What does this skill do, and when should you use it?
This Agent Skill focuses on the environment surrounding NVIDIA DOCA workloads: installation health, build prerequisites, runtime prerequisites, and setup-class failures. It covers pkg-config, headers, library paths, hugepages, devlink visibility, representors, and kernel-module prerequisites, with an NGC DOCA container fallback for hosts without an install. Deployment-shaped requests are first classified by system shape and then routed toward container or bare-metal deployment guidance. It explicitly excludes DOCA library APIs, sample-derived application development, program build details, and DOCA_ERROR_* program debugging.
Reads SKILL.md and its TASKS.md and CAPABILITIES.md companion materials to classify requests and select workflows; guides target-host checks for DOCA installation, pkg-config, headers, library paths, hugepages, devlink, representors, and kernel prerequisites; detects deployment shape and routes to container, bare-metal, or no-install paths; applies verification and safety constraints before environment changes; and requires exact target-host command output when a command channel is unavailable.
- A BlueField developer who has a built workload but needs to choose between container and bare-metal deployment.
- An engineer checking whether DOCA Flow and the surrounding build environment are healthy on a target system.
- A developer troubleshooting missing doca-*.pc files, headers, library paths, hugepages, or representors.
- A DPU operator preparing runtime prerequisites before launching a DOCA workload.
- A macOS, Windows, or unconfigured Linux user who wants to learn or build with DOCA through the NGC container path.
How do you install this skill?
- Unify the license and version metadata, and add authorship, changelog, maintenance ownership, and executable tests.
- Require explicit confirmation, state capture, and a concrete rollback plan before changing hugepages, switchdev, mlxconfig, installation state, or reboot-sensitive configuration.
- NGC/Docker paths depend on external services; validate reachability from mainland-China networks and document local or mirrored alternatives.
- Do not treat the benchmark percentages as independently reproduced evidence in this static review.
- Shell / CLI
- Network access
- Local filesystem
Docker for the NGC container Stage-1 pathA live DOCA install at /opt/mellanox/doca for validation workflows
Install the individual skill from NVIDIA’s collection with: npx skills add nvidia/skills --skill doca-setup --yes. The supplied README says cloning the repository or manually copying skill folders is unnecessary; the CLI prompts for or applies the installation destination.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/doca-setup" ~/.claude/skills/
rm -rf "$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:
- I just got a BlueField—what now?
- Is DOCA Flow available on this box?
- pkg-config cannot find doca-flow; what is missing?
After installation, invoke it with an environment- or deployment-shaped request such as “I just got a BlueField—what now?”, “Is DOCA Flow available on this box?”, or “pkg-config cannot find doca-flow; what is missing?” It classifies the request before selecting a workflow. Do not assume local shell access reaches the target host; provide a target-host command channel or have the user run the checks and return exact output.
What are this skill's strengths and limitations?
- Covers the install, build, and runtime environment layers in one troubleshooting flow.
- Separates container, bare-metal, and no-install paths instead of assuming a deployment shape.
- Includes explicit safety constraints for hugepages, mlxconfig resets, and eswitch mode changes.
- Provides an NGC DOCA container Stage-1 path for users without a DOCA installation.
- It is a thin loader whose substantive guidance depends on TASKS.md and CAPABILITIES.md companion files.
- Hardware validation requires a target-host command channel or exact user-supplied output; local evidence is not assumed to represent the target.
- It does not cover Flow or RDMA APIs, program construction, sample modification, program builds, or DOCA_ERROR_* debugging.
- The supplied material does not show concrete validation commands, a test suite, or tested-platform evidence.
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 |
|---|---|---|---|---|
| DOCA Environment Setup this page ✓ NVIDIA · Official | 57 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| DOCA Comch Host–DPU Messaging ✓ NVIDIA · Official | 53 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| DOCA Telemetry Exporter Development ✓ NVIDIA · Official | 55 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| DOCA Programming Guide ✓ NVIDIA · Official | 54 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| DOCA Version Consistency Assistant ✓ NVIDIA · Official | 53 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
How did FollowSkills review this skill?
The files explicitly require least privilege, user confirmation, disclosure of global hugepage/eswitch/install effects, no unattended reboot, and no privileged-container workarounds. However, they still prescribe sudo and global state changes, provide incomplete rollback guidance, and conflict on licensing between SKILL.md (Apache-2.0) and skill-card.md (Apache 2.0 AND CC-BY-4.0), so points are deducted.
The recognize→configure→test→debug structure, stop conditions, and error taxonomy are comparatively detailed. However, this is a static review with no command execution, many key paths depend on unprovided companion skills, and the benchmark report lacks its source dataset, so the score is conservatively limited.
Triggers, covered scenarios, and explicit non-fit boundaries are clear across host, DPU, container, and bare-metal shapes. Chinese-language interaction is not addressed, and reachability of NGC/Docker-dependent paths from mainland-China networks is not assessed, so points are deducted.
The documentation has good layering, progressive loading, dependency notes, examples, limitations, and stated ownership. It lacks a clear changelog and maintenance/update procedure; the benchmark also records missing Instructions/Examples/Author metadata and unexpected companion files, while license metadata is inconsistent.
The skill offers directly actionable checks, environment-preparation workflows, failure classification, and deployment routing, with clear core value. It is nevertheless a thin loader whose substantive outcomes depend on other skills and live environments; static review cannot establish successful completion, so the score remains below the execution-based ceiling.
Auditable commands, paths, cross-references, public documentation links, and a benchmark report are present. But the benchmark dataset is unavailable, no committed tests cover the key paths, and this review did not reproduce execution, warranting only limited credit.
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 →