-public-knowledge-map), and the safety
policy that gates env preconditions (DPA-capable BlueField,
matched DOCA + DPACC versions, DPA-side image and host-side
expected entry points agree).TASKS.md — step-by-step workflows for the six in-scope DPA
verbs: configure, build, modify, run, test,
debug. Plus a Deferred task verbs block that points
out-of-scope questions at the right next skill.The skill assumes a host where DOCA is already installed at
the standard location, a BlueField with a DPA processor is
physically present and visible to the host, the DPACC compiler
is installed at a version matched to the DOCA install per the
DOCA Compatibility Policy, and the user already knows how (at
least at a sketch level) to write the DPA-side kernel that
dpacc will compile. It does not cover installing DOCA or the
DPACC compiler — that path goes through
doca-setup.
This skill is agent guidance, not a samples or templates bundle. To keep the boundary clean, it deliberately does not contain — and pull requests should not add:
/opt/mellanox/doca/samples/doca_dpa/. The agent's job is to
route the user to those files and prescribe a minimum-diff
modification on them via the universal modify-a-sample
workflow in
doca-programming-guide,
layered with the DPA-specific overrides in
TASKS.md ## modify.meson.build,
CMakeLists.txt, …) parked inside the skill. The agent
constructs the build manifest in the user's project
directory against the user's installed DOCA + DPACC
compiler, where pkg-config --modversion doca-dpa and the
installed dpacc are the two sources of truth.samples/, bindings/, or reference/ subtree of any
kind. A mock or incomplete artifact in this skill's tree,
even one labeled "reference", is misleading: users will read
it as buildable.libdoca_dpa_dev_comm.a / libdoca_dpa_dev_verbs.a).
These are DPA-side archives shipped as part of doca-dpa
(NOT separate pkg-config modules): their symbols are called
from inside the DPA kernel and linked into the DPA image by
dpacc, not from the host. Their public guides are reachable
via
doca-public-knowledge-map.
This skill names them and routes; it does not redefine them.SKILL.md first to confirm the user's question is
in scope (host-side DPA work, not DPA-side kernel-writing).doca_dpa per-instance
context, the loaded doca_dpa_app image, the
doca_dpa_thread execution context, the kernel-launch +
completion model, the dual capability query, the
env-precondition policy, the error taxonomy, the
observability surface, and the safety policy, see
CAPABILITIES.md.Both companion files cross-link to each other,
doca-version for the canonical
DOCA version-handling rules (with the DPA overlay that DOCA
must match the DPACC compiler), and
doca-public-knowledge-map
whenever the right answer is "look it up in the public DOCA
DPA, DPACC, DPA-Comms, or DPA-Verbs guide, or in the on-disk
install layout" rather than "DPA host-side-specific guidance".
doca-public-knowledge-map —
the routing table for every public DOCA documentation
source and the on-disk layout of an installed DOCA package.
The DPA public guide is at
https://docs.nvidia.com/doca/sdk/DOCA-DPA/index.html; the
DPACC compiler guide, the DPA-Comms guide (DPA-side
communications), the DPA-Verbs guide (DPA-side verbs), and
the DPA Tools umbrella (developer / admin CLIs for DPA)
live in the same routing table and are companion surfaces
to this skill rather than redefined by it.doca-setup — env preparation,
install verification, DPACC compiler install / verification,
and the I have no install yet path with the public NGC
DOCA container. This skill assumes its preconditions are
satisfied AND that DPACC is installed at a version that
matches DOCA.doca-version — canonical
DOCA version-handling rules. This skill's ## Version compatibility cross-links the four-way match rule and adds
the DPA-specific DOCA-and-DPACC must match overlay per the
DOCA Compatibility Policy.doca-structured-tools-contract —
the bundle's structured-tools precedence rule (detect /
prefer / fall back / report). The Command appendix in
TASKS.md honors this contract.doca-programming-guide —
general DOCA programming patterns shared by every library:
the canonical pkg-config + meson build pattern, the
universal modify-a-shipped-sample first-app workflow, the
universal Core-context lifecycle, the cross-library
DOCA_ERROR_* taxonomy, and the program-side debug order.
This skill layers DPA specifics on top.doca-debug — the cross-cutting
debug ladder (install / version / build / link / runtime /
program / driver). DPA-specific debug (DPACC + DOCA version
skew, DPA not present on this BlueField generation, DPA
kernel hangs that show no host-side completion,
launch-argument shape mismatches between the host launch
call and the DPA-side function signature) overlays on top
of that ladder.DOCA DPA's DPA device-side components — the comm archive
libdoca_dpa_dev_comm.a (communication primitives the DPA
kernel itself calls, header doca_dpa_dev_comch_msgq.h) and
the verbs archive libdoca_dpa_dev_verbs.a (ibverbs-like RDMA
verbs the DPA kernel itself calls, header
doca_dpa_dev_verbs.h) — are DPA-side archives shipped
within doca-dpa, not separate pkg-config modules, and each
has its own public guide. No library skill ships for them in
this bundle yet; for any DPA-side question, route via
doca-public-knowledge-map
to the public DOCA DPA Comms and DOCA DPA Verbs guides
and to the shipped /opt/mellanox/doca/samples/doca_dpa/
samples (which include both host-side and DPA-side
translation units). Conflating them with doca-dpa is the
single most common DPA first-app design error.
评论 (0)
暂无评论,成为第一个评论者吧!