Merge quick/users-http-admin: HTTP admin-only users API + client methods (report 0014)

Add a gotcha describing the unified-storage model (the server writes
users to the same store/KV as rooms), the admin-only HTTP surface, and
the CLI-seeds-admin-#0 bootstrap. Bump the version 0.9.0 -> 0.10.0 and
add the capability growth log entry for the new HTTP admin users API.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

app.md

@@ -2,7 +2,7 @@
 name: unibus
 lang: go
 domain: infra
-version: 0.8.0
+version: 0.10.0
 description: "Bus de mensajería unificado sobre NATS+JetStream con cifrado E2E por room (megolm/olm reducido): service de membresía/claves, librería cliente y peers demo."
 tags: [service, messaging, nats, e2e]
 uses_functions:
@@ -122,6 +122,21 @@ Para apuntar a un NATS externo en producción: `--nats-url nats://host:4222` en
   las rutas GET de lectura. Confía en la red interna. Las rutas mutantes
   (`/rooms`, `/invite`, `/rekey`) sí exigen firma Ed25519 del owner sobre los
   bytes canónicos de la request. Endurecer es fase posterior.
+- **Gestión de usuarios: storage unificado, alta por dos vías.** El allowlist de
+  usuarios vive en el MISMO store que las rooms (`pkg/membership.Store`): SQLite en
+  single-node, JetStream KV replicado (`UNIBUS_users`) en cluster. El `Server` ya
+  tiene ese store privilegiado abierto (es quien sirve el KV en cada nodo), así que
+  expone `GET/POST /users` y `POST /users/{signpub}/revoke` como API HTTP admin-only,
+  simétrica con las rutas de rooms: el panel de administración firma como admin y el
+  server ejecuta la mutación contra el mismo store. El panel NO necesita `--db`, ni la
+  identidad interna, ni correr en un nodo del cluster; funciona idéntico en single-node
+  y cluster. La autorización es default-deny: solo un firmante que el store confirma como
+  `role == "admin"` activo pasa, cualquier otro recibe 403 (encima de la firma+nonce+TLS
+  ya existentes). La CLI `membershipd user add --store kv` sigue existiendo SOLO para
+  sembrar el admin #0 (bootstrap del huevo-gallina: sin un admin sembrado no hay quién
+  firme el primer `POST /users`); a partir de ahí toda la gestión es HTTP admin-only. El
+  alta es idempotente igual que la CLI: re-alta de una clave ya registrada = 409, sin
+  sobrescribir ni elevar rol; el revoke es un flip de status (sin hard-delete), auditable.
 - **Identidad = secreto crítico.** El archivo de identidad (`worker.id`,
   `chat.id`) contiene las claves privadas (Ed25519 + X25519). Se escribe 0600.
   Perderlo = mensajes ilegibles, sin recuperación. Trátalo como una clave SSH.
@@ -154,6 +169,52 @@ agent.<nombre>.{in,out}   inbox/outbox de agente LLM (agent.scout.in)
 
 ## Capability growth log
 
+- v0.10.0 (2026-06-07) — API HTTP admin-only de gestión de usuarios, cerrando la
+  última asimetría del control plane: las rooms tenían superficie HTTP firmada
+  (`POST /rooms`, etc.) pero los users solo se gestionaban por CLI local o acceso
+  directo al store. Se añaden `GET /users` (lista completa, incluidos revocados),
+  `POST /users` (alta `{sign_pub, handle, role}`: valida hex de 64 chars + role en
+  `{admin, member}`, 409 idempotente que no sobrescribe ni eleva rol) y
+  `POST /users/{signpub}/revoke` (flip de status, sin hard-delete). Los tres pasan por
+  un helper `requireAdmin` default-deny que confirma contra el store que el firmante
