Inspect multi-document Kubernetes manifests for removed API versions, missing resource requests and Service selectors that match no pod template.
Waiting for a manifest
Nothing is applied and no cluster is contacted.
The findings appear here
Removed API versions, containers with no resource requests, a Service selecting labels no pod carries, and the settings a restricted Pod Security Standard will reject.
This reads the structure of your manifests rather than validating them against the Kubernetes schema. It cannot tell you a field is misspelled or has the wrong type — only the API server can do that, and nothing here contacts one. Environment variable values are never read.
Kubernetes Manifest Inspector runs entirely on your device — your manifests are never uploaded, and the values of environment variables and Secrets are never read.
You’re on the Free plan. Compare plans
The check worth having here is the one that spans documents. A Service whose selector matches no pod template in the set is the most common Kubernetes misconfiguration there is, and it is invisible when you read either document alone: the Service is valid, the Deployment is valid, and the API server accepts both without complaint. Nothing fails. The Service simply has no endpoints, and requests to it hang or refuse with no event anywhere explaining why.
Finding that requires holding every document in the file at once and comparing each Service's selector against every workload's pod template labels — which is precisely what a person reading a 600-line manifest cannot reliably do, and what this does first.
The rest are the things that hold regardless of cluster version. A container with no resource requests lands its pod in the BestEffort QoS class, first in line for eviction when a node runs short. An API version that was removed is not deprecated — the endpoint is gone, and the error reads like a missing resource type. A container without `runAsNonRoot` will be refused outright by a namespace enforcing the restricted Pod Security Standard.
There is no Kubernetes schema here, so this cannot tell you a field is misspelled or has the wrong type. `replicas: "3"` as a string is a real error and only the API server will report it. `kubectl apply --dry-run=server` is the tool for that, and it is worth running as well as this rather than instead of it.
Environment variable names are collected; their values are never read. A manifest pasted into a tool routinely contains a real password, and the only safe way to handle that is to never look.
Almost always because its selector does not match the labels on the pod template, rather than the labels on the Deployment itself — those are two different label sets and only the template's matter. This compares them across every document and names the mismatch.
No, and it says so rather than implying otherwise. It has no schema, so a misspelled field or a wrong type will pass here and be rejected on apply. Use it for the structural and cross-document problems, and `kubectl apply --dry-run=server` for the schema.
Nothing leaves your browser, and beyond that the inspector never reads the value of an environment variable or a Secret key — only the names. So a value cannot appear in a finding even by accident.
Because of what it causes rather than what it is. Without requests the pod is BestEffort, which makes it the first thing the kubelet evicts under node pressure — so the pod that disappears is not the one causing the problem.