fix(keycloak): org create must send a domain (KC 26 rejects domainless)

Onboarding provisioning failed at CreateOrgAndInvite with
create org: 400 "You must provide at least one domain" — Keycloak 26
requires every organization to carry at least one domain, and the
payload sent none. Membership is registry-authoritative (model B2), so
we do not use Keycloak email-domain auto-join; a synthetic per-tenant
domain <slug>.tenant.breakpilot.com satisfies the constraint, is unique
(slug is unique in the registry), and never clashes with a real
deliverable mail domain. Discovered live: with the service account now
holding manage-realm the 403 is gone and this is the next real blocker
to onboarding every customer's org on KC 26.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CNdLL9BdsWm7MCyui5ffPD

.gitea/workflows/ci.yaml

@@ -94,29 +94,53 @@ jobs:
 
   image:
     needs: [shared, test]
-    if: github.event_name == 'push' && github.ref == 'refs/heads/main' && hashFiles('Dockerfile') != ''
+    # NOTE: no hashFiles() here — at job level it evaluates BEFORE checkout
+    # against an empty workspace, so the old condition was always false and
+    # this job silently never ran (deployment sat on an Aug-06 image).
+    if: github.event_name == 'push' && github.ref == 'refs/heads/main'
     runs-on: docker
+    # The runner's default job container has no docker CLI, so
+    # docker/login-action + docker/build-push-action die with "Unable to
+    # locate executable file: docker". Same pattern as the proven
+    # breakpilot-compliance build-push-deploy.yml: run in docker:27-cli
+    # (talks to the runner's daemon) and use plain docker commands.
+    container: docker:27-cli
     steps:
-      - uses: actions/checkout@v4
+      - name: Checkout
+        run: |
+          apk add --no-cache git curl openssl
+          git clone --depth 1 --branch ${GITHUB_REF_NAME} $(echo ${GITHUB_SERVER_URL} | sed -E "s#^(https?://)#\1ci:${{ secrets.GITHUB_TOKEN }}@#")/${GITHUB_REPOSITORY}.git .
 
-      - uses: docker/login-action@v3
-        with:
-          registry: repo.breakpilot.com
-          username: ${{ secrets.REGISTRY_USER }}
-          password: ${{ secrets.REGISTRY_PASS }}
-
-      - uses: docker/build-push-action@v6
-        with:
-          push: true
-          tags: |
-            repo.breakpilot.com/breakpilot/${{ github.event.repository.name }}:sha-${{ github.sha }}
-            repo.breakpilot.com/breakpilot/${{ github.event.repository.name }}:env-stage
-
-      - uses: anchore/sbom-action@v0
-        with:
-          image: repo.breakpilot.com/breakpilot/${{ github.event.repository.name }}:sha-${{ github.sha }}
-
-      - name: orca deploy stage
-        run: orca apply --env=stage --image-tag=sha-${{ github.sha }}
+      - name: Login
         env:
-          ORCA_TOKEN: ${{ secrets.ORCA_STAGE_TOKEN }}
+          REGISTRY_USER: ${{ secrets.REGISTRY_USER }}
+          REGISTRY_PASS: ${{ secrets.REGISTRY_PASS }}
+        run: echo "$REGISTRY_PASS" | docker login repo.breakpilot.com -u "$REGISTRY_USER" --password-stdin
+
+      - name: Build + push
+        run: |
+          set -eu
+          docker build \
+            -t repo.breakpilot.com/breakpilot/tenant-registry:latest \
+            -t repo.breakpilot.com/breakpilot/tenant-registry:sha-${GITHUB_SHA} \
+            .
+          docker push repo.breakpilot.com/breakpilot/tenant-registry:latest
+          docker push repo.breakpilot.com/breakpilot/tenant-registry:sha-${GITHUB_SHA}
+
+      # Notify orca to pull :latest and recreate the service (same pattern as
+      # breakpilot-core/build-pitch-deck.yml). The stage-env deploy this used to
+      # attempt does not exist in this cluster.
+      - name: orca deploy webhook
+        env:
+          ORCA_WEBHOOK_SECRET: ${{ secrets.ORCA_WEBHOOK_SECRET }}
+          ORCA_WEBHOOK_URL: http://46.225.100.82:6880/api/v1/webhooks/github
+        run: |
+          set -eu
+          PAYLOAD="{\"ref\":\"refs/heads/main\",\"repository\":{\"full_name\":\"${GITHUB_REPOSITORY}\"},\"head_commit\":{\"id\":\"${GITHUB_SHA}\",\"message\":\"ci: tenant-registry image build\"}}"
+          SIG=$(printf '%s' "$PAYLOAD" | openssl dgst -sha256 -hmac "$ORCA_WEBHOOK_SECRET" -r | awk '{print $1}')
+          curl -sSf -k \
+            -H "Content-Type: application/json" \
+            -H "X-GitHub-Event: push" \
+            -H "X-Hub-Signature-256: sha256=$SIG" \
+            -d "$PAYLOAD" \
+            "$ORCA_WEBHOOK_URL"

