NVIDIA/skills is a catalogue of portable instruction sets that teach coding agents how to use NVIDIA software for robotics, simulation, CUDA, retrieval-augmented generation, and related workflows. Developers install these skills in agents such as Claude Code or Codex, while the catalogue mirrors skills maintained in separate NVIDIA product repositories.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add NVIDIA/skills --skill doca-programming-guidegit clone --depth 1 https://github.com/NVIDIA/skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/nvidia/skills/doca-programming-guide)<a href="https://agentmods.dev/skills/nvidia/skills/doca-programming-guide"><img src="https://agentmods.dev/badge/skills/nvidia/skills/doca-programming-guide/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/nvidia/skills/doca-programming-guide"><img src="https://agentmods.dev/badge/skills/nvidia/skills/doca-programming-guide.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00176 | $0.03595 |
| Opus 5.5 | $0.00070 | $0.01438 |
| Sonnet 5.5 | $0.00035 | $0.00719 |
| Haiku 4.5 | $0.00018 | $0.00360 |
Grade A, and why
doca-programming-guide scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 290 lines — stays where its author put it; the contents beside it link to each section on GitHub.
DOCA programming guide
Examples
- "How do I write my first DOCA program?" — copy a shipped sample and follow the modify workflow, then the canonical pkg-config + meson build. Docker / NGC can compile and link; runtime needs a real BlueField or ConnectX.
- "What's the lifecycle every DOCA object follows?" — create → configure → start → connect → use → stop → destroy. Configuration finishes before start.
Instructions
Follow Loading order below. Open the tasks companion for how-to and the capabilities companion for coverage. Do not open every companion file.
Where to start: Read ## Audience to confirm the user
is consuming DOCA, not contributing to it. Then jump to the H2
that matches the verb (## modify for first-app derivation,
## build for the canonical build pattern, ## test for the test
loop, ## debug for the program-class debug ladder). Docker / NGC
is the build half, not the runtime half. The NGC container can
compile and link. Flow pipe creation, RDMA QP ops, and any call
that needs a doca_dev on a port need a real BlueField or
ConnectX. Do not promise the app will work entirely inside docker.
Example questions this skill answers well
These are the CLASSES of program-class questions the skill is built to answer, each with one worked example. Library-specific overlays (Flow / DMS / Caps / …) live in the matching library skill; this skill answers the library-agnostic shape.
- "How do I write my first DOCA program for <any library>?" —
worked example: "I have docker; help me build my first DOCA Flow
app." Answered by the modify-a-shipped-sample workflow in
modify plus the canonical build
pattern in ## build. Say that docker / NGC builds and links, but runtime needs a real NIC or DPU. - "What's the right build line for any DOCA library?" — worked
example: "How do I compile a program that calls
doca_rdma_*?" Answered by thepkg-config doca-<library>pattern inbuild (C/C++ Track 1) and the
FFI/bindings pattern in Track 2. - "What's the lifecycle every DOCA object follows?" — worked
example: "What's the right order of
doca_flow_pipe_*calls in my program?" Answered by the Core order create → configure → start → connect → use → stop → destroy in ## Capabilities and modes (cfg-create/initimplement configure and finish before start). - "
DOCA_ERROR_*came back — what does it mean and what do I do?" — worked example: "My code gotDOCA_ERROR_BAD_STATE." Answered by the cross-librarydoca_error_get_descr()rule inError taxonomy
- the program-class debug order in
debug.
- "My program built and started, but does nothing on the wire." —
worked example: "My Flow program runs cleanly but no traffic is
matched." Answered by the validate-before-commit rule in
Safety policy
and the layered program-class debug ladder indebug.
- "What does <language> consumer of DOCA look like (FFI /
bindings)?" — worked example: "How do I call DOCA Comch from Rust
without writing C?" Answered by Track 2 of
build (FFI against the public C
ABI) and the language-neutral lifecycle inCapabilities and modes.
- "How should I classify and build all the shipped DOCA samples and
applications? What's the difference between a sample and an
application?" — worked example: "I tried to build all DOCA apps
and 20/159 failed — which ones are real regressions vs missing
optional stacks?" Answered by the sample-vs-application model and
the category / dependency / skip-vs-fail taxonomy in
sample-and-app-categorization,
which separates "the SDK is broken" from "the optional GPU / RMAX / MPI stack is not on this BlueField".
What ships with it
6 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago Changed · +1 lines fdc7dfaea266
- 12d ago Changed · +27 lines · -70 tokens per session 45c2fbee9dd4
- 1mo ago First seen · 262 lines · 246 tokens per session scan A a4d12cce6918
doca-programming-guide is a skill published in the GitHub repository NVIDIA/skills (3,748 stars, last pushed today), licensed Apache-2.0. It adds 176 tokens to every session and 3,595 once invoked, about $0.0007 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other skills, from other repositories
desktop-and-local-ipc-security
Use when the diff adds or changes a loopback HTTP or WebSocket server, a bind address, a Unix socket, named pipe, XPC or D-Bus service, an Electron or Tauri main-process handler, a custom URL scheme or file association, a privileged helper or updater, or code that spawns tools inside directories a user or download…
spark-training-gotchas
Preflight and diagnose the ten known failure modes for ML training on NVIDIA DGX Spark. Use when a training run on DGX Spark fails to start, OOMs below the 128GB limit, slows down mid-run, or before any multi-hour training job on GB10.
tutorial
Interactive tutorial teaching Ouroboros hands-on.
subscription-setup
A setup workflow for creating recurring medical-learning subscriptions after a doctor’s profile has been registered.
ace-rsi-toolkit
Run ACE (Agentic Context Engineering, Zhang et al. 2025) on an agent the sequential offline way - a Reflector reads each training Trace, a Curator adds the lessons to a playbook of rules the agent reads before every task, and the final playbook is measured on a published Benchmark.
literature-review
Systematic literature review workflow: scope, search (arXiv, Semantic Scholar, Google Scholar), screen, extract, synthesize, and identify gaps. Triggers on: "literature review", "survey the literature", "related work", "systematic review", "synthesize the research", "find papers about", "research gap analysis".