Why Charter¶
The problem¶
Every application that signs users in with Entra ID needs an app registration. In most organisations these are created by hand in the portal, or by a ticket to a central identity team. Both routes have the same weaknesses:
- Slow and manual. A team waits days for a registration and a secret, and again for every redirect URI change.
- Weak evidence. The ticket says what was asked for; nothing proves the registration still looks like that six months later.
- Secrets by email. Client secrets are copied out of the portal and passed around, and they expire on dates nobody tracks, which causes outages.
- No owner. Registrations outlive the teams that asked for them.
What Charter does about it¶
| Need | How Charter meets it |
|---|---|
| Self-service without losing control | Admins create a client once; the owning team then maintains its own redirect URIs through the Kubernetes API, within a domain policy. |
| Every change authorised | Creation and admin-only changes need a requester, a different approver and a change reference. The controller re-checks this before touching Entra. |
| Every change evidenced | Each Graph mutation is recorded as attempted before the call and succeeded or failed after it, as JSON and as a Kubernetes Event. |
| Configuration stays as approved | Hand edits in the portal (name, redirect URIs, group claims, owners, extra secrets) are detected and reverted, with a drift_detected record. |
| No secrets passed by hand | The client secret is delivered straight into a Kubernetes Secret, and optionally OpenBao. Nobody sees it. |
| No surprise expiry | Secrets rotate automatically when a third of their lifetime is left; the old one stays valid for a grace window. |
| Least privilege | The controller holds only Application.ReadWrite.OwnedBy, so it can only touch registrations it created. |
| No standing credential for the controller | It authenticates with workload identity federation; there is no secret to leak. |
Prior art¶
Checked 2026-11-12.
| Option | Manages Entra apps? | Notes |
|---|---|---|
| Cluster API Azure (CAPZ) | No | Uses Entra and workload identity to provision clusters; no app-registration CRDs. |
| Azure Service Operator v2 | Yes, recently | entra.azure.com Application since v2.21.0, ServicePrincipal since v2.22.0, plus SecurityGroup. No federated credential or client secret resource. |
| Crossplane provider-azuread | Yes, broadest | Applications, service principals, federated identity credentials, passwords. Terraform-based (Upjet), so it carries a heavy runtime. |
| Charter | Yes, narrow | One opinionated resource that also does secret delivery, rotation, Dex config, approvals and auditing, which none of the above do. |
Reuse. If Azure Service Operator or Crossplane is already in place, Charter's
Graph module could be swapped for one that writes their Application and
ServicePrincipal objects, keeping the parts they lack: approvals, auditing,
rotation, secret delivery and Dex config. The reconcile shape stays the same.
Non-goals¶
- Not a general Entra management tool. Charter manages web app
registrations for sign-in, with single-tenant audience (
AzureADMyOrg). It does not manage users, groups, conditional access or API permissions. - No group assignment yet. Assigning groups to the app needs
AppRoleAssignment.ReadWrite.All, a privileged permission that should be approved separately. - Not an identity provider. Dex or Entra is the identity provider; Charter only wires them together.