Skip to content
Get packs

Guide · Agent skill

How to use the GKE Golden Base Image Finder agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

This guide explains how to install and use the gke-custom-golden-image-discovery agent skill, a set of instructions that lets an AI agent find the right golden base image for a custom Google Kubernetes Engine (GKE) node image. It is aimed at platform and DevOps engineers who know what their workload needs, such as GPUs, ARM chips or gVisor sandboxing, but not the exact image name. By the end you will know how to install the skill, ask for what you need in plain language, and check the result.

GKE Golden Base Image Findergoogle/skills · Apache-2.0

Quick start

  1. 1Download the ZIP. Download the skill ZIP from the skill page on this site and unzip it.
  2. 2Place the folder. Move the unzipped folder into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
  3. 3Check prerequisites. Make sure the gcloud CLI and curl are available, and that the agent can read references/attribution.md.
  4. 4Ask in plain language. Describe your cluster: GKE version, OS, architecture, accelerators, gVisor or cgroup needs.
  5. 5Review the inferred settings. Check the defaults the agent assumed and correct any that do not match your need.
  6. 6Verify the base image name. Confirm the returned image name against the node-to-base-image map for your GKE minor version before use.

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file of instructions that an AI agent loads when your request matches the skill's description. This skill is for building custom GKE node images, which need a pre-approved golden base image to start from.

You describe the cluster you want in plain language, including GKE version, operating system, CPU architecture, GPUs or TPUs, gVisor sandboxing and cgroup mode. The agent turns that into technical settings, fetches Google's published node-to-base-image map for your GKE minor version, and returns the matching base image name. It also tells you which settings it assumed when you did not specify them.

  • Maps plain-language requirements to settings such as image_family, architecture, accelerators, gVisor and cgroup mode.
  • Looks up the matching base image name in Google's published image map.
  • Reports the defaults it assumed, and tries a fallback (cgroup v1 or gVisor off) if no exact match is found.

Install the skill with a ZIP download or the Claude apps

Download the ZIP from the skill page on this site and unzip it. For a personal setup, place the unzipped folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

  • Personal install: ~/.claude/skills/ (available across your projects).
  • Project install: .claude/skills/ inside the project folder.
  • Claude apps: Settings → Capabilities → Skills, then upload.

Check what the skill needs before you start

The skill needs the gcloud CLI for any gcloud commands the agent runs, and curl or an equivalent HTTP client to fetch the node-to-base-image JSON map from gstatic.com. It also needs read access to references/attribution.md, which is bundled with the skill.

The skill does not run scripts. It does use the network, because the image map is fetched live.

  • gcloud CLI, for any gcloud commands the agent runs.
  • curl or an equivalent HTTP client, to fetch the JSON map from gstatic.com.
  • Read access to references/attribution.md (bundled).

Trigger the skill with natural requests

You do not need a special command. Ask in plain language about a golden base image, custom GKE node creation, or your cluster configuration. The skill is designed to trigger on requests about finding the golden base image, mapping cluster parameters to image names, or querying GKE base image maps.

Include your GKE version if you have it. The skill requires a GKE version of 1.34.1-gke.2909000 or later and warns you if the version is unsupported. If you leave out other details, the skill applies its defaults.

  • Give the GKE version when you can, for example 1.34.1-gke.2909000.
  • Mention hardware (GPU, TPU, T4, A100, H100, L4) or ARM/Ampere if relevant.
  • Mention gVisor, a sandbox or signed modules if your security setup needs them.

Understand the defaults the skill assumes

When you do not specify a value, the skill infers it. Operating system defaults to COS, architecture to X86_64, gVisor to false, accelerators to false, signed modules to false, TPU to false, and cgroup mode to CGROUP_MODE_V2. Cgroup v1 is used only if you explicitly ask for legacy behaviour.

Read the assumptions the agent reports back. If a default does not match your real need, say so and ask again with the correct setting.

  • OS: COS by default; Ubuntu if you ask for Ubuntu or standard Linux.
  • Architecture: X86_64 by default; ARM64 if you mention ARM or Ampere.
  • Cgroup mode: CGROUP_MODE_V2 by default.

Example request and what to expect

A request such as the following gives the agent enough to work with. The skill's own example describes a setup on GKE 1.34.1-gke.2909000 with COS and new H100 GPUs. The agent infers X86_64 with accelerators enabled and returns a base image name in the form gke-1341-gke2909000-cos-125-19216-0-115-c-nvda.

Treat the name as a lookup result, not a guarantee. Confirm it against the image map for your version before you build anything.

  • Example prompt: I need a golden base image for GKE 1.34.1-gke.2909000 with COS and H100 GPUs.
  • Expected reply: the inferred settings, then the base image name from the map.
gke-1341-gke2909000-cos-125-19216-0-115-c-nvda

Know the safety limits and when to use another skill

Before running any gcloud, REST or MCP command, the agent must read references/attribution.md. For terminal commands it prefixes gcloud with CLOUDSDK_METRICS_REQUEST_ATTRIBUTION set to agent-skills/1.0 (skill:gke-custom-golden-image-discovery), and adds the same string to User-Agent headers on direct HTTP calls. The tag identifies only the skill and its version and carries no user data.

This skill is not for general cluster creation or standard node pool management. The skill itself points those tasks to gke-cluster-creation or gke-basics.

  • Runs scripts: no. Uses network: yes, to fetch the image map from gstatic.com.
  • Do not use gcloud config set for the attribution tag, since it would persist beyond the task.
  • Use gke-cluster-creation or gke-basics for general cluster and node pool work.

What people use it for

  • Finding the COS base image for a custom node image with H100 GPUs on a given GKE version.
  • Locating an ARM64 (Ampere) base image for a custom node pool.
  • Checking whether a specific GKE version has a base image for a gVisor-enabled configuration.
  • Mapping a list of cluster parameters to the exact base image name for a Terraform or script review.

Frequently asked questions

How do I install the gke golden base image agent skill?

Download the ZIP from the skill page on this site and unzip it. Put the folder in ~/.claude/skills/ for personal use, or in a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.

What do I need to run this skill?

You need the gcloud CLI for any gcloud commands the agent runs, curl or an equivalent HTTP client to fetch the image map from gstatic.com, and read access to the bundled references/attribution.md. The skill does not run scripts but does use the network.

Is the GKE golden base image finder free to use?

The source does not state pricing or licence terms beyond the skill's Apache-2.0 licence, so check the licence in the repository before you rely on it. The gcloud CLI and the image map fetch are separate tools whose terms are not described here.

Which GKE versions does the skill support?

The skill requires a GKE version of 1.34.1-gke.2909000 or later. If you give an invalid or unsupported version, the agent tells you it is unsupported.

Why did the skill return a different image than I expected?

The skill applies defaults when you do not specify a value, such as COS, X86_64 and cgroup v2. If a default does not match your workload, state the correct setting. If no exact match exists, it tries cgroup v1 or gVisor off and tells you it did so.

Does this skill create GKE clusters?

No. It is for finding the golden base image for custom node images. For general cluster creation or standard node pool management, the skill points you to gke-cluster-creation or gke-basics.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

This skill suits platform and DevOps engineers who need the exact golden base image name for a custom GKE node image and can verify the result. It is not a guaranteed-correct tool, because it infers defaults that may not match your real need and fetches the image map live. If you need general cluster or node pool work, use the skills the source points to instead.

Download the skill