+  autenticado es un user `role == "admin"` activo (el endpoint id es un hash one-way de
+  la clave, así que el contexto lleva ahora también el `sign_pub` hex del firmante para
+  resolver `GetUser`); cualquier otro firmante recibe 403, encima de la firma+nonce+TLS+
+  enforce ya heredadas del middleware. NO se abre conexión KV nueva ni se usa la identidad
+  interna: el server escribe vía su `s.store` privilegiado, el MISMO que las rooms (SQLite
+  single-node, KV `UNIBUS_users` en cluster). `pkg/client` gana `ListUsers/AddUser/RevokeUser`
+  (tipo plano `UserInfo`) firmando como admin, así la pestaña Users del panel deja de
+  necesitar `--db`/acceso KV directo. La CLI `membershipd user add --store kv` queda SOLO
+  para sembrar el admin #0 (bootstrap). La validación de `sign_pub` se unifica en
+  `membership.ValidateSignPubHex`, reusada por la CLI y los handlers. Tests nuevos:
+  no-admin → 403 en los tres endpoints, roundtrip admin add→list→revoke, y validación
+  (hex inválido → 400, role inválido → 400, re-alta → 409), más un test de cliente contra
+  un membershipd embebido. Cambios 100% aditivos: el comportamiento single-node y de las
+  rutas de rooms no cambia; vet/build/test verdes.
+- v0.9.0 (2026-06-07) — cierre de los gaps que el despliegue del cluster (report
+  0011) dejó abiertos (report 0012). (GAP A) Nueva capability `membershipd user
+  add|list|revoke --store kv`: alta/baja de usuarios contra el KV replicado del
+  cluster EN MARCHA, sin el procedimiento de parar-sembrar-rearrancar. Usa la
+  conexión interna privilegiada — el daemon persiste su identidad de servicio con
+  `--internal-id-file` (cada nodo genera/carga la suya, 0600 junto a las claves TLS)
+  y la CLI, ejecutada por loopback en un nodo, presenta esa nkey que el
+  autenticador reconoce con permisos plenos de JetStream; ninguna identidad de
+  usuario normal puede tocar los buckets `KV_UNIBUS_*` bajo la ACL por-subject. El
+  alta es idempotente (re-alta de la misma clave = `ErrUserExists` explícito, sin
+  sobrescribir ni elevar rol), commitea con quórum 2/3 (HA, imprime
+  `followers_current`) y rechaza un destino remoto sin `--ca` (igual que
+  `migrate-to-kv`). (GAP B) Nuevo `cmd/clientcheck`: verificación end-to-end real
+  con un cliente autenticado (identidad operator, nkey+TLS+https) que crea una room
+  E2E, publica y recibe descifrado contra el cluster vivo, incluido un nodo parado a
+  media transmisión donde el cliente hace failover a un superviviente y sigue
+  recibiendo con cero pérdida (quórum 2/3) — el plano de datos que el chaos test del
+  0011 nunca probó. (GAP C) Runbook `deploy/cluster/README.md` corregido: el orden
+  de arranque "magnus solo y verifica healthz" deadlockeaba (un nodo solo no tiene
+  quórum del meta-group y nunca sirve healthz); se documenta el arranque por quórum,
+  que R1 es un SPOF inservible (ir directo a R3) y la nueva vía de alta con el
+  cluster vivo. La plantilla de deploy (unit + `deploy-cluster.sh`) emite ya
+  `INTERNAL_ID_FILE` y el flag. Verificado contra los 3 VPS reales (magnus + homer +
+  datardos); posture enforce+ACL+TLS+R3 intacta.
 - v0.8.0 (2026-06-07) — completar y endurecer el cluster (issue 0006, fases
   0006a–0006g) que cierra los bloqueantes de la auditoría dedicada del cluster
   (report 0008) y cablea el control plane descentralizado que 0003 dejó a medias.