Dockerfile

@@ -3,7 +3,7 @@
 #   /tenant-registry  — long-running API server
 #   /migrate          — one-shot schema migrator (Orca init container in prod)
 
-FROM golang:1.24-alpine AS build
+FROM golang:1.25-alpine AS build
 WORKDIR /src
 COPY go.mod go.sum ./
 RUN go mod download

internal/keycloak/adapter.go

@@ -75,11 +75,12 @@ type Adapter interface {
 	SyncClaims(ctx context.Context, userID string, c Claims) error
 
 	// Memberships resolves the tenants user userID (the Keycloak user id,
-	// i.e. the JWT `sub`) belongs to, as Keycloak records them. The realm
-	// has no Organizations yet, so this reads the user's attribute
-	// projection — zero or one memberships. When the realm migrates to
-	// Organizations this becomes an org-membership query and callers keep
-	// working unchanged. Returns ErrUserNotFound for an unknown user id.
+	// i.e. the JWT `sub`) belongs to, as Keycloak records them. Since the
+	// realm enabled Organizations (2026-09-01) this is an org-membership
+	// query: one membership per enabled org the user belongs to (alias =
+	// tenant slug, org attribute tenant_id = registry UUID). The legacy
+	// user-attribute projection no longer grants membership on its own.
+	// Returns ErrUserNotFound for an unknown user id.
 	Memberships(ctx context.Context, userID string) ([]Claims, error)
 
 	// Health pings the admin endpoint. Used by readyz and the cluster cold-

internal/keycloak/client_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"encoding/json"
 	"errors"
+	"io"
 	"net/http"
 	"net/http/httptest"
 	"strings"
@@ -24,6 +25,7 @@ type stubKC struct {
 	emailCalls  atomic.Int32
 	healthCalls atomic.Int32
 	syncCalls   atomic.Int32
+	lastOrgBody string
 
 	tokenFails atomic.Bool // when true, /token returns 401 once
 }
@@ -53,6 +55,8 @@ func newStubKC(t *testing.T) *stubKC {
 	mux.HandleFunc("/admin/realms/test-realm/organizations", func(w http.ResponseWriter, r *http.Request) {
 		s.orgCalls.Add(1)
 		if r.Method == http.MethodPost {
+			body, _ := io.ReadAll(r.Body)
+			s.lastOrgBody = string(body)
 			w.Header().Set("Location", s.srv.URL+"/admin/realms/test-realm/organizations/org-xyz")
 			w.WriteHeader(http.StatusCreated)
 			return
@@ -135,6 +139,12 @@ func TestHTTPAdapter_createOrgAndInvite(t *testing.T) {
 		t.Errorf("call counts: org=%d user=%d member=%d email=%d",
 			s.orgCalls.Load(), s.userCalls.Load(), s.memberCalls.Load(), s.emailCalls.Load())
 	}
+	// KC 26 rejects a domainless org; the adapter must send a synthetic
+	// per-slug domain so onboarding actually provisions.
+	if !strings.Contains(s.lastOrgBody, `"domains"`) ||
+		!strings.Contains(s.lastOrgBody, "acme.tenant.breakpilot.com") {
+		t.Errorf("org create body missing synthetic domain: %s", s.lastOrgBody)
+	}
 }
 
 func TestHTTPAdapter_emailMissingAdminEmailRejected(t *testing.T) {

internal/keycloak/orgs.go

@@ -13,6 +13,11 @@ import (
 
 // ─── organizations API ───────────────────────────────────────────────────
 
+// orgDomainSuffix namespaces the synthetic org domain. The slug is unique in
+// the registry, so "<slug>.tenant.breakpilot.com" is unique per organization
+// and never a real deliverable mail domain we might clash with.
+const orgDomainSuffix = ".tenant.breakpilot.com"
+
 type orgCreate struct {
 	Name        string              `json:"name"`
 	Alias       string              `json:"alias"`
@@ -50,6 +55,16 @@ func (a *HTTPAdapter) CreateOrgAndInvite(ctx context.Context, in InviteInput) (*
 		Name:        in.Name,
 		Alias:       in.Slug,
 		Description: fmt.Sprintf("Auto-provisioned from tenant-registry %s", in.TenantID),
+		// Keycloak 26 rejects an organization with no domain ("You must
+		// provide at least one domain"). Membership is registry-authoritative
+		// (model B2), so we do NOT use Keycloak's email-domain auto-join; a
+		// synthetic per-tenant domain derived from the unique slug satisfies
+		// the constraint without depending on the customer's real mail domain
+		// (which may be a shared public domain and would collide across
+		// tenants). Unverified is fine — verification only gates auto-join.
+		Domains: []map[string]any{
+			{"name": in.Slug + orgDomainSuffix, "verified": false},
+		},
 		Attributes: map[string][]string{
 			"tenant_id": {in.TenantID},
 		},

