Ask Kubernetes

Attach a cluster and ask in plain words. The model says what the question needs and what it is about; Verbiqo finds that among the cluster’s names, reads only what is needed, works out its findings, and the model reads further where they point.

Attach

/k8s prod

Verbiqo shows the platform, version and node count, and whether it reads the whole cluster or some namespaces.

What to ask

You want to know	Ask, for example	What Verbiqo does first
Why something fails	Why is checkout crashing?	Finds crash loops, OOM kills, image pulls, missing Secrets or ConfigMaps, pending pods with the scheduler’s reason, pods not ready, and reads the crashed run’s log when logs are on.
A rollout	Did the last deploy break the API?	Compares the newest revision with the one before, and gives the rollback command when the new one is the cause.
What changed	What changed in the cluster at 3am?	A timeline of rollouts with their images, restarts, OOM kills, node changes, failed jobs and warnings, beside your databases’ and logs’ events.
Room and cost	Do we have room for another replica? What is over-provisioned?	Each node’s requests and use, the largest pod that still fits, quotas, and containers far above or below what they asked for.
Reliability	What would break if a node went down?	Single replicas, missing probes, `:latest` images, PDBs that block drains, replicas all on one node.
Upgrades	Are we ready to upgrade?	Version support, kubelets behind, deprecated APIs still called, what would block a drain.
Networking	The api pods can’t reach other services by name, why?	Service ports and endpoints, DNS, network policies (DNS blocked, all traffic dropped), ingresses, the network plugin and IP addresses.
Service mesh	Users get 503 from reviews through Istio, why?	Subsets with no pods, routes to nothing, mTLS conflicts, authorization policies, sidecars missing or not ready; with logs on, the sidecar’s response flags counted.
Security	How secure is this cluster?	Pods that can take over their node, credentials in manifests, cluster-admins, roles made of wildcards, namespaces without Pod Security or network policy, anonymous access.
What Verbiqo can see	What should I enable?	Each item in Settings that help, as ready or not.

Every check and the words that bring it in: Kubernetes checks.

What happened earlier

Ask what changed at 3am? or what happened last night? Kubernetes keeps its events for about an hour, so Verbiqo reads the rest from the objects themselves, which carry their own times: each Deployment’s ReplicaSets (a rollout, with its revision and images), each container’s last termination (a restart or OOM kill, with the exit code), node conditions (NotReady, pressure, cordoned), failed Jobs, and the warning events still kept. A rollout or a restart days ago is still on the timeline.

/k8s prod
/prom prod
/splunk prod

Prometheus (cAdvisor, kube-state-metrics) adds a day of CPU throttling and peak memory against limits; Splunk or Elasticsearch with the container logs adds events and logs older than the hour the cluster keeps.

With your databases and logs

/k8s prod
/pg orders
/splunk prod

One timeline: the rollout at 02:58, the connection spike on the database at 03:00, the errors in the logs after. Verbiqo places events in time; it does not say one caused another.

From a terminal

In verbiqo chat, attach with /k8s, or verbiqo chat --k8s prod "why is checkout crashing?". One report, or one read in kubectl’s words:

verbiqo k8s run prod problems
verbiqo k8s command prod "get pods -n shop -o wide"

Reports for run: overview, problems, pods, restarts, workloads, nodes, events, changes, capacity, top-pods, reliability, upgrade, security, rbac, storage, network, mesh, jobs, autoscaling, quotas.

Reads for command, and for the model: get, describe, top, events, logs (when allowed), report, version, api-resources. Verbiqo parses them and sends the request itself; kubectl is never run, and nothing that changes the cluster is accepted.