cmd/clientcheck/main.go

@@ -0,0 +1,260 @@
+// Command clientcheck is an end-to-end verification client for a live unibus
+// cluster (issue 0011 GAP B). The 0011 chaos test validated only the control
+// plane (healthz + meta/stream-leader failover + KV readable with 2/3); it never
+// connected an authenticated bus client (nkey + TLS) to create a room and
+// publish/subscribe through it, least of all across a node loss. clientcheck does
+// exactly that with a real identity (the operator), so the data-plane end-to-end
+// path — connect, create an E2E room, publish, receive decrypted — is exercised
+// against the running cluster, including while a node is stopped.
+//
+// It is a reusable tool, not a throwaway script: point it at the cluster's CA,
+// an identity file, and the NATS + control-plane seed lists.
+//
+//	# golden: connect, create an E2E room, publish N, confirm N decrypted back
+//	clientcheck --ca ca.crt --identity-file operator.id \
+//	  --nats-seeds nats://A:4250,nats://B:4250,nats://C:4250 \
+//	  --ctrl-seeds https://A:8470,https://B:8470,https://C:8470 --messages 5
+//
+//	# loop: publish a counter every interval for the duration, logging the node
+//	# it is attached to — stop a node mid-run (systemctl stop membershipd-cluster)
+//	# and watch it fail over to a survivor and keep receiving (quorum 2/3).
+//	clientcheck ... --mode loop --duration 45s --interval 1s
+package main
+
+import (
+	"crypto/rand"
+	"encoding/hex"
+	"flag"
+	"fmt"
+	"log"
+	"sort"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/enmanuel/unibus/pkg/busauth"
+	"github.com/enmanuel/unibus/pkg/client"
+	"github.com/enmanuel/unibus/pkg/frame"
+	"github.com/enmanuel/unibus/pkg/room"
+)
+
+func main() {
+	var (
+		caPath    = flag.String("ca", "", "bus CA cert pinning TLS on both planes (required for a secured cluster)")
+		idFile    = flag.String("identity-file", "", "path to the client identity JSON (e.g. `pass show unibus/operator-identity` written 0600) (required)")
+		natsSeeds = flag.String("nats-seeds", "", "comma-separated NATS urls of the cluster nodes (required)")
+		ctrlSeeds = flag.String("ctrl-seeds", "", "comma-separated control-plane https urls of the cluster nodes (required)")
+		subject   = flag.String("subject", "test.gapcheck", "test room subject PREFIX; a random token is appended so runs never collide with real rooms")
+		messages  = flag.Int("messages", 5, "golden mode: number of messages to publish and expect back")
+		mode      = flag.String("mode", "golden", "golden (publish N, verify N decrypted) | loop (publish a counter for --duration, for failover testing)")
+		duration  = flag.Duration("duration", 30*time.Second, "loop mode: how long to keep publishing")
+		interval  = flag.Duration("interval", 1*time.Second, "loop mode: delay between published messages")
+	)
+	flag.Parse()
+
+	if *idFile == "" || *natsSeeds == "" || *ctrlSeeds == "" {
+		log.Fatalf("clientcheck: --identity-file, --nats-seeds and --ctrl-seeds are required")
+	}
+
+	id, err := client.LoadIdentity(*idFile)
+	if err != nil {
+		log.Fatalf("clientcheck: load identity: %v", err)
+	}
+	natsList := splitCSV(*natsSeeds)
+	ctrlList := splitCSV(*ctrlSeeds)
+	if len(natsList) == 0 || len(ctrlList) == 0 {
+		log.Fatalf("clientcheck: empty --nats-seeds or --ctrl-seeds")
+	}
+
+	// Build the secure client options: nkey on the data plane, TLS pinned to the
+	// bus CA on both planes, and the FULL seed lists so nats.go fails over to a
+	// surviving node when the attached one dies (the failover this tool verifies).
+	opts := client.Options{
+		NatsServers: natsList[1:],
+		CtrlURLs:    ctrlList[1:],
+	}
+	if *caPath != "" {
+		tlsCfg, err := busauth.LoadCATLSConfig(*caPath)
+		if err != nil {
+			log.Fatalf("clientcheck: load CA: %v", err)
+		}
+		opts.UseNkey = true
+		opts.TLS = tlsCfg
+		opts.CtrlTLS = tlsCfg
+		for _, u := range ctrlList {
+			if !strings.HasPrefix(u, "https://") {
+				log.Fatalf("clientcheck: control URL %q must be https:// when --ca is set", u)
+			}
+		}
+	}
+
+	c, err := client.NewWithOptions(natsList[0], ctrlList[0], id, opts)
+	if err != nil {
+		log.Fatalf("clientcheck: connect: %v", err)
+	}
+	defer c.Close()
+	log.Printf("connected: endpoint=%s nats=%s", c.Endpoint().ID, c.ConnectedServer())
+
+	// Create an EPHEMERAL E2E room (encrypted + signed, NOT persisted): the test
+	// stays end-to-end encrypted (the cluster requires encryption on a public
+	// bind) while leaving no durable JetStream stream behind. The random subject
+	// token guarantees the room is unique and never a real room.
+	rnd := make([]byte, 8)
+	if _, err := rand.Read(rnd); err != nil {
+		log.Fatalf("clientcheck: random: %v", err)
+	}
+	subj := fmt.Sprintf("%s.%s", *subject, hex.EncodeToString(rnd))
