kubernetes

Cluster baseline scaffolding, RBAC diagnosis and generation, workload hardening, and structured pod/scheduling debug for plain Kubernetes across all distributions.

SKILL.md

GitHub

Kubernetes Command

Structured guidance for cluster baseline standards, RBAC, workload patterns, and operational debugging on Kubernetes.

Activation

/platform-skills:kubernetes baseline   # generate namespace, RBAC, network policy, PDB, quota scaffold
/platform-skills:kubernetes rbac       # diagnose 401/403; generate Role/RoleBinding; simulate access
/platform-skills:kubernetes workload   # Deployment, HPA, probes, securityContext hardening
/platform-skills:kubernetes debug      # pod crashloop, OOMKill, pending scheduling, image pull errors

Interactive Wizard (fires when no mode is provided)

When invoked with no arguments, ask before proceeding:

Q1 — Mode?

What do you need?
  1. baseline  — namespace, RBAC, network policy, PDB, resource quota scaffold
  2. rbac      — diagnose 401/403, generate Role/RoleBinding, simulate access
  3. workload  — Deployment, HPA, probes, securityContext, PDB hardening
  4. debug     — crashloop, OOMKill, pending, image pull, general troubleshoot

Enter 1–4 or mode name:

Q2 — Context (after mode selected):

  • baseline: Which namespace? What team owns it? Any special requirements (privileged workloads, GPU, ingress)?
  • rbac: Paste the 403 error message or describe which service account needs which access.
  • workload: Paste the Deployment manifest or describe what the workload does.
  • debug: Paste the pod describe output and recent logs, or describe the symptom.

Mode: baseline

Triggers: baseline, scaffold, new namespace, set up namespace, platform baseline

Read references/kubernetes.md before responding.

Generate the minimum platform baseline for a new namespace. Ask for namespace name and team before generating.

Namespace with ownership labels

apiVersion: v1
kind: Namespace
metadata:
  name: app-team
  labels:
    team: app-team
    environment: production
    managed-by: platform
  annotations:
    contact: platform@org.com

Resource quota

apiVersion: v1
kind: ResourceQuota
metadata:
  name: app-team-quota
  namespace: app-team
spec:
  hard:
    requests.cpu: "4"
    requests.memory: 8Gi
    limits.cpu: "8"
    limits.memory: 16Gi
    pods: "20"
    persistentvolumeclaims: "10"

LimitRange — container defaults

apiVersion: v1
kind: LimitRange
metadata:
  name: container-defaults
  namespace: app-team
spec:
  limits:
    - type: Container
      default:
        cpu: 500m
        memory: 512Mi
      defaultRequest:
        cpu: 100m
        memory: 128Mi

Default-deny network policy

apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: default-deny-all
  namespace: app-team
spec:
  podSelector: {}
  policyTypes:
    - Ingress
    - Egress
---
# Allow DNS egress — required for pod name resolution
apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: allow-dns-egress
  namespace: app-team
spec:
  podSelector: {}
  policyTypes:
    - Egress
  egress:
    - ports:
        - port: 53
          protocol: UDP
        - port: 53
          protocol: TCP

PodDisruptionBudget

apiVersion: policy/v1
kind: PodDisruptionBudget
metadata:
  name: app-pdb
  namespace: app-team
spec:
  minAvailable: 1          # keep at least 1 pod running during voluntary disruptions
  selector:
    matchLabels:
      app: app-team

Validation

kubectl apply --dry-run=server -f namespace-baseline/

# Confirm quota is applied
kubectl describe resourcequota app-team-quota -n app-team

# Confirm network policy is in place
kubectl get networkpolicy -n app-team

# Test DNS still works from a pod
kubectl run dns-test --image=busybox --restart=Never --rm -it -n app-team \
  -- nslookup kubernetes.default

Handoffs:

  • Policy enforcement on these namespaces → /platform-skills:kyverno or /platform-skills:opa
  • Secrets strategy for the namespace → /platform-skills:secrets
  • Measure the result against NSA or the provider's CIS benchmark → /platform-skills:kubescape cluster (kubescape scan framework nsa --include-namespaces app-team)
  • Allow rules for the default-deny policy, drafted from the traffic the workloads actually send → /platform-skills:kubescape-operator netpol

Mode: rbac

Triggers: 401, 403, forbidden, unauthorized, role, binding, service account, RBAC, can-i

Read references/kubernetes.md → RBAC troubleshooting section before responding.

