Failure diagnostics¶
kubeagent scans your cluster, finds unhealthy pods, and explains why they
are failing — covering the most common pod failure modes.
Read-only operation¶
kubeagent talks to the cluster directly via the official Kubernetes Go client
(client-go) — the same library kubectl and operators use — and operates
read-only. It never creates, updates, patches, or deletes cluster resources.
Failure modes detected¶
CrashLoopBackOff¶
The container keeps restarting. Kubernetes backs off exponentially between
attempts. kubeagent surfaces the exit code and last termination reason so you
can spot crash loops without tailing logs manually.
ImagePullBackOff / ErrImagePull¶
The image cannot be pulled — either the image tag does not exist or the node
lacks credentials for the registry. kubeagent reports the image reference and
the pull error from the pod's conditions.
OOMKilled¶
The container exceeded its memory limit and was killed by the kernel OOM killer.
kubeagent annotates the finding with the container's configured requests and
limits (see Resource context) so you can judge whether to
raise the limit or reduce memory pressure.
Pending / Unschedulable¶
No node can place the pod. This covers insufficient CPU or memory, a missing
taint toleration, an unsatisfied node affinity, or no nodes at all.
kubeagent reports the scheduler message from the pod's events.
Status¶
kubeagent scan performs a read-only, whole-cluster scan and reports
CrashLoopBackOff, ImagePullBackOff/ErrImagePull, OOMKilled, and
Pending/Unschedulable pods, in text or JSON.
The optional --explain flag makes a single Claude API call to summarize
findings in plain English. The deterministic core still works offline with no
API key.