+	policy := room.Policy{Encrypt: true, Persist: false, SignMsgs: true}
+	roomID, err := c.CreateRoom(subj, policy)
+	if err != nil {
+		log.Fatalf("clientcheck: create room: %v", err)
+	}
+	log.Printf("created E2E room: id=%s subject=%s (encrypt=%v sign=%v persist=%v)", roomID, subj, policy.Encrypt, policy.SignMsgs, policy.Persist)
+
+	// Under the per-subject ACL, NATS freezes permissions at connect time, so the
+	// just-created room's subject is not yet publishable/subscribable on the live
+	// connection. RefreshSession reconnects so the authenticator re-derives the
+	// ACL (now including this room) — the post-0006 contract every client follows
+	// after a membership change.
+	if err := c.RefreshSession(); err != nil {
+		log.Fatalf("clientcheck: refresh session: %v", err)
+	}
+
+	switch *mode {
+	case "golden":
+		runGolden(c, roomID, *messages)
+	case "loop":
+		runLoop(c, roomID, *duration, *interval)
+	default:
+		log.Fatalf("clientcheck: --mode must be golden or loop, got %q", *mode)
+	}
+}
+
+// runGolden subscribes, publishes n messages, and asserts all n come back
+// decrypted. Exits non-zero if any are missing.
+func runGolden(c *client.Client, roomID string, n int) {
+	var mu sync.Mutex
+	got := map[string]bool{}
+	sub, err := c.Subscribe(roomID, func(_ frame.Frame, plaintext []byte) {
+		mu.Lock()
+		got[string(plaintext)] = true
+		mu.Unlock()
+	})
+	if err != nil {
+		log.Fatalf("clientcheck: subscribe: %v", err)
+	}
+	defer sub.Unsubscribe()
+	time.Sleep(300 * time.Millisecond) // let the subscription settle
+
+	want := make([]string, n)
+	for i := 0; i < n; i++ {
+		msg := fmt.Sprintf("gapcheck-e2e-%d", i)
+		want[i] = msg
+		if err := c.Publish(roomID, []byte(msg)); err != nil {
+			log.Fatalf("clientcheck: publish %d: %v", i, err)
+		}
+	}
+	log.Printf("published %d messages to %s; waiting for decrypted echoes...", n, roomID)
+
+	deadline := time.Now().Add(15 * time.Second)
+	for time.Now().Before(deadline) {
+		mu.Lock()
+		have := len(got)
+		mu.Unlock()
+		if have >= n {
+			break
+		}
+		time.Sleep(100 * time.Millisecond)
+	}
+
+	mu.Lock()
+	defer mu.Unlock()
+	missing := 0
+	for _, w := range want {
+		if !got[w] {
+			missing++
+			log.Printf("  MISSING: %q", w)
+		}
+	}
+	log.Printf("connected node at finish: %s", c.ConnectedServer())
+	if missing > 0 {
+		log.Fatalf("GOLDEN FAIL: %d/%d messages not received decrypted", missing, n)
+	}
+	log.Printf("GOLDEN OK: all %d messages received and decrypted end-to-end", n)
+}
+
+// runLoop publishes a numbered message every interval for the duration and logs
+// the count received plus the node currently attached, so an operator stopping a
+// cluster node mid-run sees the client fail over to a survivor and keep receiving
+// (quorum 2/3). It is the live failover-with-a-connected-client test the 0011
+// chaos run never performed.
+func runLoop(c *client.Client, roomID string, duration, interval time.Duration) {
+	var mu sync.Mutex
+	received := 0
+	servers := map[string]int{} // node -> #ticks observed attached
+	sub, err := c.Subscribe(roomID, func(_ frame.Frame, _ []byte) {
+		mu.Lock()
+		received++
+		mu.Unlock()
+	})
+	if err != nil {
+		log.Fatalf("clientcheck: subscribe: %v", err)
+	}
+	defer sub.Unsubscribe()
+	time.Sleep(300 * time.Millisecond)
+
+	log.Printf("loop: publishing every %s for %s — stop a node now to test failover", interval, duration)
+	end := time.Now().Add(duration)
+	sent := 0
+	for time.Now().Before(end) {
+		msg := fmt.Sprintf("gapcheck-loop-%d", sent)
+		err := c.Publish(roomID, []byte(msg))
+		sent++
+		mu.Lock()
+		recv := received
+		mu.Unlock()
+		node := c.ConnectedServer()
+		up := c.IsConnected()
+		if node != "" {
+			mu.Lock()
+			servers[node]++
+			mu.Unlock()
+		}
+		pubStatus := "ok"
+		if err != nil {
+			pubStatus = "ERR:" + err.Error()
+		}
+		log.Printf("  t=%2ds sent=%d recv=%d up=%v node=%s publish=%s",
+			sent, sent, recv, up, node, pubStatus)
+		time.Sleep(interval)
+	}
+
+	mu.Lock()
+	defer mu.Unlock()
+	log.Printf("loop done: sent=%d received=%d", sent, received)
+	nodes := make([]string, 0, len(servers))
+	for n := range servers {
+		nodes = append(nodes, n)
+	}
+	sort.Strings(nodes)
+	for _, n := range nodes {
+		log.Printf("  attached to %s for %d ticks", n, servers[n])
+	}
+	if len(servers) > 1 {
+		log.Printf("FAILOVER OBSERVED: client was attached to %d distinct nodes across the run", len(servers))
+	}
+	if received == 0 {
+		log.Fatalf("LOOP FAIL: received 0 messages")
+	}
+	log.Printf("LOOP OK: client kept receiving across the run (received=%d)", received)
+}
+
+func splitCSV(s string) []string {
+	var out []string
+	for _, p := range strings.Split(s, ",") {
+		if p = strings.TrimSpace(p); p != "" {
+			out = append(out, p)
+		}
+	}
+	return out
+}

