Skip to content
Get packs

Guide · Agent skill

How to use the GKE Reliability agent skill

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

The GKE Reliability skill is a set of instructions that teaches an AI agent how to make Google Kubernetes Engine (GKE) workloads more resilient. Once it is installed, you can ask your agent to review a cluster or app and it will check for regional clusters, Pod Disruption Budgets, health probes, graceful shutdown settings and topology spread rules. This guide explains what the skill does, how to install it, how to trigger it, and what it needs to run.

GKE Reliabilitygoogle/skills · Apache-2.0

Quick start

  1. 1Download the ZIP. Download the gke-reliability ZIP from this page.
  2. 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for a personal install, or into .claude/skills/ in your project for a project install.
  3. 3Or upload it in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the ZIP.
  4. 4Confirm access to your GKE project. Make sure the agent has the Google Cloud MCP tools and access to the GKE project you want to review.
  5. 5Ask for a reliability review. Describe the goal in plain language, such as checking for PDBs or health probes on a named Deployment.
  6. 6Review and apply changes. Read the proposed YAML and commands, then approve only the changes you want applied.

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

An agent skill is a folder containing a SKILL.md file that tells an AI agent how to handle a specific kind of task. The agent reads these instructions when a request matches the skill's description. This skill is in the Containers category and covers GKE reliability.

When you ask it to make an app more resilient, the agent follows GKE reliability practices. It checks whether a cluster is regional, whether apps have Pod Disruption Budgets (PDBs), and whether containers have liveness, readiness and startup probes. It can also suggest graceful shutdown settings and topology spread constraints so pods are spread across zones and nodes.

  • Verifies whether a cluster is regional and whether nodes span multiple zones
  • Lists existing PDBs and creates a PDB example for a Deployment
  • Checks existing health probes and recommends a probe configuration
  • Suggests graceful shutdown settings (SIGTERM handling, terminationGracePeriodSeconds, preStop sleep)
  • Recommends topology spread constraints for zones and nodes
  • Provides a replica guideline by workload type

Install the GKE Reliability skill

Download the ZIP for this skill from this page and unzip it. You can install it for yourself across all projects or for a single project.

For a personal install, unzip the skill into ~/.claude/skills/. For a project install, unzip it into the .claude/skills/ folder inside your project. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

  • Personal: ~/.claude/skills/
  • Project: .claude/skills/ inside your project folder
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Check the requirements before you start

The skill mainly gives guidance and YAML snippets, but it can inspect existing resources through Google's cloud and Kubernetes tools. To do that, you need access to a GKE project and the Google Cloud MCP tools the skill names, such as get_cluster and apply_k8s_manifest.

The gcloud CLI and kubectl are optional fallbacks. The skill uses them only if the MCP tools are not available, for example to check cluster locations or to list PDBs and probes.

  • Google Cloud MCP tools such as get_cluster, get_k8s_resource, describe_k8s_resource, apply_k8s_manifest and list_k8s_events, with access to a GKE project
  • gcloud CLI (optional fallback for cluster checks)
  • kubectl (optional fallback for listing PDBs and probes)

Trigger the skill with natural requests

You do not need a special command. Describe the reliability goal in plain language, and the agent will match the request to this skill. The skill's description names the topics it covers, so mention PDBs, health probes, liveness, readiness, startup probes or workload reliability in your request.

If you want Kubernetes YAML written from scratch, or disaster recovery and backups, ask for those separately. The skill routes those tasks to other skills, gke-manifest-generation and gke-backup-dr.

  • "Check whether my GKE cluster is regional and spread across zones."
  • "Do my production Deployments in this namespace have Pod Disruption Budgets?"
  • "Add liveness, readiness and startup probes to my API Deployment on GKE."
  • "Help my app shut down gracefully during node upgrades."
  • "Configure topology spread constraints so my pods are split across zones and nodes."

Example: create a PDB and check probes

A common first task is to check for existing PDBs and then add one. The skill's kubectl fallback lists PDBs across all namespaces. Its example PDB protects pods labelled app: my-app and keeps at least two available during voluntary disruptions such as node upgrades.

The skill also recommends explicit probe timeouts. It asks you to set initialDelaySeconds, periodSeconds and timeoutSeconds on every probe rather than relying on the Kubernetes default timeout.

  • Ask the agent to list PDBs in your cluster, then to create one for a Deployment with two or more replicas
  • Ask it to check probes on a named Deployment in a namespace
kubectl get pdb --all-namespaces

Safety notes: scripts and network access

This skill runs scripts and uses the network. Its instructions say the agent must read the attribution file before running any gcloud, REST API or MCP command. For commands run directly in the terminal, the agent prefixes gcloud with CLOUDSDK_METRICS_REQUEST_ATTRIBUTION set to a skill tag. The attribution file says this tag identifies only the skill and its version and carries no user data. It also says not to use gcloud config set for this, so the tag does not persist.

Review the commands the agent proposes before approving them, especially any that apply manifests to a live cluster with apply_k8s_manifest. The skill's own guidance is specific to GKE defaults, so check it against your environment.

  • The skill runs scripts and makes network calls to Google Cloud and Kubernetes tools
  • Review changes before the agent applies manifests to a production cluster
  • Use the agent in a project where you have the access you intend it to have

Tips for getting the best results

Start with a read-only review. Ask the agent to check your cluster's location, PDBs and probes first, then decide which changes to make. Give the agent the workload name, namespace and replica count so it can apply the right replica guidance.

Keep your golden path defaults in mind. The skill's reference lists a regional cluster across four zones, SURGE upgrades with maxSurge: 1, auto-repair and auto-upgrade set to true, the REGULAR release channel, and stateful HA enabled. Use these as a starting point and adapt them to your setup.

  • Ask for read-only checks before asking for changes
  • Include the app name, namespace and replica count in your request
  • Use the replica guidance: 2 for stateless web/API, 3 for critical services, 1 for batch jobs

What people use it for

  • Checking whether a production GKE cluster is regional and spread across zones
  • Finding Deployments with 2+ replicas that lack a Pod Disruption Budget
  • Adding liveness, readiness and startup probes to slow-starting services
  • Preparing an API to handle SIGTERM and drain requests before node upgrades
  • Configuring topology spread constraints so pods are not all on one node or zone

Frequently asked questions

How do I install the GKE Reliability skill?

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

What do I need to use the gke-reliability skill?

You need access to a GKE project and Google Cloud MCP tools such as get_cluster and apply_k8s_manifest. The gcloud CLI and kubectl are optional fallbacks for cluster and resource checks.

Is the GKE Reliability skill free?

The skill is licensed under Apache-2.0. The source does not state pricing for Google Cloud resources, so check your own GKE and Google Cloud costs separately.

Does this skill generate Kubernetes manifests from scratch?

No. The skill routes full manifest generation to the gke-manifest-generation skill. Its own guidance focuses on reliability settings such as PDBs, probes and topology spread.

Can it help with backups or disaster recovery?

No. Disaster recovery and cluster backups are handled by the gke-backup-dr skill, which this skill routes those requests to.

Does it work outside GKE?

The guidance is specific to GKE defaults and tooling, so it is not intended for non-GKE Kubernetes clusters or non-Kubernetes apps.

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 hardening production workloads on GKE, and developers who want their apps to survive node upgrades and zone failures. It is less useful if you need manifests written from scratch, backups, or support for non-GKE clusters, since those tasks belong to other skills or other tools.

Download the skill