What problem does it solve? Running Kubernetes clusters in production means constant Day-2 work: Pods stuck in CrashLoopBackOff, rollouts that hang, nodes under memory pressure, RBAC denials, and PVCs that never bind. This Skill provides a structured SRE playbook that starts from Events and kubectl describe before any action, so incidents are diagnosed at the root cause instead of being masked by blind restarts. ## Core Features & Use Cases - Incident triage decision trees: Maps symptoms like CrashLoopBackOff, OOMKilled (exit 137), Evicted, Pending/FailedScheduling, service-not-reachable, and PVC-Pending to concrete diagnose-and-fix steps using kubectl get|describe|logs --previous|events|top|debug. - Workload and lifecycle operations: Covers Deployment rollout/rollback strategy (maxSurge/maxUnavailable), StatefulSet and DaemonSet ops, probe tuning, graceful shutdown, and version-skew-aware cluster upgrades with kubeadm. - Capacity, scheduling, and autoscaling: Explains requests vs limits, QoS classes and eviction order, LimitRange/ResourceQuota, affinity/taints/topologySpreadConstraints, HPA (autoscaling/v2), VPA, Cluster Autoscaler, Karpenter, and KEDA. - Security, networking, storage, and DR: Guides RBAC verification with auth can-i, Pod Security Admission, NetworkPolicy default-deny, EndpointSlices/CoreDNS debugging, PV/PVC lifecycle and reclaim policies, plus etcd snapshots, CSI VolumeSnapshots, and Velero backup/restore. - Use Case: A Deployment rollout is stuck and Pods show OOMKilled. The Skill walks you through kubectl describe Events, logs --previous, and kubectl top to confirm memory-limit pressure, then right-sizes requests/limits and verifies the rollout completes. ## Quick Start Ask the AI to triage why a specific Pod is in CrashLoopBackOff and recommend the fix using the Kubernetes operations playbook.