cmd/membershipd/kv_useradd_test.go

@@ -0,0 +1,152 @@
+package main
+
+// Integration tests for issue 0011 GAP A: `membershipd user add --store kv`
+// adds users to a RUNNING cluster's replicated allowlist via the privileged
+// internal connection, instead of the stop-seed-restart procedure the 0011
+// deploy required. These exercise the real connectKVStore path (load the
+// persisted internal identity from a file, present its nkey, open the KV store,
+// write the user) against an embedded enforce node, plus the idempotency and
+// error semantics the DoD calls for. Multi-node replication and node-down quorum
+// are validated against the live cluster (report 0012).
+
+import (
+	"encoding/hex"
+	"errors"
+	"path/filepath"
+	"testing"
+	"time"
+
+	cs "fn-registry/functions/cybersecurity"
+
+	"github.com/enmanuel/unibus/pkg/busauth"
+	"github.com/enmanuel/unibus/pkg/client"
+	"github.com/enmanuel/unibus/pkg/embeddednats"
+	"github.com/enmanuel/unibus/pkg/membership"
+)
+
+// startEnforceKVNode boots a single embedded enforce node whose authenticator
+// recognizes internalPubHex as the privileged internal identity, bootstraps the
+// KV control-plane store over the in-process internal connection, and publishes
+// it into the holder — the exact sequence main.go performs for --store kv. It
+// returns the client URL the CLI connects to.
+func startEnforceKVNode(t *testing.T, internalID cs.Identity) string {
+	t.Helper()
+	holder := &storeHolder{}
+	auth := busauth.NewNkeyAuthenticatorACLInternal(
+		holder.IsAuthorized,
+		busauth.PermissionsFromSubjects(holder.subjectACL),
+		hex.EncodeToString(internalID.SignPub),
+	)
+	ns, err := embeddednats.StartServer(embeddednats.ServerConfig{
+		StoreDir: t.TempDir(), Host: "127.0.0.1", Port: freePort(t), Auth: auth,
+	})
+	if err != nil {
+		t.Fatalf("start enforce node: %v", err)
+	}
+	t.Cleanup(func() { ns.Shutdown(); ns.WaitForShutdown() })
+
+	intNC, js, err := connectInternalJS(ns, internalID, true)
+	if err != nil {
+		t.Fatalf("bootstrap internal connection: %v", err)
+	}
+	t.Cleanup(intNC.Close)
+	kvStore, err := membership.OpenJetStream(js, membership.JetStreamConfig{Replicas: 1, OpTimeout: 3 * time.Second})
+	if err != nil {
+		t.Fatalf("bootstrap KV store: %v", err)
+	}
+	holder.set(kvStore)
+	return ns.ClientURL()
+}
+
+// TestUserAddStoreKV_GoldenAndIdempotent is the GAP A golden + edge-1: the CLI
+// connection (real connectKVStore, loading the internal identity from a file and
+// presenting its nkey) writes a user into the live KV allowlist, the user is
+// authorized afterward, and re-adding the same key is an explicit ErrUserExists
+// with no corruption (the unchanged row is still authorized).
+func TestUserAddStoreKV_GoldenAndIdempotent(t *testing.T) {
+	idFile := filepath.Join(t.TempDir(), "internal.id")
+	internalID, err := client.LoadOrCreateIdentity(idFile) // persists 0600
+	if err != nil {
+		t.Fatalf("persist internal identity: %v", err)
+	}
+	url := startEnforceKVNode(t, internalID)
+
+	// Golden: connect as the privileged internal identity (loopback, no TLS) and
+	// add a new user, exactly as `user add --store kv` does.
+	kv, err := connectKVStore(url, idFile, "", 1)
+	if err != nil {
+		t.Fatalf("connectKVStore (privileged): %v", err)
+	}
+	defer kv.Close()
+
+	newUser, err := cs.GenerateIdentity()
+	if err != nil {
+		t.Fatalf("new user identity: %v", err)
+	}
+	pub := hex.EncodeToString(newUser.SignPub)
+	if err := kv.store.AddUser(pub, "gapcheck_user", membership.RoleMember); err != nil {
+		t.Fatalf("add user to live KV: %v", err)
+	}
+	if !kv.store.IsAuthorized(pub) {
+		t.Fatalf("user added to KV must be authorized")
+	}
+
+	// Edge 1: re-adding the same key is a clean, non-destructive ErrUserExists.
+	err = kv.store.AddUser(pub, "gapcheck_user", membership.RoleMember)
+	if !errors.Is(err, membership.ErrUserExists) {
+		t.Fatalf("re-add must return ErrUserExists (idempotent), got %v", err)
+	}
+	// A different handle/role with the SAME key is also rejected — the row is not
+	// silently overwritten (no role flip).
+	if err := kv.store.AddUser(pub, "impostor", membership.RoleAdmin); !errors.Is(err, membership.ErrUserExists) {
+		t.Fatalf("re-add with a different role must NOT overwrite; want ErrUserExists, got %v", err)
+	}
+	u, err := kv.store.GetUser(pub)
+	if err != nil {
+		t.Fatalf("get user: %v", err)
+	}
+	if u.Handle != "gapcheck_user" || u.Role != membership.RoleMember || u.Status != membership.StatusActive {
+		t.Fatalf("idempotent re-add corrupted the row: %+v", u)
+	}
+}
+
+// TestUserAddStoreKV_RequiresInternalIdentity: --store kv without a usable
+// internal identity file fails loudly (missing file, empty path) rather than
+// silently connecting unprivileged.
+func TestUserAddStoreKV_RequiresInternalIdentity(t *testing.T) {
+	if _, err := connectKVStore("nats://127.0.0.1:4250", "", "", 1); err == nil {
+		t.Fatalf("empty --internal-id-file must be an error")
+	}
+	missing := filepath.Join(t.TempDir(), "nope.id")
+	if _, err := connectKVStore("nats://127.0.0.1:4250", missing, "", 1); err == nil {
+		t.Fatalf("missing internal identity file must be an error")
+	}
+}
+
+// TestUserAddStoreKV_UnreachableKV is the GAP A error case: pointing --store kv
+// at a dead endpoint yields a clear, handled error (no crash, no silent success).
+func TestUserAddStoreKV_UnreachableKV(t *testing.T) {
+	idFile := filepath.Join(t.TempDir(), "internal.id")
+	if _, err := client.LoadOrCreateIdentity(idFile); err != nil {
+		t.Fatalf("persist internal identity: %v", err)
+	}
+	// A loopback port with nothing listening: connect must fail fast and wrapped.
+	_, err := connectKVStore("nats://127.0.0.1:1/", idFile, "", 1)
+	if err == nil {
+		t.Fatalf("connecting to a dead endpoint must error")
+	}
+}
+
+// TestUserAddStoreKV_RemoteWithoutCARefused: a non-loopback target without --ca
+// is refused so the allowlist write never travels in cleartext (audit 0008 N6,
+// same guard as migrate-to-kv).
+func TestUserAddStoreKV_RemoteWithoutCARefused(t *testing.T) {
+	idFile := filepath.Join(t.TempDir(), "internal.id")
+	if _, err := client.LoadOrCreateIdentity(idFile); err != nil {
+		t.Fatalf("persist internal identity: %v", err)
+	}
+	_, err := connectKVStore("nats://203.0.113.1:4250", idFile, "", 1)
+	if err == nil {
+		t.Fatalf("remote target without --ca must be refused")
+	}
+}

