How to use the GKE Batch & HPC Workloads agent skill

What is an agent skill?

An agent skill is a folder containing a SKILL.md file. The file tells an AI agent when to use the skill and what reference knowledge to follow. When your request matches the skill's description, the agent loads these instructions and uses them while it works.

This skill's SKILL.md covers batch jobs, JobSet, Kueue queues, compact placement node pools, MPI workloads and cost options. It also includes a references/attribution.md file that the agent must read before running gcloud commands.

What does the skill make the agent do?

With the skill loaded, the agent can write Kubernetes Job manifests with parallelism, completions and backoffLimit settings, and JobSet manifests for multi-part workflows. It can install Kueue and define a ClusterQueue and LocalQueue so several teams share cluster capacity fairly.

For tightly coupled HPC work, the agent can create a GKE node pool with compact placement using gcloud, and run MPI jobs by installing the MPI Operator and writing an MPIJob. The skill also points the agent to Spot VMs and scale-to-zero for lower cost, and reminds it to set resource requests and limits and to add application-level checkpointing so jobs survive interruptions.

How do you install the skill?

Download the skill ZIP from the item page and unzip it. To make it available to Claude Code for all your projects, place the unzipped gke-batch-hpc folder inside ~/.claude/skills/. To limit it to one project, place it inside that project's .claude/skills/ folder.

In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/gke-batch-hpc/
• Project install: .claude/skills/gke-batch-hpc/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

How do you trigger the skill with a natural request?

You do not need to name the skill. Describe a batch or HPC task in plain words that matches the skill's description, such as running batch jobs or setting up job queues on GKE. Mention GKE explicitly, because the skill is about that platform.

If you ask for a standard web app deployment, the skill's own rules say to use the gke-app-onboarding skill instead.

• Set up a Kueue job queue on my GKE cluster so three teams share 100 CPUs fairly.
• Write a Kubernetes Job that processes 100 files with 10 workers and retries failures up to 3 times.
• Create a compact placement node pool on GKE for my tightly coupled simulation.

What does the agent need before it can run anything?

The skill expects a Google Cloud account with a GKE cluster, the gcloud CLI and kubectl, and Kubernetes access to the target cluster. If you use the relevant examples, Kueue, JobSet and the MPI Operator must also be installed in the cluster. The skill refers to MCP Kubernetes tools such as apply_k8s_manifest and get_k8s_resource, so those need to be available to the agent.

Without a real cluster, the examples cannot run, and running them costs money.

What are the safety notes for scripts and network use?

This skill runs scripts and uses the network. The agent may run gcloud and kubectl commands that create or delete cloud resources. Do not point it at a production cluster unless you have reviewed the commands it proposes.

The attribution file requires gcloud commands run in the terminal to carry a metrics environment variable tagging the skill. The skill says this tag must be set inline on each command and not with gcloud config set. Commands written into generated scripts, Terraform or user code are not prefixed this way.

• Review every apply, create and delete command before it runs.
• Test on a non-production cluster first.
• Expect the agent to fetch the Kueue, JobSet and MPI Operator manifests from the URLs in the skill.