Manage KubeSphere Clusters with 32 Agent Skills

KubeSphere is the open-source container platform for Kubernetes multi-cloud, datacenter, and edge management — and its 17k-star repo carries a payload the main README never mentions: a skills/ directory with 32 SKILL.md skills (as of October 16, 2026) that turn a coding agent into a KubeSphere operator. You get read-only cluster queries, DevOps pipeline operations, and WizTelemetry observability workflows, all driven from chat.

Why This Skill Matters

The README documents none of this — the skills live only as folders under skills/, each with a SKILL.md whose frontmatter states when to use it, and a few add an explicit out-of-scope section. That structure makes the collection unusually honest: the kubesphere-cluster-management skill scopes itself to read-only queries and warns never to create, update, or delete through kubectl — the console stays the place for management operations.

The thirty-two skills group into a few clear clusters: a DevOps set of six (overview, Jenkins, Argo CD, pipelines, tenant-scoped operation, credentials) for CI/CD work; a WizTelemetry observability set — logging, events, auditing, notification, tracing, alerting rules, a Vector-based data pipeline, and the whizard-telemetry platform service whose APIServer backs the set; platform operations like kubeeye (cluster inspection with OPA, PromQL, and other rule types), openpitrix (the Helm/YAML app store), volcano (batch training jobs and queues), openkruise (CloneSet and SidecarSet rollouts), and servicemesh (Istio, Kiali, Jaeger); plus a gateway pair that cross-references itself — kubesphere-gateway for the ingress-nginx path and kubesphere-gateway-api for the newer Kubernetes Gateway API on Traefik. The directory also holds nine skills this article does not unfold — kubesphere-core, kubesphere-extension-management, kubesphere-fluid, kubesphere-network-extension-operations, nodegroup, opensearch, and three frontend-forge skills; treat the directory itself as the full list.

Installation

The repo documents no install command, so installation is the standard SKILL.md convention: clone and copy the skills you want into your agent's skills directory. To take the cluster-query skill, for example:

git clone https://github.com/kubesphere/kubesphere
cp -r kubesphere/skills/kubesphere-cluster-management ~/.claude/skills/

Pick per role rather than copying all 32 — a platform engineer wants the DevOps set, an SRE the WizTelemetry set. You will need kubectl pointed at a KubeSphere-managed cluster for the skills that drive it.

Real Workflow: A Read-Only Fleet Check

1. Install the cluster-management skill as above and confirm your kubectl context.
2. Ask for the report:

用 kubesphere-cluster-management 列出所有托管集群的名称、连接类型、
KubeSphere 版本、Kubernetes 版本和 Ready 状态，整理成一张汇总表。
只做查询，不要执行任何写操作。

3. The skill runs kubectl get cluster with a formatted template — the SKILL.md's own commands print name, connection type, KS version, K8s version, and the Ready/KSCoreReady conditions.
4. Follow up in the same session: the multi-tenant-management skill can create users, workspaces, projects, and role bindings — while refusing delete operations and custom roles, per its own scope.

For scheduled hygiene, point the same setup at kubeeye: its SKILL.md covers deploying KubeEye, creating InspectRule and InspectPlan resources, and retrieving inspection reports.

Tips

• Read the whole SKILL.md before relying on a skill — several include an explicit out-of-scope section, and the collection's boundaries are written down rather than left to guesswork.
• The two gateway skills reference each other for a reason: pick kubesphere-gateway-api for the newer Kubernetes Gateway API standard, kubesphere-gateway for ingress-nginx.
• Directory listings drift — count what you copy against the skills/ folder rather than this article's number.
• The DevOps skills document KubeSphere API operations as well as kubectl paths, so tenant-scoped users without cluster-admin have a documented route.

When Not to Use This

The skills deliberately route write-heavy management back to the KubeSphere console — if your workflow is mostly creating and deleting resources, the console UI is the intended tool. They are also KubeSphere-specific: a plain Kubernetes cluster without KubeSphere will not answer these queries. And with no documented install command or marketplace entry, this collection asks more setup of you than a one-command plugin — worth it for platform teams, less so for a quick taste.

See the leaderboard for more skills.