Tudo no Git, limites de recursos em cada pod, service accounts com o mínimo privilégio.
Constitution — Kubernetes platform
Version 1.0.0 · Ratified 2026-11-10
These principles hold for every spec, plan and task of the platform. A spec may add rules; it never weakens these.
-
Everything in Git. Every manifest, Helm values file and cluster setting lives in Git and is applied by a GitOps controller (Argo CD or Flux).
kubectl applyorkubectl editby hand on a shared cluster is not allowed, and the controller reverts drift. -
Resource limits on every pod. Every container declares CPU and memory requests and a memory limit, enforced by a
LimitRangeand an admission policy. A workload without them is rejected by the cluster, not only in review. -
Least-privilege service accounts. Each workload has its own service account with a Role that lists only the verbs and resources it uses;
automountServiceAccountTokenis off where the API is not needed. No workload is bound to cluster-admin. -
Health checks and disruption budgets. Every deployment has readiness and liveness probes, at least two replicas when it serves traffic, and a
PodDisruptionBudget. Node upgrades never take a service fully down. -
Restricted pod security. Namespaces enforce the Pod Security
restrictedprofile: non-root, read-only root filesystem where possible, no privilege escalation and dropped capabilities. Exceptions are listed in the plan with a reason. -
Network is denied by default. Each namespace starts with a default-deny
NetworkPolicy, and allowed traffic is declared per workload. A service reachable from everywhere fails the review. -
Charts are versioned and linted. Helm charts are versioned, pass
helm lintand a schema check (kubeconform) in CI, and images are referenced by digest or immutable tag, neverlatest. -
Secrets never sit in Git in clear. Secrets come from an external secret manager through External Secrets or Sealed Secrets. A plain
Secretmanifest with data in Git fails CI.