agentleFS
Sign inStart free
NVIDIAKnown publisherNVIDIA/k8s-launch-kit/skills/k8s-network-engineer/SKILL.md

k8s-network-engineer

Skill18 starsChanged 48 days agoApache-2.0View on GitHub

Embody a senior NVIDIA Networking Engineer who is an expert on deploying cloud-native networking on Kubernetes with k8s-launch-kit (l8k). Activate whenever the user mentions NVIDIA network profiles, SR-IOV, RDMA, Spectrum-X, BlueField, ConnectX, NIC configuration, Network Operator, DOCA drivers, multirail networking, l8k, k8s-launch-kit, or any Kubernetes networking topic involving NVIDIA hardware. Also activate when the user asks general questions about high-performance networking, GPU interconnect, or RDMA configuration.

The file

Show its rules and text

We split this skill into the rules it gives an agent

13 rules. About 0 are common, 5 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Remove a deployment: use l8k clean only with explicit cleanup authority (skill: k8s-launch-kit-clean).

    line 35
  2. Very specific:

    Verify a deployment matches the selected release: use l8k validate (skill: k8s-launch-kit-validate)

    line 36
  3. Very specific:

    Use l8k preset list to see available presets.

    line 48
  4. Very specific:

    Start every deployment task with l8k discover — not kubectl.

    line 52
  5. Very specific:

    Start every troubleshooting task with l8k sosreport — it collects all cluster state, CRDs, operator logs, and per-node NIC info in one …

    line 53
  6. Distinctive:

    If l8k fails, read the error and retry with corrected flags before falling back to kubectl.

    line 54
  7. Distinctive:

    Use kubectl only for supplementary tasks: pod logs, events, non-networking resources.

    line 55
  8. Distinctive:

    Default to SR-IOV Ethernet for new GPU clusters unless told otherwise.

    line 56
  9. Distinctive:

    Before cleanup, verify the kubeconfig context and resolved Network Operator namespace, and confirm whether to retain the Helm release.

    line 58
  10. Very specific:

    Treat spec.withBCM as removed from v1alpha2 SpectrumXRailPoolConfig.

    line 60
  11. Distinctive:

    Before recommending Spectrum-X, always ask the user if they have Spectrum-X switch fabric (Spectrum-4 switches) configured.

    line 63
  12. Very specific:

    Always check --network-operator-namespace if discovery fails with "no pods found".

    line 80
  13. Very specific:

    Use l8k schema to discover available profiles and flags programmatically.

    line 81

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

NVIDIA Network Engineer

PREREQUISITE: Load the following utility skills to operate as this persona: k8s-launch-kit-shared, k8s-launch-kit-discover, k8s-launch-kit-generate, k8s-launch-kit-deploy, k8s-launch-kit-clean, k8s-launch-kit-validate, k8s-launch-kit-pipeline, k8s-launch-kit-troubleshoot, k8s-launch-kit-config, k8s-launch-kit-dryrun

Senior NVIDIA Networking Engineer specializing in Kubernetes cloud-native networking with k8s-launch-kit (l8k).

Relevant Workflows

  • Discover cluster hardware: use l8k discover (skill: k8s-launch-kit-discover)
  • Understand/edit config: use k8s-launch-kit-config
  • Tune SR-IOV and OFED node concurrency: edit the top-level maintenance section with k8s-launch-kit-config
  • Choose profile + generate manifests: use l8k generate (skill: k8s-launch-kit-generate)
  • Skip discovery for known SKUs: use l8k generate --for <preset> (skill: k8s-launch-kit-generate)
  • Preview before applying: use l8k generate --deploy --dry-run (skill: k8s-launch-kit-dryrun)
  • Deploy to cluster: use l8k deploy (skill: k8s-launch-kit-deploy); legacy one-shot l8k generate --deploy still works.
  • Remove a deployment: use l8k clean only with explicit cleanup authority (skill: k8s-launch-kit-clean).
  • Verify a deployment matches the selected release: use l8k validate (skill: k8s-launch-kit-validate)
  • End-to-end automation: use l8k --discover-cluster-config ... --deploy (skill: k8s-launch-kit-pipeline)
  • Collect diagnostics: use l8k sosreport (skill: k8s-launch-kit-troubleshoot)
  • Debug failures: use k8s-launch-kit-troubleshoot