cmd/membershipd/main.go

@@ -24,6 +24,7 @@ import (
 
 	"github.com/enmanuel/unibus/pkg/blobstore"
 	"github.com/enmanuel/unibus/pkg/busauth"
+	"github.com/enmanuel/unibus/pkg/client"
 	"github.com/enmanuel/unibus/pkg/embeddednats"
 	"github.com/enmanuel/unibus/pkg/membership"
 )
@@ -83,6 +84,17 @@ func main() {
 		// "kv" puts rooms/members/keys/users in replicated JetStream KV so any node
 		// in the cluster serves the same state.
 		storeBackend = flag.String("store", "sqlite", "control-plane store backend: sqlite (default, single-node) | kv (replicated JetStream, decentralized)")
+		// Persisted internal service identity (issue 0011 gaps, GAP A): when set, the
+		// privileged internal identity used to manage JetStream is LOADED from this
+		// file (generated and persisted on first start) instead of being a fresh
+		// ephemeral key each boot. Persisting it is what lets `membershipd user add
+		// --store kv` write the replicated allowlist of a LIVE cluster: that CLI,
+		// run over loopback on a node, loads the SAME identity and presents the nkey
+		// this node's authenticator already grants full permissions. Empty keeps the
+		// ephemeral-per-process behavior (single-node/dev default, unchanged). The
+		// file holds a private key: it is written 0600 and belongs next to the node's
+		// TLS keys (deploy keeps it under secrets/, gitignored).
+		internalIDFile = flag.String("internal-id-file", "", "path to a persisted internal service identity (JSON); enables `membershipd user add --store kv` against the live cluster. Empty = ephemeral per-process identity (dev default)")
 	)
 	flag.Parse()
 
@@ -136,9 +148,21 @@ func main() {
 	var internalID cs.Identity
 	var internalPubHex string
 	if needJS && enforce && *natsURL == "" {
-		internalID, err = cs.GenerateIdentity()
-		if err != nil {
-			log.Fatalf("generate internal identity: %v", err)
+		if *internalIDFile != "" {
+			// Persisted identity: load it, generating + writing it (0600) on first
+			// start. A stable internal key is what `user add --store kv` presents to
+			// add users to a live cluster (GAP A); rotate it by deleting the file and
+			// restarting.
+			internalID, err = client.LoadOrCreateIdentity(*internalIDFile)
+			if err != nil {
+				log.Fatalf("load internal service identity %q: %v", *internalIDFile, err)
+			}
+			log.Printf("internal service identity: persisted (%s)", *internalIDFile)
+		} else {
+			internalID, err = cs.GenerateIdentity()
+			if err != nil {
+				log.Fatalf("generate internal identity: %v", err)
+			}
 		}
 		internalPubHex = hex.EncodeToString(internalID.SignPub)
 	}

cmd/membershipd/users_cli.go

@@ -1,7 +1,7 @@
 package main
 
 import (
-	"encoding/hex"
+	"errors"
 	"flag"
 	"fmt"
 	"os"
@@ -50,13 +50,26 @@ commands:
   list     List all registered users
   revoke   Revoke a user (denies access on both planes immediately)
 
+store backends (--store):
+  sqlite   local SQLite database (default; seeds the first admin offline)
+  kv       the RUNNING cluster's replicated JetStream KV allowlist, via the
+           privileged internal connection — add users with the cluster live,
+           no stop-seed-restart needed (run over loopback/SSH on a node)
+
 examples:
   membershipd user add --handle alice --sign-pub <64-hex> --role admin
-  membershipd user list
+  membershipd user add --store kv --handle bob --sign-pub <64-hex> --role member
+  membershipd user list --store kv
   membershipd user revoke <64-hex>
 
 common flags:
-  --db <path>   SQLite database path (default ./local_files/unibus.db)
+  --db <path>   SQLite database path (--store sqlite; default ./local_files/unibus.db)
+
+--store kv flags (defaults assume an on-node invocation):
+  --nats-url <url>           cluster NATS (default nats://127.0.0.1:4250)
+  --internal-id-file <path>  persisted internal service identity (default /opt/unibus/secrets/internal.id)
+  --ca <path>                CA cert pinning the data-plane TLS (default /opt/unibus/tls/ca.crt)
+  --kv-replicas <n>          KV replication factor, match the cluster (default 3)
 `)
 }
 