Step 1 — Diagnose 401 vs 403

Code Layer First check
401 Unauthorized Authentication failed kubectl auth whoami — confirm identity
403 Forbidden Authorized identity, missing permission kubectl auth can-i — confirm what is allowed

Step 2 — Simulate the failing request

# Test exactly what the service account can do
kubectl auth can-i <verb> <resource> \
  --as=system:serviceaccount:<namespace>:<sa-name> \
  -n <namespace>

# Examples
kubectl auth can-i get pods \
  --as=system:serviceaccount:app-team:app-controller \
  -n app-team

kubectl auth can-i list secrets \
  --as=system:serviceaccount:app-team:app-controller \
  -n app-team

Step 3 — Find existing bindings for the service account

kubectl get rolebindings,clusterrolebindings -A -o json \
  | jq -r '
    .items[]
    | select(
        .subjects[]?
        | select(.kind=="ServiceAccount"
            and .name=="<sa-name>"
            and .namespace=="<namespace>")
      )
    | "\(.kind)/\(.metadata.namespace)/\(.metadata.name) -> \(.roleRef.name)"'

Step 4 — Generate minimum Role and RoleBinding

apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
  name: pod-reader
  namespace: app-team
rules:
  - apiGroups: [""]
    resources: ["pods"]
    verbs: ["get", "list", "watch"]
  - apiGroups: ["apps"]
    resources: ["deployments"]
    verbs: ["get", "list", "watch"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
  name: pod-reader-binding
  namespace: app-team
subjects:
  - kind: ServiceAccount
    name: app-controller
    namespace: app-team
roleRef:
  kind: Role
  name: pod-reader
  apiGroup: rbac.authorization.k8s.io

Scope selection

Situation Use
Workload needs access within one namespace RoleBinding → Role
Reuse a cluster role, scoped to one namespace RoleBinding → ClusterRole
Controller needs cross-namespace or node access ClusterRoleBinding → ClusterRole

Prefer the narrowest scope. Audit ClusterRoleBinding usage regularly:

kubectl get clusterrolebindings -o json \
  | jq -r '.items[] | select(.subjects[]?.kind=="ServiceAccount") | "\(.metadata.name) -> \(.roleRef.name)"'

Validation

kubectl auth can-i get pods \
  --as=system:serviceaccount:app-team:app-controller \
  -n app-team
# expected: yes

kubectl auth can-i delete pods \
  --as=system:serviceaccount:app-team:app-controller \
  -n app-team
# expected: no

Rollback: Delete the RoleBinding. Roles and bindings are additive — removing them revokes the access without affecting running workloads beyond that access.


Mode: workload

Triggers: deployment, statefulset, HPA, probes, securityContext, harden, resources, requests, limits

Read references/kubernetes.md before responding.

Hardened Deployment template

apiVersion: apps/v1
kind: Deployment
metadata:
  name: app
  namespace: app-team
  labels:
    app: app
    version: "1.0.0"
spec:
  replicas: 2
  selector:
    matchLabels:
      app: app
  template:
    metadata:
      labels:
        app: app
    spec:
      serviceAccountName: app-sa    # dedicated SA, not default
      securityContext:
        runAsNonRoot: true
        runAsUser: 1000
        fsGroup: 2000
        seccompProfile:
          type: RuntimeDefault
      containers:
        - name: app
          image: ghcr.io/org/app:1.0.0
          ports:
            - containerPort: 8080
          resources:
            requests:
              cpu: 100m
              memory: 128Mi
            limits:
              memory: 256Mi           # no CPU limit — throttles unnecessarily
          securityContext:
            allowPrivilegeEscalation: false
            readOnlyRootFilesystem: true
            capabilities:
              drop: ["ALL"]
          livenessProbe:
            httpGet:
              path: /healthz
              port: 8080
            initialDelaySeconds: 10
            periodSeconds: 10
            failureThreshold: 3
          readinessProbe:
            httpGet:
              path: /ready
              port: 8080
            initialDelaySeconds: 5
            periodSeconds: 5
            failureThreshold: 2
          startupProbe:             # protects slow-starting apps from liveness kills
            httpGet:
              path: /healthz
              port: 8080
            failureThreshold: 30
            periodSeconds: 10

HPA — only when requests and metrics are defined

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: app-hpa
  namespace: app-team
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: app
  minReplicas: 2
  maxReplicas: 10
  metrics:
    - type: Resource
      resource:
        name: cpu
        target:
          type: Utilization
          averageUtilization: 70

For event-driven or custom metric scaling → /platform-skills:keda

Common hardening gaps

Gap Fix
No resources.requests Pod is unschedulable under resource pressure; add requests
resources.limits.cpu set Causes throttling even when cores are idle; remove CPU limit unless hard isolation is needed
runAsRoot: true or no runAsNonRoot Run as non-root; most app images support UID 1000+
No readinessProbe Pod receives traffic before it is ready; add readiness probe
No startupProbe for slow apps Liveness kills pods during slow start; add startup probe
default service account Creates implicit access to cluster APIs; create a dedicated SA

Find these gaps across a whole manifest tree or cluster at once: kubescape scan framework nsa <path> (or with no path for the current cluster) lists each as a failed control with the affected resources → /platform-skills:kubescape.


