What problem does it solve? Debugging Kubernetes workloads normally requires direct cluster access and deep kubectl knowledge. This Skill provides a structured debugging methodology and ready-made scripts that route all queries through a k8s-gateway, so you can investigate pod crashes, CrashLoopBackOff, OOMKilled, and deployment issues on remote clusters without direct API access. ## Core Features & Use Cases - Gateway-First Cluster Discovery: Lists connected clusters via the k8s-gateway and routes every query with a --cluster-id, eliminating the need for local kubeconfig or direct API access. - Events-Before-Logs Methodology: Enforces a proven diagnostic order (events first, then logs) that resolves most issues faster, with a lookup table mapping event reasons like OOMKilled and ImagePullBackOff to concrete remediation steps. - Complete Diagnostic Toolkit: Scripts for listing pods and namespaces, fetching events and logs, describing pods, deployments, and nodes, and comparing resource usage against limits via metrics-server. - Use Case: A payment pod in production is stuck in CrashLoopBackOff. You list clusters, pull the pod's events to find an OOMKilled reason, check resource limits versus actual usage, and get a structured root-cause report with a recommended memory limit increase. ## Quick Start Ask the agent to debug why a specific pod is crashing in a given namespace, and it will discover the cluster, check events, and report the root cause.
