What does this skill do, and when should you use it?
This skill targets external developers using the DOCA Ethernet C library for high-rate packet I/O. It covers RX/TX queue lifecycles on physical ports, representors, and SFs, queue-type selection, capability queries, send tasks, and DOCA_ERROR_* diagnosis. It assumes the DOCA SDK is already installed at the standard path and excludes DOCA installation, flow-rule programming, host-to-DPU control messaging, and RDMA data movement. Its companion files provide capability guidance and configure, build, modify, run, test, and debug workflows.
It guides users to inspect the installed DOCA Ethernet library with pkg-config doca-eth and read libraries, headers, samples, and applications under /opt/mellanox/doca. It helps configure doca_eth_rxq and doca_eth_txq objects, choose regular, cyclic, managed mempool, or shared mempool RX types, size burst and scatter-gather settings from device capability queries, check checksum offload support, and submit send tasks carrying doca_buf packet buffers. It also classifies DOCA_ERROR_* failures across lifecycle, permission, capability, queue-full, driver, and steering layers, routing installation, steering, and cross-cutting debug issues to related skills.
- A C or C++ developer using a BlueField DPU or ConnectX NIC needs to bring up RX and TX queues on a physical port or representor.
- A packet-processing developer needs to select an RX queue type for fixed-size frames, buffer-ring layouts, or managed receive memory.
- A user-space forwarding or telemetry developer needs to submit packet send tasks through doca_eth_txq using doca_buf objects.
- A developer has a started RX queue with no packets, or receives DOCA_ERROR_AGAIN while sending at high rate.
- A Rust, Go, or Python binding maintainer needs lifecycle, capability, permission, and error rules for wrapping the DOCA Ethernet C ABI.
How do you install this skill?
- Before executing advice involving sudo, promiscuous mode, DOCA Flow, port state, or hardware/driver diagnostics, confirm the target device, authorization, impact, and a recoverable procedure.
- Verify the installed DOCA version, doca-eth/doca-common consistency, device capability-query results, and all companion files; do not infer availability from documentation claims, aggregate benchmark results, or publisher provenance.
- The source material contains inconsistent SKILL.md paths, SKILLCARD source metadata, repository names, and license declarations; publication should normalize the metadata and provide reproducible test data.
- Chinese-language output behavior and mainland-China reachability of external DOCA documentation are not established; the local SDK and shipped samples should be treated as the primary sources.
- Shell / CLI
- Local filesystem
DOCA SDK installed at /opt/mellanox/docaLinux Ubuntu 22.04/24.04 or RHEL/SLESBlueField DPU or ConnectX NICpkg-config
Install the skill from NVIDIA/skills with the skills CLI:
npx skills add nvidia/skills --skill doca-eth --yes
The CLI prompts for an installation destination; use --agent codex, --agent claude-code, --agent cursor, or --agent kiro-cli to target a client. The skill becomes available when an agent loads it for a relevant task. The source does not specify a manual-copy destination.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/doca-eth" ~/.claude/skills/
rm -rf "$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:
- Configure a doca_eth_rxq and doca_eth_txq on one BlueField representor and troubleshoot why the RX queue receives no packets.
First confirm that DOCA is installed, then give the agent a concrete Ethernet queue request such as: “Configure a doca_eth_rxq and doca_eth_txq on one BlueField representor and troubleshoot why the RX queue receives no packets.” The skill uses TASKS.md for configure, build, modify, run, test, and debug workflows, and CAPABILITIES.md for queue modes, RX types, send interfaces, capability queries, and error taxonomy. Route flow rules, DOCA installation, host-to-DPU messaging, and RDMA questions to the named related skills.
What are this skill's strengths and limitations?
- Covers RX/TX lifecycle, RX-type taxonomy, capability discovery, send tasks, and error classification.
- Clearly separates the DOCA Ethernet queue surface from the DOCA Flow steering surface.
- Provides language-neutral guidance for C/C++ and FFI consumers of the same C ABI.
- Accounts for installed DOCA version and device-specific capability checks.
- Requires the DOCA SDK to be preinstalled and the user to have permission to open a doca_dev.
- Requires a BlueField DPU or ConnectX NIC; the stated Linux coverage is Ubuntu 22.04/24.04 or RHEL/SLES.
- Ships no directly buildable application source, standalone build manifest, bindings, or samples subtree.
- The supplied material does not include the contents of TASKS.md or CAPABILITIES.md and does not establish testing across other platforms or hardware combinations.
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 Ethernet Queue Development this page ✓ NVIDIA · Official | 55 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| DOCA Common Foundation ✓ NVIDIA · Official | 55 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| DOCA RDMA Programming Guide ✓ NVIDIA · Official | 56 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| DOCA SHA Acceleration ✓ NVIDIA · Official | 56 · Use with care | ★ 3.6k | 13d ago | Apache-2.0 |
| DOCA Hardware Telemetry Counter Reader ✓ NVIDIA · Official | 60 · Recommended | ★ 3.6k | 13d ago | Apache-2.0 |
The skill explicitly positions DOCA Ethernet as the queue surface and DOCA Flow as the steering surface. DOCA installation is routed to doca-setup, host-to-DPU control messaging to doca-comch, and RDMA data movement to doca-rdma.
How did FollowSkills review this skill?
The material clearly scopes the skill to DOCA Ethernet queues, requires capability, permission, and port-state checks, and routes installation, steering, and RDMA elsewhere. It notes that sudo or mlnx-group access may be needed, but lacks explicit per-action user confirmation, rollback guidance, and complete data-flow disclosure. No malware or credential-theft red line is evident, so the score remains relatively high but is reduced for incomplete authorization and recovery boundaries.
SKILL.md, CAPABILITIES.md, and TASKS.md provide a coherent loading, configuration, run, test, and debug structure with error categories and escalation paths. However, the workflow depends on a local DOCA SDK, specific hardware, and companion skills not supplied here, and some claims are stated too absolutely. No key path was executed during this static review, so the score is capped below 10 by the calibration rule.
The audience, BlueField/ConnectX environment, queue types, trigger phrases, and non-fit boundaries are unusually explicit, and the material addresses C/C++ and FFI consumers. It nevertheless requires Linux, a preinstalled DOCA SDK, and actual hardware; Chinese-language output support is not specified, and several capabilities depend on local device and version state. These limitations reduce the score.
The information architecture is strong, with a loader, capability reference, task workflows, evaluation artifacts, and a skill card. Apache-2.0 and related ownership, issue, and security paths are described. However, version fields remain TBD-at-sign-time, the SKILLCARD path and license metadata conflict with the assessed material, the benchmark lacks its source dataset, and the report records schema issues plus missing author and recommended sections. These governance and maintenance gaps warrant deductions.
The skill gives directly useful decision rules for queue creation, RX-type selection, capability queries, TX submission, and error isolation, covering the core task with meaningful anti-misuse guidance. It is explicitly a thin loader, supplies no runnable code or complete examples, depends on companion files not included in the assessment, and provides no static proof that outputs are directly usable. The score therefore uses the static maximum of 7.
Evidence includes committed workflows, evaluation configuration, benchmark summaries, and a skill card, but the evaluation dataset is unavailable and the reported results are aggregate figures without reproducible coverage. External references and version/license claims are present, yet they do not establish independent corroboration. Because this is a static review with no execution, the score is limited to 5.
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 →