kubeagent¶
Read-only Kubernetes troubleshooting, explained.
kubeagent scans a cluster, finds unhealthy pods, and explains why they're
failing — talking to the cluster through the official Kubernetes Go client
(client-go), strictly read-only.
What it catches¶
- CrashLoopBackOff — containers stuck restarting
- ImagePullBackOff — bad image or registry auth
- OOMKilled — hit the memory limit (shown with the container's requests/limits)
- Pending / Unschedulable — no node can place the pod
Beyond pods¶
- Service health — Services with no ready endpoints and LoadBalancers with no address, backing-aware
- NetworkPolicy hints — which policies select a stuck pod
- Connectivity diagnostics — actionable "API server unreachable" messages
- Credential lint — opt-in scan for secrets stored in the clear
- Resource context — cluster CPU/memory plus per-OOMKill limits
- Platform facts — CNI, ingress, storage, distro, runtime, cloud
See it¶
$ kubeagent scan
Cluster: Healthy — 3/3 nodes Ready
Platform: Cilium CNI · Traefik ingress · Kubernetes v1.30 · containerd
Resources (cluster):
CPU 24.0 cores · req 6.2 (25%) · lim 18.0 (75%) · used 3.1 (12%)
Memory 96Gi · req 22Gi (22%) · lim 70Gi (72%) · used 18Gi (18%)
⚠ shop/checkout Deployment 1/3 Degraded · 12 restarts, last 2m ago
⚠ CrashLoopBackOff: Container repeatedly crashes after starting
Service issues:
⚠ shop/checkout ClusterIP no ready endpoints
Optional --explain makes a single Claude API call to summarize findings in
plain English — the deterministic core still works fully offline.