kubestellar-console – skill Agent & Tự động hóa cho agent AI

KubeStellar Console

Overview

KubeStellar Console is an open-source multi-cluster Kubernetes dashboard (CNCF project) with AI-powered operations. It ships with kc-agent, an MCP server that bridges coding agents to kubeconfig and Kubernetes APIs, plus 10+ built-in agent skills for development, testing, and operations.

When to Use This Skill

• Use when managing multiple Kubernetes clusters across edge and cloud
• Use when you need AI-assisted Kubernetes troubleshooting and debugging
• Use when running performance tests, cache compliance checks, or CI debugging on a Kubernetes dashboard
• Use when integrating with CNCF projects (Argo, Kyverno, Istio, and 20+ others)

How It Works

Step 1: Install kc-agent

brew tap kubestellar/tap && brew install kc-agent

Step 2: Start the MCP server

kc-agent

This bridges the active kubeconfig context to any MCP-compatible coding agent. Do not start it from a cluster-admin or write-capable context unless the user explicitly accepts that risk.

Step 3: Use built-in agent skills

The project ships with agent skills accessible via CLAUDE.md and AGENTS.md:

• @perf-test — Dashboard performance testing and TTFI analysis
• @cache-test — Card cache compliance testing (IndexedDB warm return)
• @nav-test — Navigation performance testing
• @ui-compliance-test — Card loading compliance (8 criteria, 150+ cards)
• @ci-status — CI pipeline monitoring and status checks
• @rca — Root cause analysis for CI/test failures
• @tdd — Test-driven development workflow
• @k8s-debug — Kubernetes debugging and troubleshooting

Key Features

• Multi-cluster management across edge and cloud
• Real-time streaming observability
• 20+ CNCF project integrations (Argo, Kyverno, Istio, etc.)
• GitHub OAuth authentication
• Supply chain security (SBOM, SLSA)
• SQLite WASM caching with stale-while-revalidate pattern
• 15+ themes with dark/light mode

Security & Safety Notes

• Critical risk: kc-agent bridges your active kubeconfig context to MCP-compatible agents. If that context carries cluster-admin, write permissions, or secret read access, agents inherit those capabilities.
• Do not rely on RBAC objects alone: creating a ServiceAccount or ClusterRoleBinding does not change the credentials kc-agent uses. Start kc-agent only after switching KUBECONFIG/context to dedicated least-privilege credentials and verifying them.
• Recommended read-only scope: avoid resources='*', because it includes sensitive objects such as Secrets. Prefer an explicit non-secret resource list and verify access before starting the MCP server:

  kubectl create serviceaccount kc-agent -n default
  kubectl create clusterrole kc-agent-readonly \
    --verb=get,list,watch \
    --resource=pods,services,deployments.apps,replicasets.apps,statefulsets.apps,daemonsets.apps,namespaces,nodes,events,configmaps
  kubectl create clusterrolebinding kc-agent-readonly \
    --clusterrole=kc-agent-readonly \
    --serviceaccount=default:kc-agent
  kubectl auth can-i get secrets --as=system:serviceaccount:default:kc-agent
  kubectl auth can-i list pods --as=system:serviceaccount:default:kc-agent
  

• The first can-i command must return no; the second should return yes. Then create or select a kubeconfig that actually authenticates as that ServiceAccount before running kc-agent.
• Do not expose kc-agent on a public network without authentication.
• Review [SECURITY-AI.md](https://github.com/kubestellar/console/blob/main/docs/security/SECURITY-AI.md) for prompt injection and agent drift mitigations.

Limitations

• This skill requires an external binary (kc-agent) installed separately via Homebrew.
• Do not treat agent output as a substitute for environment-specific validation or expert review.
• Stop and ask for clarification if required permissions or safety boundaries are unclear.

Links

• [GitHub](https://github.com/kubestellar/console)
• [Website](https://console.kubestellar.io)
• [CLAUDE.md](https://github.com/kubestellar/console/blob/main/CLAUDE.md)
• [AGENTS.md](https://github.com/kubestellar/console/blob/main/AGENTS.md)