internal/keycloak/users.go

@@ -18,7 +18,33 @@ type userRepresentation struct {
 	Attributes map[string][]string `json:"attributes"`
 }
 
-// Memberships implements Adapter against GET /admin/realms/{realm}/users/{id}.
+// memberOrgRepresentation is the slice of OrganizationRepresentation the
+// membership query needs: the alias IS the tenant slug and the org
+// attribute "tenant_id" carries the registry tenant UUID (both written by
+// CreateOrgAndInvite, or provisioned by the realm admin for pre-existing
+// tenants).
+type memberOrgRepresentation struct {
+	ID         string              `json:"id"`
+	Alias      string              `json:"alias"`
+	Enabled    bool                `json:"enabled"`
+	Attributes map[string][]string `json:"attributes"`
+}
+
+// Memberships implements Adapter with Keycloak Organizations as the
+// authoritative membership source (prod shape, realm orgs enabled
+// 2026-09-01):
+//
+//  1. GET /users/{id} — preserves ErrUserNotFound semantics and supplies
+//     the per-user claim attributes (org_roles / products / plan /
+//     tenant_status) that SyncClaims maintains.
+//  2. GET /organizations/members/{id}/organizations — the memberships.
+//
+// One Claims entry per ENABLED organization: TenantID = org attribute
+// "tenant_id" (registry UUID), TenantSlug = org alias. A disabled org
+// grants no membership. A user in no organization has zero memberships —
+// the legacy user-attribute tenant projection is NO LONGER consulted for
+// membership, so a stale tenant_id/tenant_slug user attribute cannot
+// grant access the org model has revoked.
 func (a *HTTPAdapter) Memberships(ctx context.Context, userID string) ([]Claims, error) {
 	var u userRepresentation
 	resp, err := a.adminCall(ctx, http.MethodGet, "/users/"+userID, nil, &u)
@@ -33,25 +59,34 @@ func (a *HTTPAdapter) Memberships(ctx context.Context, userID string) ([]Claims,
 		_ = resp.Body.Close()
 		return nil, fmt.Errorf("keycloak get user: %d", resp.StatusCode)
 	}
-	return claimsFromAttributes(u.Attributes), nil
-}
 
-// claimsFromAttributes builds the membership list from a user's attribute
-// projection. A user with no tenant_id and no tenant_slug attribute simply
-// has no memberships — that is a valid state, not an error.
-func claimsFromAttributes(attrs map[string][]string) []Claims {
-	c := Claims{
-		TenantID:     attrValue(attrs, "tenant_id"),
-		TenantSlug:   attrValue(attrs, "tenant_slug"),
-		OrgRoles:     attrValues(attrs, "org_roles"),
-		Products:     attrValues(attrs, "products"),
-		Plan:         attrValue(attrs, "plan"),
-		TenantStatus: attrValue(attrs, "tenant_status"),
+	var orgs []memberOrgRepresentation
+	resp, err = a.adminCall(
+		ctx, http.MethodGet, "/organizations/members/"+userID+"/organizations", nil, &orgs,
+	)
+	if err != nil {
+		return nil, err
 	}
-	if c.TenantID == "" && c.TenantSlug == "" {
-		return []Claims{}
+	if resp.StatusCode/100 != 2 {
+		_ = resp.Body.Close()
+		return nil, fmt.Errorf("keycloak member organizations: %d", resp.StatusCode)
 	}
-	return []Claims{c}
+
+	claims := []Claims{}
+	for _, org := range orgs {
+		if !org.Enabled {
+			continue
+		}
+		claims = append(claims, Claims{
+			TenantID:     attrValue(org.Attributes, "tenant_id"),
+			TenantSlug:   org.Alias,
+			OrgRoles:     attrValues(u.Attributes, "org_roles"),
+			Products:     attrValues(u.Attributes, "products"),
+			Plan:         attrValue(u.Attributes, "plan"),
+			TenantStatus: attrValue(u.Attributes, "tenant_status"),
+		})
+	}
+	return claims, nil
 }
 
 func attrValue(attrs map[string][]string, key string) string {

internal/keycloak/users_test.go

@@ -9,9 +9,13 @@ import (
 	"testing"
 )
 