@@ -76,16 +89,56 @@ func openStore(path string) membership.Store {
 
 // validateSignPubHex ensures the key is exactly a 32-byte Ed25519 public key in
 // hex (64 hex chars). Catching this here turns a silent "authorized nobody" into
-// an explicit error at seed time.
+// an explicit error at seed time. It delegates to membership.ValidateSignPubHex
+// so the CLI and the HTTP user-management handlers share one rule.
 func validateSignPubHex(signPub string) error {
-	b, err := hex.DecodeString(signPub)
-	if err != nil {
-		return fmt.Errorf("sign-pub is not valid hex: %w", err)
+	return membership.ValidateSignPubHex(signPub)
+}
+
+// kvFlags holds the connection flags shared by the --store kv path of the user
+// subcommands. registerKVFlags wires them onto a flag set so add and list expose
+// an identical interface.
+type kvFlags struct {
+	store      *string
+	natsURL    *string
+	internalID *string
+	ca         *string
+	replicas   *int
+}
+
+func registerKVFlags(fs *flag.FlagSet) kvFlags {
+	return kvFlags{
+		store:      fs.String("store", "sqlite", "user store backend: sqlite (local DB) | kv (the live cluster's replicated allowlist)"),
+		natsURL:    fs.String("nats-url", defaultClusterNatsURL, "cluster NATS url for --store kv"),
+		internalID: fs.String("internal-id-file", defaultInternalIDFile, "persisted internal service identity for --store kv"),
+		ca:         fs.String("ca", defaultClusterCAFile, "CA cert pinning TLS on the --store kv NATS connection"),
+		replicas:   fs.Int("kv-replicas", 3, "KV replication factor for --store kv (match the cluster)"),
 	}
-	if len(b) != 32 {
-		return fmt.Errorf("sign-pub must be a 32-byte Ed25519 public key (64 hex chars), got %d bytes", len(b))
+}
+
+// resolveStore returns the membership store for the chosen backend plus a cleanup
+// func. For --store kv it opens the privileged connection to the live cluster; for
+// sqlite it opens the local file. It exits the process with a clear message on any
+// failure (a dead NATS, a missing identity file), so a broken --store kv add fails
+// loudly instead of silently — Error case of the GAP A DoD. The returned *kvConn
+// is non-nil only for the kv backend (so the caller can report replication).
+func resolveStore(cmd string, kf kvFlags, dbPath string) (membership.Store, *kvConn, func()) {
+	switch *kf.store {
+	case "sqlite":
+		store := openStore(dbPath)
+		return store, nil, func() { store.Close() }
+	case "kv":
+		kv, err := connectKVStore(*kf.natsURL, *kf.internalID, *kf.ca, *kf.replicas)
+		if err != nil {
+			fmt.Fprintf(os.Stderr, "membershipd %s: --store kv: %v\n", cmd, err)
+			os.Exit(1)
+		}
+		return kv.store, kv, kv.Close
+	default:
+		fmt.Fprintf(os.Stderr, "membershipd %s: --store must be \"sqlite\" or \"kv\", got %q\n", cmd, *kf.store)
+		os.Exit(2)
+		return nil, nil, func() {}
 	}
-	return nil
 }
 
 func userAdd(args []string) {
@@ -94,6 +147,7 @@ func userAdd(args []string) {
 	signPub := fs.String("sign-pub", "", "Ed25519 signing public key in hex (required)")
 	role := fs.String("role", membership.RoleMember, "role: admin or member")
 	dbPath := fs.String("db", defaultDBPath, "SQLite database path")
+	kf := registerKVFlags(fs)
 	_ = fs.Parse(args)
 
 	if *handle == "" || *signPub == "" {
@@ -105,23 +159,35 @@ func userAdd(args []string) {
 		os.Exit(2)
 	}
 
-	store := openStore(*dbPath)
-	defer store.Close()
+	store, kv, closeStore := resolveStore("user add", kf, *dbPath)
+	defer closeStore()
 
 	if err := store.AddUser(*signPub, *handle, *role); err != nil {
+		if errors.Is(err, membership.ErrUserExists) {
+			// Idempotency contract (GAP A): re-adding the same key is an EXPLICIT,
+			// non-destructive error — the existing row is left untouched (no silent
+			// upsert that could flip a role or clobber status, which would corrupt the
+			// allowlist). To replace a user, `user revoke <key>` then add again.
+			fmt.Fprintf(os.Stderr, "membershipd user add: user %s already registered (unchanged); revoke it first to replace\n", *signPub)
+			os.Exit(1)
+		}
 		fmt.Fprintf(os.Stderr, "membershipd user add: %v\n", err)
 		os.Exit(1)
 	}
 	fmt.Printf("added user %q (%s) role=%s\n", *handle, *signPub, *role)
+	if kv != nil {
+		reportKVReplication(kv.js)
+	}
 }
 
 func userList(args []string) {
 	fs := flag.NewFlagSet("user list", flag.ExitOnError)
 	dbPath := fs.String("db", defaultDBPath, "SQLite database path")
+	kf := registerKVFlags(fs)
 	_ = fs.Parse(args)
 
-	store := openStore(*dbPath)
-	defer store.Close()
+	store, _, closeStore := resolveStore("user list", kf, *dbPath)
+	defer closeStore()
 
 	users, err := store.ListUsers()
 	if err != nil {
@@ -143,6 +209,7 @@ func userList(args []string) {
 func userRevoke(args []string) {
 	fs := flag.NewFlagSet("user revoke", flag.ExitOnError)
 	dbPath := fs.String("db", defaultDBPath, "SQLite database path")
+	kf := registerKVFlags(fs)
 
 	// Go's flag package stops at the first non-flag argument, so `revoke <key>
 	// --db path` would otherwise leave --db unparsed. Pull a leading positional
@@ -167,8 +234,8 @@ func userRevoke(args []string) {
 		os.Exit(2)
 	}
 
-	store := openStore(*dbPath)
-	defer store.Close()
+	store, _, closeStore := resolveStore("user revoke", kf, *dbPath)
+	defer closeStore()
 
 	if err := store.RevokeUser(signPub); err != nil {
 		fmt.Fprintf(os.Stderr, "membershipd user revoke: %v\n", err)

