Guide · Agent skill

How to use the GKE ComputeClasses Helper agent skill

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

The gke-compute-classes skill is an add-on guide for an AI agent that helps you write and review Google Kubernetes Engine (GKE) ComputeClass YAML, which tells a cluster what kinds of machines to create for your workloads. After this guide you will know what the skill does, how to install it from the ZIP or the Claude apps, how to phrase requests so the agent uses it, and what safety limits to keep in mind before you apply anything.

GKE ComputeClasses Helpergoogle/skills · Apache-2.0 Download the skill

Quick start

1Download and unzip. Download the skill ZIP from this page and unzip it so the gke-compute-classes folder is intact.
2Place the folder. Copy it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder, or upload the ZIP under Settings → Capabilities → Skills in the Claude apps.
3Check the requirements. Confirm that gcloud, jq and kubectl are installed and that you have a GKE cluster and a project with the right IAM permissions.
4Ask a specific question. Describe your goal in plain words, for example a Spot workload with on-demand fallback, including your zone and machine family constraints.
5Review before deploying. Check the draft against your commitments and zones, remove nothing you did not understand, and test in a non-production cluster first.

What does the gke-compute-classes skill make the agent do?

An agent skill is a folder of instructions and supporting files that an AI agent loads when a request matches the skill's description. This one covers GKE ComputeClasses. Its stated focus is configuring, optimizing and troubleshooting them, mainly for Spot VMs with on-demand fallback, specific accelerators (GPUs and TPUs), specific machine families, restricting ComputeClass access, and pending pods tied to node pool auto-creation.

The skill's own assessment says the agent can draft example ComputeClass YAML, explain scheduling rules such as taints and tolerations, and help debug pods stuck in Pending. It also ships example YAML templates and a shell script that live-tails cluster autoscaler events using gcloud and jq.

Draft example ComputeClass YAML for Spot with on-demand fallback, accelerators or machine families
Explain taints, tolerations and related scheduling rules
Help debug pods stuck in Pending or showing noScaleUp
Provide example templates in the assets folder, including a GPU inference example and an RBAC example for editing ComputeClasses
Out of scope: cluster-level Node Auto Provisioning and general GKE cluster creation

How do you install the gke-compute-classes skill?

Download the skill ZIP from this page and unzip it. For a personal setup, place the unzipped gke-compute-classes folder inside ~/.claude/skills/. For a single project, place it inside that project's .claude/skills/ folder so it applies only there.

If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the ZIP there.

Personal install: ~/.claude/skills/gke-compute-classes/
Project install: .claude/skills/gke-compute-classes/ inside your project
Claude apps: Settings → Capabilities → Skills → upload the ZIP

What do you need before the skill is useful?

The skill lists its requirements plainly. You need the gcloud CLI, jq, and kubectl (the skill uses kubectl for cluster context inference). You also need an existing GKE cluster and a Google Cloud project, and an account with appropriate IAM permissions. The skill assumes you already know Kubernetes and gcloud; it is not a beginner course.

Before running any gcloud, REST API or MCP command, the skill's rules require the agent to read its Usage Attribution reference. For commands typed directly into a terminal, the skill asks that gcloud be prefixed with a specific CLOUDSDK_METRICS_REQUEST_ATTRIBUTION value.

gcloud CLI
jq
kubectl
A GKE cluster and Google Cloud project with suitable IAM permissions

CLOUDSDK_METRICS_REQUEST_ATTRIBUTION="agent-skills/1.0 (skill:gke-compute-classes)" gcloud

How do you trigger the skill with a natural request?

You do not need a special command. Describe the goal in plain words that match the skill's use cases. The skill is meant to load for requests about Spot VMs with on-demand fallback, targeting GPUs or TPUs, choosing a machine family, restricting who can use a ComputeClass, or a pod stuck Pending after node pool auto-creation.

The skill's guidance is designed to give a first draft quickly and then refine it. Expect placeholders such as <YOUR-ZONE-HERE> and a reminder to align machine families with your committed use discounts or reservations.

Spot VMs with a fallback to on-demand machines
Use L4 or H100 GPUs for my inference pods
Cap our committed N4 cores and spill extra demand to other families
Why is my pod Pending with noScaleUp?
Restrict which teams can use this ComputeClass

What should you check in generated YAML?

Treat every generated template as a draft. The skill requires that initial YAML be labelled EXAMPLE TEMPLATE - DO NOT DEPLOY, and that machine family fields carry a comment telling you to align them with your existing CUDs or reservations. Review zone choices and machine families against what you actually have.

The skill's rules also say a GPU pod that stays Pending is often missing a toleration for the nvidia.com/gpu taint. Check that first when debugging accelerator pods. Spot tolerations are a different case: the skill says to verify them on your own cluster rather than assume they are needed.

Confirm the EXAMPLE TEMPLATE label has been removed only after review
Align machineFamily with your committed use discounts or reservations
Check zones against your own availability
Keep any intentional dedication taint and its matching toleration
Do not treat a Pending pod as a missing Spot toleration unless the events name that taint

tolerations:
- key: nvidia.com/gpu
  operator: Exists

What about the autoscaler event script?

The assets folder includes log-autoscaler-events.sh, which live-tails cluster autoscaler events using gcloud and jq. Use it to watch what the autoscaler reports while a pod is waiting. Read the script before running it, because it runs commands and uses the network. The skill does not document its arguments in the material provided here, so check the script itself for options.

Script path: assets/log-autoscaler-events.sh
Requires gcloud and jq
Runs commands and uses the network

What people use it for

Drafting a ComputeClass that prefers Spot VMs and falls back to on-demand machines
Creating a GPU ComputeClass for inference, including the toleration a GPU pod needs
Capping a committed machine family to your CUD core count while letting excess demand spill to other families
Investigating a pod stuck in Pending with noScaleUp by reading autoscaler events

Frequently asked questions

How do I install the gke-compute-classes skill?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into your project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

Is the gke-compute-classes skill free and what licence does it use?

The skill is listed under the Apache-2.0 licence in the google/skills repository. The page does not state a price, so check the repository for any terms beyond the licence.

What do I need to run the gke-compute-classes skill?

You need the gcloud CLI, jq and kubectl, plus an existing GKE cluster and Google Cloud project with suitable IAM permissions. The skill assumes Kubernetes and gcloud experience.

Why is my GKE pod stuck in Pending with noScaleUp?

The skill's guidance says a GPU pod stuck Pending is often missing a toleration for the nvidia.com/gpu:NoSchedule taint, so add tolerations with key nvidia.com/gpu and operator Exists. For other cases, run the autoscaler event script to see what the autoscaler reports.

Does the skill set up cluster-level Node Auto Provisioning?

No. The skill states that cluster-level Node Auto Provisioning and general GKE cluster creation are out of scope. It describes ComputeClass node pool auto-creation as a separate feature.

Is it safe to apply the YAML the agent generates?

Not without review. The templates are labelled EXAMPLE TEMPLATE - DO NOT DEPLOY, and the machine family choice depends on your own commitments and zone availability. The skill also runs scripts and uses the network, so read them first.

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

Use this skill if you run GKE, know Kubernetes and gcloud, and need ComputeClass YAML drafted or reviewed, or need help with pods stuck in Pending. Pick something else if you need cluster creation or cluster-level Node Auto Provisioning, or if you are new to GKE and want a step-by-step introduction.

Download the skill