KubeMD v0.1.0 — Evidence-first Kubernetes diagnosis with case memory

KubeMD v0.1.0

Evidence-first runtime diagnosis for Kubernetes failures — with case memory. A DSH (DeepSeek Harness) skill that diagnoses live, broken clusters, not manifests.

What ships

• SKILL.md — 7-phase diagnosis procedure (feedback loop → signals → falsifiable hypotheses → dry-run fix → case record)
• 5 verified playbooks: CrashLoopBackOff / OOMKilled / Network / Pending / NodeNotReady
• signal-map.md symptom→signal→command cheatsheet
• Scripts: collect-signals.sh (one-shot signal collection), record-case.sh (case library append)
• cases.yaml example library + full verification report (real fault-injected kind cluster, 5 scenarios, no hard errors)

Install (30s)

git clone https://github.com/guiyi-labs/kubemd ~/.dsh/skills/dsh-k8s-diagnosis

CLI twin

The same deterministic engine ships as a go install CLI: aiops diagnose / aiops cases (https://github.com/guiyi-labs/aiops-platform, v0.1.0 binaries available).

Verification report