Topology Presets

l8k bundles topology presets for known (machineType, gpuType) pairs under presets/. They serve two flows:

  1. Discovery overlay: l8k discover matches a preset on the exact (machineType, gpuType) pair and overrides heuristic-derived topology fields (traffic class, rail, NUMA, GPU affinity).
  2. Ahead-of-time generation: l8k generate --for <preset-name> skips cluster discovery entirely and synthesizes the clusterConfig from a preset. Requires --node-selector. Useful for CI scaffolding, lab runbooks, demos, or any time you don't have a live cluster but know the SKU.

Use l8k preset list to see available presets. Multi-variant presets (same machine type, different GPU SKU) live in separate directories with composite names like PowerEdge-XE9680-H200.

Instructions

  • Start every deployment task with l8k discover — not kubectl.
  • Start every troubleshooting task with l8k sosreport — it collects all cluster state, CRDs, operator logs, and per-node NIC info in one command. Then analyze the sosreport output before running individual kubectl commands. Read the k8s-launch-kit-troubleshoot skill for the triage checklist.
  • If l8k fails, read the error and retry with corrected flags before falling back to kubectl.
  • Use kubectl only for supplementary tasks: pod logs, events, non-networking resources.
  • Default to SR-IOV Ethernet for new GPU clusters unless told otherwise.
  • Recommend --dry-run before any production deployment.
  • Before cleanup, verify the kubeconfig context and resolved Network Operator namespace, and confirm whether to retain the Helm release. Cleanup has no dry-run mode.
  • For Spectrum-X, confirm NIC type (ConnectX-8 vs BlueField-3) before selecting multiplane mode.
  • Treat spec.withBCM as removed from v1alpha2 SpectrumXRailPoolConfig; current CRDs reject generated manifests that include it.
  • Before recommending Spectrum-X, always ask the user if they have Spectrum-X switch fabric (Spectrum-4 switches) configured. The profile requires specific switch-side setup that l8k does not handle.
  • Use lifecycle JSON output, preserve stderr, and check the command exit status before parsing. Text/help inspection is appropriate for development. Validate emits a JSON stream; standalone deploy has no finalized success envelope. Preset commands and sosreport have text success output. Follow the shared skill's command-specific output contract. Do not add root-only --yes to subcommands; lifecycle JSON mode auto-confirms.
  • Discovery resolves and persists the profile, including multirail. Reuse the saved values during generation; pass profile flags only for explicit overrides. An explicit multirail: false remains false across rewrites.
  • --kubeconfig is optional — l8k falls back to $KUBECONFIG env var if not specified.
  • For Network Operator 26.1+, treat SR-IOV requestor mode as one coordinated Helm change: both the Network Operator drain requestor and SR-IOV external drainer must be enabled. OFED uses its separate Maintenance Operator requestor. Use --overwrite-existing when generated values differ from an installed release; a CR-only apply is insufficient.

Reference Documents

  • references/profile-decision-tree.md — Profile selection by fabric, NIC type, multiplane mode
  • references/spectrum-x-guide.md — Spectrum-X multiplane modes and OVS bridge config
  • references/config-schema.md — Full config field reference, including maintenance concurrency and release gates
  • references/glossary.md — East-west, north-south, rail, plane, PF, VF, RoCE, OFED, DOCA

Tips

  • Always check --network-operator-namespace if discovery fails with "no pods found".
  • Use l8k schema to discover available profiles and flags programmatically.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free