Kubernetes Specialist guides Claude through cluster work in five steps: analyze the workload, design the architecture, write declarative manifests, secure them with RBAC, NetworkPolicies and Pod Security settings, and validate with kubectl. The skill ends with a set of kubectl commands for rollout status, pod events, logs, resource use and rollback.
Eleven reference files cover workloads, networking, configuration, storage, Helm charts, troubleshooting, custom operators, service mesh, GitOps, cost optimization and multi-cluster management.
What You Get
- Deployment, StatefulSet, DaemonSet, Job and CronJob manifests with limits, probes and security contexts
- RBAC roles and bindings plus default-deny NetworkPolicies with explicit allow rules
- ConfigMap, Secret and storage definitions, including external and sealed secrets
- Helm charts with values, helpers, hooks and tests
- Troubleshooting steps with kubectl logs, describe, events and debug commands
- Notes on operators, Istio and Linkerd, ArgoCD and Flux, cost tuning and multi-cluster setups
Common Use Cases
- Deploying an app with limits, probes and a locked-down security context
- Writing least-privilege RBAC and network isolation for a namespace
- Packaging an application as a Helm chart
- Debugging a crash loop or a pod stuck pending
- Right-sizing workloads with VPA and HPA settings
- Setting up GitOps delivery with ArgoCD or Flux
What Makes It Unique
It covers the full Kubernetes surface in one skill, from a single manifest to custom operators, service mesh, GitOps and multi-cluster disaster recovery. A MUST NOT list bans the default service account, the latest tag, root containers, secrets in ConfigMaps and open network access, and a validation block gives the commands to confirm a rollout and to audit a service account.
Under The Hood
The skill is a SKILL.md with eleven reference files and no scripts, and the references load by topic. The workflow tells the agent to run kubectl against a live cluster: rollout status, get pods, describe, logs, top, auth can-i and rollout undo, which can change the cluster. The references go further: they run commands inside pods and copy files in and out, apply the ArgoCD manifest from a raw GitHub URL, add Helm repositories and install charts, and download and run installer scripts for Istio, Linkerd, clusterctl and Submariner. The agent also writes manifests and Helm chart files. Examples reference secrets by name and use placeholder values. Treat any cluster it is pointed at as one it can modify.