What does this skill do, and when should you use it?
This skill teaches AI agents how to use NVIDIA’s doca_flow_tune to tune a DOCA Flow pipeline. It covers snapshots of live or captured state, three tuning axes—rule placement, resource hints or table sizing, and hardware-offload mode—and matching measurements such as rule-install rate, lookup latency, and hardware-counter delta. It supports online and offline workflows and interpretation of CSV, JSON, and Mermaid outputs. It also guides minimum-diff changes back into the Flow program, with smoke validation before state-changing application.
Reads the local DOCA installation, shipped configuration templates, and available scripts; attaches to a running doca-flow application or consumes captured JSON/CSV files; guides the dump, monitor, web, analyze, and visualize workflow; interprets dumper, monitor, analyzer, and visualizer outputs; selects a tuning axis and measurement; proposes an optimization; and routes the recommendation back into the surrounding Flow program for controlled testing.
- A BlueField operator wants a read-only snapshot of existing pipes and hardware or software counter progress before recommending a change.
- A performance engineer has a doca-flow-perf baseline and needs to choose between rule placement, table sizing, and hardware-offload tuning.
- A DOCA Flow developer wants to inspect a captured pipeline through offline analysis and visualization without re-instrumenting the application.
- An operator needs to diagnose attachment failures, empty output, degenerate Mermaid diagrams, or disagreement between analysis and live counters.
- A developer needs to apply a recommendation to a C/C++ Flow program and validate it with warm-up, steady-state, and before/after checks.
How do you install this skill?
- Before execution, verify the local DOCA version, installed templates, --help output, Flow application, Unix socket, device scope, and full configuration; the skill does not provide a fixed command contract.
- Any Flow-program change or online state-changing action may affect the dataplane; obtain explicit confirmation and retain a recoverable deployment/rollback plan.
- Do not externally share dumper, analyze, visualize, or log artifacts without redaction; the documentation requests full unredacted capture but does not define sensitive-data handling.
- Core use depends on the NVIDIA DOCA installation, BlueField/ConnectX hardware, and external documentation; reachability from mainland-China networks is not established.
- The benchmark dataset is unavailable, and the SKILLCARD contains TBD fields plus path/license metadata inconsistencies.
- Shell / CLI
- Local filesystem
DOCA SDK at /opt/mellanox/docaLinuxBlueField DPU or ConnectX NICrunning or captured doca-flow applicationNode.js/npx for the documented CLI installation
Install the skill with the documented Skills CLI command: npx skills add nvidia/skills --skill doca-flow-tune --yes. The repository README says the skill becomes available when an agent later encounters a relevant task. Actual use additionally requires the DOCA SDK at /opt/mellanox/doca, Linux, an attached BlueField DPU or ConnectX NIC, and a running or captured doca-flow application.
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/doca-flow-tune" ~/.claude/skills/
rm -rf "$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:
- My Flow rule-install rate is low on BlueField; use doca_flow_tune to determine whether rule placement or table sizing is the likely axis.
Ask an agent a concrete tuning question, for example: “My Flow rule-install rate is low on BlueField; use doca_flow_tune to determine whether rule placement or table sizing is the likely axis.” The skill directs the agent to TASKS.md, starting with configure and continuing through run and test, while using the installed version’s help and flow_tune_cfg*.json templates as sources for the evolving command and schema surface. Do not use it for DOCA installation, Flow application development, or baseline measurement; those are explicitly out of scope.
What are this skill's strengths and limitations?
- Covers both live observation and offline captured-snapshot analysis.
- Separates tuning axes from measurements, which supports baseline-driven optimization.
- Provides a snapshot-to-analysis-to-visualization workflow with recommendation and smoke-test guidance.
- Supports read-only operation while explicitly flagging state-changing application as high stakes.
- Targets operators, performance engineers, Flow developers, and AI agents.
- Requires a specific DOCA SDK installation, Linux, compatible NVIDIA hardware, and an existing doca-flow application.
- Does not create Flow pipes or replace DOCA setup, Flow API development, or baseline measurement skills.
- Does not ship a fixed flag inventory, output schema examples, wrappers, parsers, or pre-baked recommendations.
- Recommendations and outputs are installation-, device-, firmware-, NUMA-, pipeline-, and DOCA-version-specific.
- The supplied material provides no independent test results or platform coverage matrix.
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 Flow Tune this page ✓ NVIDIA · Official | 54 · Use with care | ★ 3.8k | today | Apache-2.0 |
| DOCA Compress Hardware Offload ✓ NVIDIA · Official | 58 · Recommended | ★ 3.8k | today | Apache-2.0 |
| DOCA Flow DPA Performance ✓ NVIDIA · Official | 58 · Recommended | ★ 3.8k | today | Apache-2.0 |
| DOCA AES-GCM Acceleration ✓ NVIDIA · Official | 57 · Use with care | ★ 3.8k | today | Apache-2.0 |
| DOCA SHA Acceleration ✓ NVIDIA · Official | 56 · Use with care | ★ 3.8k | today | Apache-2.0 |
The repository explicitly distinguishes this skill from doca-flow-perf: doca-flow-perf measures baselines, while doca-flow-tune recommends optimizations on top of them. doca-flow-dpa-perf is the relevant baseline tool for a DPA-offloaded path. doca-flow is the base library whose pipelines this tool observes and tunes; this skill does not teach the Flow API or create the application.
How did FollowSkills review this skill?
The documentation specifies read-only-by-default operation, confirmation before changes, snapshots, before/after comparison, and data-flow surfaces such as the Unix socket, configuration, logs, and live state. Least-privilege details, rollback/recovery procedures, and log redaction are incomplete, and state changes depend on the surrounding Flow application, so points are deducted.
The workflow, prerequisites, and layered error taxonomy are relatively thorough and provide diagnostic routing. However, this is a static review: commands, dependencies, key paths, and abnormal-input behavior were not executed or reproduced; exact CLI and schema details are intentionally externalized, so the score stays conservative under the static cap.
Audience, trigger phrases, scope, and exclusions are clearly stated for live or captured offline/online scenarios. The skill nevertheless requires a specific DOCA, BlueField/ConnectX environment; exact parameters require the installed version and external documentation; Chinese interaction and mainland-China reachability are not evidenced, so points are deducted.
The information architecture, loading order, task routing, limitations, and related-skill guidance are strong, and SKILL.md declares Apache-2.0. Versioning, maintenance ownership, and change history are incomplete; the SKILLCARD contains TBD fields and path/license inconsistencies, and the benchmark reports missing recommended Instructions/Examples sections, so points are deducted.
The skill covers the core configuration, snapshot, analysis, visualization, debugging, and recommendation-to-code workflow and emphasizes before/after evidence. It supplies no runnable wrappers, fixed representative outputs, or complete commands; outcomes depend on the local DOCA installation and surrounding application, so static evidence supports only a conservative score.
The files contain task procedures, an error taxonomy, an evaluation report, and internal cross-references. The evaluation dataset is unavailable, the report is self-reported, and no independently reproducible CI/test coverage for the key paths is included; important contracts depend on the installed environment and external guides, so evidence is limited.
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 →