Roadmap¶
Shipped¶
- v1 —
kubeagent scan: deterministic whole-cluster scan and diagnosis of CrashLoopBackOff, ImagePullBackOff/ErrImagePull, OOMKilled, and Pending/Unschedulable pods - v2 — optional
--explainflag: one Claude API call summarizes findings in plain English; the deterministic core still works offline with no API key - Resource context — compact CPU/memory summary (allocatable, reserved,
limits, live usage); OOMKilled findings annotated with the container's
requests/limits; sent to
--explain— see Resource context - Platform facts — CNI, ingress, storage provisioner, Kubernetes version +
distribution, container runtime, and cloud detected read-only and shown under
the cluster verdict; sent to
--explain— see Platform facts - Service health — flags selector-based Services with zero ready endpoints and LoadBalancer Services with no external address; backing-workload annotations distinguish expected-empty from broken — see Service health
- NetworkPolicy hints — when a workload is degraded with no detector finding, names the NetworkPolicies whose podSelector matches its pods — see NetworkPolicy hints
- Connectivity diagnostics — when the API server is unreachable, prints an actionable diagnosis (down control plane, timeout, TLS/cert error, 401/403, DNS) instead of a raw transport error — see Connectivity diagnostics
- Credential lint —
scan --lint-secretsflags credentials stored in the clear in ConfigMaps and pod env literals; reports location and pattern only, never the value, and never sends findings to--explain— see Credential lint
Version history
GitHub Releases and the CHANGELOG are the source of truth for what shipped in each version.