kubernetes-troubleshooting

Extract the bundle into scratch space, inventory logs, events, and kubectl outputs, then run broad keyword searches for errors before forming hypotheses. This Skill guides that workflow and produces a root-cause report with file and line citations.

Check previous container logs, exit codes, termination reasons, restart cadence, probes, config and env references, and dependency readiness. Correlate these with Service, Endpoints, and DNS state before concluding the cause.

Yes, the Skill explicitly handles OpenShift must-gather archives along with tarballs, kubectl outputs, Helm and Kustomize files, node data, and CNI/DNS/ingress/storage artifacts. It reads must-gather indexes and collection metadata before raw logs.

No, it prefers offline analysis of evidence files and read-only tools like rg, tar, jq, and yq. kubectl is used only against offline files unless live-cluster access is explicitly requested and available.

Labeling claims as Fact, Inference, or Unknown prevents premature conclusions from noisy logs. Every important claim must cite a file path, line number, or command section so the root cause is defensible.

The Skill names the missing discriminator, such as cluster-wide events, pod descriptions, previous container logs, Service and Endpoint YAML, node conditions, kubelet logs, CNI logs, or manifests from before and after the change.