More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Kubernetes Cluster Ops with kubectl

    1

    Operate Kubernetes with kubectl safely: context checks before any mutation, a pod triage runbook that reads evidence in the right order, watched rollouts, and rollback that always exists.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Kubernetes Cluster Ops with kubectl

    Kubernetes Cluster Ops with kubectl

    Example session with this skill installed

    We deployed a new version of our checkout service to production. The rollout seems stuck — some new pods keep restarting and traffic still hits the old version. Walk me through what's wrong and how to fix it without making it worse.

    • Read your context and instructions
    • Compiled the kubernetes cluster ops

    TRIAGE — run in the skill's order: observe, diagnose from evidence, mutate once.

    Context first: kubectl config current-context confirms production, namespace shop stated before anything mutating. kubectl get pods -n shop -l app=checkout shows the picture: old ReplicaSet pods all Running, new ReplicaSet pods split — some Running, three in CrashLoopBackOff. The rollout isn't stuck; it's paused because the new version isn't healthy, and traffic still hitting the old version is the Service doing its job (endpoints only include ready pods).

    Phase step: the failing pods are CrashLoopBackOff, so the runbook goes to the death log before anything else: kubectl logs <pod> -n shop --previous. It shows the process dying at startup: fatal: failed to load config: open /etc/checkout/flags.yaml: no such file or directory. Current-attempt logs show the identical line — the container never starts serving.

    Root cause: the new image expects a config file the deployment doesn't mount. kubectl get pod <pod> -n shop -o yaml confirms the new pod spec has no volume for the flags ConfigMap — the manifest update added the image bump but dropped the volume mount that the previous revision had. Scheduling, image pull, registry auth: all clean per describe. The pod is a symptom of a manifest diff, nothing cluster-wide.

    Fix, one mutation: restore the ConfigMap volume and mount in the deployment manifest, kubectl apply --dry-run=server first, then apply for real. The crashed pods pick up the corrected spec as the controller replaces them. kubectl rollout status deployment/checkout -n shop --timeout=5m completes.

    Verify from outside: the checkout health endpoint returns the new version string and 200s; endpoints now list the new pods. The previous ReplicaSet is retained, so kubectl rollout undo is still available if the new version misbehaves under real load. Incident note records the root cause as a manifest regression, with the death-log line as evidence.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    kubectl is unforgiving in a specific way: the commands are short, the blast radius is a whole cluster, and the most common disaster is running the right command against the wrong context. This skill is the operating discipline that prevents it. It opens with the first rule — confirm context and namespace out loud before any mutation — and builds everything on top: an observing protocol that starts wide (nodes, all pods, sorted events) and narrows to the workload, with the jsonpath patterns that find sick pods in a 400-pod namespace without scrolling. The pod triage runbook walks a decision tree keyed on phase: Pending goes to describe and Events, CrashLoopBackOff goes to the previous container's logs first (the death log is usually the answer), Running-but-wrong earns an exec only after the cheap steps are exhausted. Deploys are treated as declared state plus watched convergence: server-side dry-run, apply, rollout status to a verdict, and verification from outside the cluster's own bookkeeping, because a pod can be Running and serving errors. Scaling, drains, and PodDisruptionBudgets are covered with the rule that a hanging drain is the system protecting availability, not an obstacle to force through. A worked example resolves a half-crashed rollout to one ConfigMap fix with almost no mutations. Leverage is built in: a symptom-to-step decision table, an incident note template, and a safety checklist. Edge cases cover missing metrics-server, Service selector drift, expiring events, and sidecar aiming, and the quality gate requires a written root cause and a live rollback path before any incident is called done.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for DevOps & Cloud.

    • It operates existing clusters; it doesn't design cluster architecture, provision infrastructure, or write a GitOps pipeline (it tells you when to graduate to one).
    • Commands assume a broadly standard Kubernetes distribution; managed offerings occasionally name things differently, and cluster-scoped steps need cluster-scoped permissions.
    • Live kubectl top data requires metrics-server; the skill's fallbacks work without it but are coarser.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 24 days ago

    Permissions

    Terminal / Shell
    Listed24 days ago
    Updated21 days ago

    What's inside

    Frequently Asked Questions

    Because the controller recreates them identically — same image, same missing mount — and you've destroyed the previous container's logs, which the runbook identifies as the single most useful piece of evidence. The skill's order exists so the diagnosis survives: describe, death log, then decide.

    Yes, for the half GitOps doesn't cover: live diagnosis. When Argo reports a degraded sync, the triage runbook is how you find which pod, which event, which log line explains it — and the context and rollback discipline applies to any manual intervention you make alongside the pipeline.

    The skill's first rule, mechanized: state context and namespace before mutations, keep the context visible in your prompt, use a separate KUBECONFIG per environment for anything destructive, and dry-run server-side before applies. Most wrong-cluster incidents are process failures, and the checklist is the process.

    New in DevOps & Cloud