Mode: debug

Triggers: crashloop, OOMKilled, pending, ImagePullBackOff, ErrImagePull, error, not starting, debug

Apply the structured checklist:

Step 1 — Collect evidence

# Pod state and recent events
kubectl get pod <pod-name> -n <namespace>
kubectl describe pod <pod-name> -n <namespace>

# Container logs — current and previous (if restarted)
kubectl logs <pod-name> -n <namespace>
kubectl logs <pod-name> -n <namespace> --previous

# Recent events in the namespace
kubectl get events -n <namespace> --sort-by='.lastTimestamp' | tail -20

Step 2 — Classify by symptom

CrashLoopBackOff

# Application is starting and exiting — check exit code and logs
kubectl describe pod <pod-name> -n <namespace> | grep -A5 "Last State"
kubectl logs <pod-name> -n <namespace> --previous

Common causes: application error on startup, missing env var or secret, wrong entrypoint, liveness probe killing a slow-starting pod (add startupProbe).

OOMKilled

# Check memory usage vs limit
kubectl top pod <pod-name> -n <namespace> --containers
kubectl get pod <pod-name> -n <namespace> \
  -o jsonpath='{.spec.containers[*].resources}'

# Check OOM events
kubectl describe pod <pod-name> -n <namespace> | grep -i oom

Fix: increase resources.limits.memory or profile the application for a memory leak. Never remove the limit — set it correctly.

Pending

# Events show why scheduling failed
kubectl describe pod <pod-name> -n <namespace> | grep -A10 "Events:"
Pending reason Cause Fix
Insufficient cpu/memory No node has enough capacity Add nodes, reduce requests, or check Karpenter/Cluster Autoscaler
didn't match node selector/affinity Node labels don't match Fix the nodeSelector or affinity block
Unschedulable: taint Node has a taint the pod doesn't tolerate Add the matching toleration to the pod spec
PVC not bound PersistentVolumeClaim is pending kubectl describe pvc <name> -n <ns> — check StorageClass and provisioner

ImagePullBackOff / ErrImagePull

kubectl describe pod <pod-name> -n <namespace> | grep -A5 "Failed"

Common causes: image tag does not exist, registry is private and imagePullSecrets is missing, incorrect registry hostname.

# Verify the image exists
docker manifest inspect ghcr.io/org/app:1.0.0

# Check imagePullSecrets are mounted
kubectl get pod <pod-name> -n <namespace> \
  -o jsonpath='{.spec.imagePullSecrets}'

Validation

After applying a fix, confirm the pod reaches Running and passes readiness:

kubectl rollout status deployment/<name> -n <namespace>
kubectl get pod -n <namespace> -l app=<label> -w

Rollback

# Roll back the Deployment to the previous revision
kubectl rollout undo deployment/<name> -n <namespace>

# Check rollout history
kubectl rollout history deployment/<name> -n <namespace>

Common mistakes

  • Setting resources.limits.cpu — CPU limits throttle containers even when other cores are idle. Set requests but omit the CPU limit unless hard isolation is specifically required.
  • Using the default service account — it may have accumulated bindings over time and creates implicit API access. Always create a dedicated service account per workload.
  • No startupProbe for slow-starting apps — liveness probes kill pods during startup if initialDelaySeconds is too short. Use a startup probe to gate liveness until the app is running.
  • ClusterRoleBinding when RoleBinding was sufficient — grants access cluster-wide. Audit all ClusterRoleBindings with service account subjects.
  • kubectl apply -f in production without --dry-run=server first — server-side dry-run catches admission errors, invalid API versions, and webhook rejections before they apply.

Reference

Full guidance: references/kubernetes.md

For policy enforcement: /platform-skills:kyverno or /platform-skills:opa

For event-driven autoscaling: /platform-skills:keda

For node provisioning: /platform-skills:karpenter

Examples:

  • examples/kubernetes/deployment-baseline.yaml
  • examples/kubernetes/namespace-baseline.yaml
  • examples/kubernetes/network-policy-default-deny.yaml
  • examples/kubernetes/pod-disruption-budget.yaml