How to use the GKE Production Readiness Review agent skill

What does the gke-productionize skill make the agent do?

The skill acts as an orchestrator. The agent first inspects the current setup using gcloud and kubectl commands, then works through a checklist covering scalability, observability, reliability, security, backup and disaster recovery, ingress, cost and upgrades.

For each area it hands off to a more specialised skill, such as gke-workload-scaling or gke-reliability. It then produces a red, amber or green (RAG) status per area and an overall readiness score to help you prioritise fixes. According to the skill's own assessment, it does not implement most of the changes itself.

How do you install the skill?

Download the ZIP file for the skill from this page. Then choose one of two routes.

For Claude Code, unzip the download into ~/.claude/skills/ to make it available for all your projects, or into a project's .claude/skills/ folder to limit it to that project.

In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

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

What do you need before running a review?

The skill expects the Google Cloud CLI (gcloud) with access to your Google Cloud project, and kubectl with access to the cluster. Your Google Cloud account needs permission to describe GKE clusters.

The skill refers to companion skills by name, including gke-app-onboarding, gke-workload-scaling, gke-observability, gke-reliability, gke-platform-security, gke-workload-security, gke-backup-dr, gke-service-networking, gke-cost-optimization, gke-upgrades and gke-golden-path. These are referenced but are not included in the files provided with this skill, so make sure they are available to your agent or expect the agent to be unable to run those checks in full.

• gcloud CLI with access to the Google Cloud project
• kubectl with access to the cluster
• A Google Cloud account allowed to describe GKE clusters
• Companion skills listed above, which are not bundled here

How do you trigger the skill with a natural request?

You do not need a special command. Ask in plain language using words like productionize, prepare, assess, audit or review, together with a GKE cluster or workload and the fact that it is going live. The skill's description is written to match requests of this kind.

Name the cluster, location and project when you can, because the discovery commands use them. If you only care about one app, say so, and the agent will focus on that workload.

• Review my GKE cluster prod-cluster in europe-west1 before we go live.
• Audit the payments deployment in the payments namespace for production readiness.
• Prepare our staging workload on GKE for production and tell me what to fix first.
• Assess this cluster for production: check scaling, security and backups.

What does the agent check and how is the score calculated?

Discovery starts with cluster details, including whether the cluster is Autopilot or Standard and its release channel. For a specific workload it checks the deployment, namespace Pod Security Standards labels, service accounts, resource requests and limits, liveness, readiness and startup probes, HPA, PDB and NetworkPolicies.

Each domain is classified per check as pass, fail-critical (production-blocking) or fail-minor (an improvement). A domain is Red if it has any fail-critical check, Amber if it has only fail-minor checks, and Green if all checks pass. Domains score 100 for Green, 50 for Amber and 0 for Red.

Security, Reliability and Backup/DR are weighted 2x, and the other domains 1x. The overall score is the weighted average, rounded to the nearest integer. Not-applicable domains, such as Backup/DR for fully stateless workloads, are excluded and the report notes this. A score of 90 or more with no Red domains is 'Production ready'; 70 to 89 with no Red domains is 'Ready with follow-ups'; anything else is 'Not production ready'.

• Red: one or more fail-critical checks
• Amber: no fail-critical checks, but one or more fail-minor checks
• Green: all checks pass

What safety points should you know before running it?

The skill does not run scripts. It does use the network, and it may run read commands against live infrastructure using your own credentials. Treat that as a real production action: review what the agent is about to run, and do not point it at a production cluster without that review.

Every gcloud command the skill runs directly in the terminal is prefixed with a usage attribution variable that identifies the skill. This tag carries no user data. The skill also says it seeks your confirmation before applying any state-changing implementation.

• No scripts are run by the skill itself
• Network access is used
• Read commands can hit live clusters under your credentials
• Changes are proposed and confirmed before they are applied