-// stubUsersKC is a users-endpoint-only KC look-alike; stubKC (client_test.go)
-// covers the org/invite paths and doesn't register GET /users/{id}.
-func stubUsersKC(t *testing.T, users map[string]userRepresentation) *httptest.Server {
+// stubUsersKC is a users+organizations KC look-alike; stubKC (client_test.go)
+// covers the org-create/invite paths and doesn't register these reads.
+func stubUsersKC(
+	t *testing.T,
+	users map[string]userRepresentation,
+	memberOrgs map[string][]memberOrgRepresentation,
+) *httptest.Server {
 	t.Helper()
 	mux := http.NewServeMux()
 	mux.HandleFunc("/realms/test-realm/protocol/openid-connect/token", func(w http.ResponseWriter, _ *http.Request) {
@@ -28,6 +32,14 @@ func stubUsersKC(t *testing.T, users map[string]userRepresentation) *httptest.Se
 		w.Header().Set("Content-Type", "application/json")
 		_ = json.NewEncoder(w).Encode(u)
 	})
+	mux.HandleFunc("GET /admin/realms/test-realm/organizations/members/{id}/organizations", func(w http.ResponseWriter, r *http.Request) {
+		orgs, ok := memberOrgs[r.PathValue("id")]
+		if !ok {
+			orgs = []memberOrgRepresentation{}
+		}
+		w.Header().Set("Content-Type", "application/json")
+		_ = json.NewEncoder(w).Encode(orgs)
+	})
 	srv := httptest.NewServer(mux)
 	t.Cleanup(srv.Close)
 	return srv
@@ -40,10 +52,11 @@ func usersAdapter(srv *httptest.Server) *HTTPAdapter {
 }
 
 func TestHTTPAdapter_Memberships(t *testing.T) {
-	srv := stubUsersKC(t, map[string]userRepresentation{
+	users := map[string]userRepresentation{
 		"u-1": {ID: "u-1", Username: "test@breakpilot.com", Enabled: true, Attributes: map[string][]string{
-			"tenant_id":     {"acme-001"},
-			"tenant_slug":   {"acme"},
+			// legacy projection attrs — MUST NOT grant membership on their own
+			"tenant_id":     {"stale-legacy-001"},
+			"tenant_slug":   {"stale"},
 			"tenant_status": {"active"},
 			"plan":          {"Scale"},
 			"org_roles":     {"IT_ADMIN", "FINANCE"},
@@ -51,10 +64,30 @@ func TestHTTPAdapter_Memberships(t *testing.T) {
 			"products": {"compliance##certifai"},
 		}},
 		"u-2": {ID: "u-2", Username: "bare@breakpilot.com", Enabled: true},
-	})
+		"u-3": {ID: "u-3", Username: "multi@breakpilot.com", Enabled: true},
+		"u-4": {ID: "u-4", Username: "attrs-only@breakpilot.com", Enabled: true, Attributes: map[string][]string{
+			"tenant_id":   {"acme-001"},
+			"tenant_slug": {"acme"},
+		}},
+	}
+	memberOrgs := map[string][]memberOrgRepresentation{
+		"u-1": {{
+			ID: "org-1", Alias: "acme", Enabled: true,
+			Attributes: map[string][]string{"tenant_id": {"2f875d6a-1d94-433a-b2ec-8529451a2d89"}},
+		}},
+		"u-3": {
+			{ID: "org-1", Alias: "acme", Enabled: true,
+				Attributes: map[string][]string{"tenant_id": {"2f875d6a-1d94-433a-b2ec-8529451a2d89"}}},
+			{ID: "org-2", Alias: "globex", Enabled: true,
+				Attributes: map[string][]string{"tenant_id": {"7c3f2b10-0000-4000-8000-000000000042"}}},
+			{ID: "org-3", Alias: "disabled-co", Enabled: false,
+				Attributes: map[string][]string{"tenant_id": {"9e9e9e9e-0000-4000-8000-000000000099"}}},
+		},
+	}
+	srv := stubUsersKC(t, users, memberOrgs)
 	a := usersAdapter(srv)
 
