Please report security issues privately through GitHub's private vulnerability reporting on this repository, not as a public issue.

We aim to acknowledge reports within a few working days.

KMate is pre-1.0. Only the latest commit on main receives fixes.

KMate is pre-1.0 and not yet hardened for exposure to untrusted networks. The current implementation includes:

• Multi-user accounts. Each account has a global role (viewer, operator or admin). A global admin can reach every cluster; everyone else reaches a cluster only through an explicit grant, optionally limited to specific namespaces, and clusters without a grant are not listed.
• Local password login and optional OIDC single sign-on with any standards-compliant provider (authorization-code flow with PKCE). Password login can be turned off with localLogin: false once SSO works; it stays on when no issuer is configured.
• Kubernetes user/group impersonation, on by default in the agent chart (rbac.impersonate: true). User-scoped operations impersonate the signed-in user with their configured Kubernetes groups plus the kmate:<global role> group; cached reads come from an agent-run informer that checks the caller's access first. Within the KMate grant, your Kubernetes RBAC decides what each user may do. With impersonation off, every user shares the agent's ServiceAccount and the cluster shows a "shared identity" badge.
• Agents enrol with a single-use token that expires after 24 hours. Agent mutual TLS with certificate issuance and automatic renewal is available; it defaults to agentMTLS: optional for migration, so bearer tokens are still accepted. Set agentMTLS: required once every agent reports a certificate.

Until it is hardened, run the hub on a private network. The hub chart defaults to insecure: true (plaintext inside the cluster): configure a TLS-terminating ingress or gateway (neither is enabled by default), or set insecure: false with tls.secretName to terminate in the hub.

By default the agent chart grants read-only verbs (get, list, watch) on all resources, including Secrets (rbac.secretsRead: true), plus an impersonator ClusterRole. rbac.write and rbac.exec are opt-in. Set rbac.secretsRead: false to exclude Secrets from the generated reader/writer roles and deny Secret and Helm requests; the agent keeps get/update/patch on its own identity Secret and namespace-scoped permission to create Secrets.

See docs/06-security.md for the threat model and docs/12-oidc.md for single sign-on setup.