guiyi-labs/kubemd

README excerpt

KubeMD — The Kubernetes Surge Doctor Evidence-first runtime diagnosis for Kubernetes failures — with case memory. A DSH (DeepSeek Harness) skill that diagnoses live, broken clusters , not manifests. When a pod is CrashLooping, a node goes NotReady, or a Service stops answering, KubeMD runs a disciplined loop: capture context → build a red-capable feedback loop → collect signals → rank falsifiable hypotheses → fix with dry-run semantics → record the case for instant recall next time . Different from KubeShark-style skills: they prevent hallucinations while writing YAML . KubeMD finds out why your running workload is broken — and never forgets a fix. Install (30 seconds) That's it. DSH auto-discovers skills in /.dsh/skills/ . No restart needed. DSH (DeepSeek Harness) — everything is a plugin. Skills are instruction bundles + scripts that agents load on demand. 💡 Install as a skill directory: the repo layout we ship is exactly a DSH skill bundle. Or copy the folder and rename to dsh-k8s-diagnosis under /.dsh/skills/ . Demo CLI running against a real fault-injected kind cluster (diagnose → 4 findings → case recall): What it does Included Path Purpose SKILL.md The 7-phase procedure (sh…

Category