-	t.Run("attribute projection becomes one membership", func(t *testing.T) {
+	t.Run("org membership becomes the claim, org attrs are authoritative", func(t *testing.T) {
 		got, err := a.Memberships(context.Background(), "u-1")
 		if err != nil {
 			t.Fatalf("memberships: %v", err)
@@ -63,8 +96,14 @@ func TestHTTPAdapter_Memberships(t *testing.T) {
 			t.Fatalf("want 1 membership, got %d", len(got))
 		}
 		c := got[0]
-		if c.TenantID != "acme-001" || c.TenantSlug != "acme" || c.Plan != "Scale" || c.TenantStatus != "active" {
-			t.Errorf("scalar claims wrong: %+v", c)
+		// tenant identity comes from the ORG (alias + tenant_id attribute),
+		// never from the user's legacy projection attributes
+		if c.TenantID != "2f875d6a-1d94-433a-b2ec-8529451a2d89" || c.TenantSlug != "acme" {
+			t.Errorf("org identity wrong: %+v", c)
+		}
+		// per-user claim attrs still ride along
+		if c.Plan != "Scale" || c.TenantStatus != "active" {
+			t.Errorf("user claim attrs wrong: %+v", c)
 		}
 		if len(c.OrgRoles) != 2 || c.OrgRoles[0] != "IT_ADMIN" || c.OrgRoles[1] != "FINANCE" {
 			t.Errorf("org_roles wrong: %v", c.OrgRoles)
@@ -74,7 +113,7 @@ func TestHTTPAdapter_Memberships(t *testing.T) {
 		}
 	})
 
-	t.Run("user without tenant attributes has zero memberships", func(t *testing.T) {
+	t.Run("user in no org has zero memberships", func(t *testing.T) {
 		got, err := a.Memberships(context.Background(), "u-2")
 		if err != nil {
 			t.Fatalf("memberships: %v", err)
@@ -84,6 +123,29 @@ func TestHTTPAdapter_Memberships(t *testing.T) {
 		}
 	})
 
+	t.Run("legacy tenant attributes alone grant NO membership", func(t *testing.T) {
+		got, err := a.Memberships(context.Background(), "u-4")
+		if err != nil {
+			t.Fatalf("memberships: %v", err)
+		}
+		if len(got) != 0 {
+			t.Fatalf("attribute projection must not grant membership, got %+v", got)
+		}
+	})
+
+	t.Run("multiple orgs give multiple memberships, disabled org skipped", func(t *testing.T) {
+		got, err := a.Memberships(context.Background(), "u-3")
+		if err != nil {
+			t.Fatalf("memberships: %v", err)
+		}
+		if len(got) != 2 {
+			t.Fatalf("want 2 memberships (disabled org skipped), got %d", len(got))
+		}
+		if got[0].TenantSlug != "acme" || got[1].TenantSlug != "globex" {
+			t.Errorf("slugs wrong: %+v", got)
+		}
+	})
+
 	t.Run("unknown user is ErrUserNotFound", func(t *testing.T) {
 		_, err := a.Memberships(context.Background(), "nope")
 		if !errors.Is(err, ErrUserNotFound) {
@@ -92,15 +154,24 @@ func TestHTTPAdapter_Memberships(t *testing.T) {
 	})
 }
 
-func TestMock_Memberships(t *testing.T) {
-	m := NewMock()
-	if _, err := m.Memberships(context.Background(), "ghost"); !errors.Is(err, ErrUserNotFound) {
-		t.Fatalf("want ErrUserNotFound, got %v", err)
-	}
-	want := Claims{TenantSlug: "acme", OrgRoles: []string{"IT_ADMIN"}}
-	m.Claims["u-1"] = want
-	got, err := m.Memberships(context.Background(), "u-1")
-	if err != nil || len(got) != 1 || got[0].TenantSlug != "acme" {
-		t.Fatalf("got %+v err %v", got, err)
+func TestHTTPAdapter_Memberships_OrgQueryFailure(t *testing.T) {
+	mux := http.NewServeMux()
+	mux.HandleFunc("/realms/test-realm/protocol/openid-connect/token", func(w http.ResponseWriter, _ *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_ = json.NewEncoder(w).Encode(map[string]any{"access_token": "t", "expires_in": 60})
+	})
+	mux.HandleFunc("GET /admin/realms/test-realm/users/{id}", func(w http.ResponseWriter, _ *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_ = json.NewEncoder(w).Encode(userRepresentation{ID: "u-1", Enabled: true})
+	})
+	mux.HandleFunc("GET /admin/realms/test-realm/organizations/members/{id}/organizations", func(w http.ResponseWriter, _ *http.Request) {
+		w.WriteHeader(http.StatusInternalServerError)
+	})
+	srv := httptest.NewServer(mux)
+	t.Cleanup(srv.Close)
+
+	_, err := usersAdapter(srv).Memberships(context.Background(), "u-1")
+	if err == nil {
+		t.Fatal("want error when the org query fails, got nil")
 	}
 }