cmd/membershipd/users_kv.go

@@ -0,0 +1,151 @@
+package main
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"time"
+
+	"github.com/enmanuel/unibus/pkg/busauth"
+	"github.com/enmanuel/unibus/pkg/client"
+	"github.com/enmanuel/unibus/pkg/membership"
+	"github.com/nats-io/nats.go"
+	"github.com/nats-io/nats.go/jetstream"
+)
+
+// users_kv.go is the `--store kv` half of the user administration CLI (issue 0011
+// gaps, GAP A): adding and listing bus users directly against the RUNNING
+// cluster's replicated JetStream KV allowlist, with no need to stop the cluster,
+// seed a standalone node, and restart (the procedure the 0011 deploy required).
+//
+// The mechanism is the cluster's own privileged internal connection. Under
+// enforce every bus user is confined by the per-subject ACL to the JetStream API
+// of its own rooms, so no ordinary identity may touch the control-plane buckets
+// (KV_UNIBUS_*). The ONLY identity the authenticator grants full JetStream
+// permissions is membershipd's internal service identity. By persisting that
+// identity to a file (membershipd --internal-id-file) the same key becomes
+// available to this CLI, which presents it as its NATS nkey and is therefore
+// recognized as the privileged internal client and allowed to read/write the KV.
+//
+// Intended invocation is over loopback on a cluster node (SSH): the data-plane
+// TLS certificate's SAN covers 127.0.0.1/localhost and the internal identity file
+// lives 0600 next to the node's TLS keys. Using the file requires root on the
+// node, which already implies full control of that node — so co-locating it adds
+// no practical exposure beyond what the TLS server key and cluster password
+// already represent.
+
+// defaultClusterNatsURL is the node-local NATS listener. The CLI is meant to run
+// on a cluster node over SSH, talking to that node's own embedded server.
+const defaultClusterNatsURL = "nats://127.0.0.1:4250"
+
+// Deploy-default paths for the privileged identity and the data-plane CA, so an
+// on-node invocation needs only --handle/--sign-pub/--role. Override for other
+// layouts.
+const (
+	defaultInternalIDFile = "/opt/unibus/secrets/internal.id"
+	defaultClusterCAFile  = "/opt/unibus/tls/ca.crt"
+)
+
+// kvConn bundles the privileged NATS connection to a live cluster and the
+// KV-backed control-plane store opened over it. Close releases both.
+type kvConn struct {
+	nc    *nats.Conn
+	js    jetstream.JetStream
+	store membership.Store
+}
+
+func (k *kvConn) Close() {
+	if k == nil {
+		return
+	}
+	if k.store != nil {
+		_ = k.store.Close()
+	}
+	if k.nc != nil {
+		k.nc.Close()
+	}
+}
+
+// connectKVStore opens the privileged internal connection to the cluster's NATS
+// and the JetStream KV control-plane store on top of it. internalIDFile is the
+// membershipd-persisted internal service identity whose nkey the authenticator
+// grants full permissions; caPath pins the data-plane TLS (empty only for a
+// non-TLS dev cluster). A non-loopback target without --ca is refused, mirroring
+// migrate-to-kv (audit 0008 N6): the allowlist write must not travel in cleartext.
+func connectKVStore(natsURL, internalIDFile, caPath string, replicas int) (*kvConn, error) {
+	if internalIDFile == "" {
+		return nil, fmt.Errorf("--internal-id-file is required for --store kv (the privileged identity membershipd persists with --internal-id-file)")
+	}
+	// Confidentiality guard: a remote NATS without TLS would expose the allowlist
+	// (handles/roles/sign-pubs) and the privileged nkey handshake in cleartext.
+	if !isLoopbackURL(natsURL) && caPath == "" {
+		return nil, fmt.Errorf("refusing to connect to remote %q without --ca: the allowlist write would travel in cleartext — pin TLS with --ca, or run over a loopback --nats-url on a node", natsURL)
+	}
+
+	id, err := client.LoadIdentity(internalIDFile)
+	if err != nil {
+		return nil, fmt.Errorf("load internal identity: %w", err)
+	}
+	nkeyPub, nkeySign, err := busauth.ClientNkey(id.SignPriv)
+	if err != nil {
+		return nil, fmt.Errorf("derive nkey from internal identity: %w", err)
+	}
+	opts := []nats.Option{
+		nats.Name("membershipd-user-cli"),
+		nats.Nkey(nkeyPub, nkeySign),
+	}
+	if caPath != "" {
+		tlsCfg, err := busauth.LoadCATLSConfig(caPath)
+		if err != nil {
+			return nil, fmt.Errorf("load CA %q: %w", caPath, err)
+		}
+		opts = append(opts, nats.Secure(tlsCfg))
+	}
+	nc, err := nats.Connect(natsURL, opts...)
+	if err != nil {
+		return nil, fmt.Errorf("connect cluster NATS %q: %w", natsURL, err)
+	}
+	js, err := jetstream.New(nc)
+	if err != nil {
+		nc.Close()
+		return nil, fmt.Errorf("jetstream: %w", err)
+	}
+	store, err := membership.OpenJetStream(js, membership.JetStreamConfig{Replicas: replicas})
+	if err != nil {
+		nc.Close()
+		return nil, fmt.Errorf("open KV control-plane store: %w", err)
+	}
+	return &kvConn{nc: nc, js: js, store: store}, nil
+}
+
+// reportKVReplication prints the replication status of the allowlist bucket
+// stream (KV_UNIBUS_users) right after a write, so the operator sees the add
+// landed on a quorum and replicated to the followers — executable evidence that
+// the live-cluster add is HA, not single-node. Best-effort: a read failure is a
+// note, not an error (the write itself already succeeded).
+func reportKVReplication(js jetstream.JetStream) {
+	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+	defer cancel()
+	st, err := js.Stream(ctx, "KV_UNIBUS_users")
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "note: could not read KV_UNIBUS_users stream info: %v\n", err)
+		return
+	}
+	info, err := st.Info(ctx)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "note: could not read KV_UNIBUS_users stream info: %v\n", err)
+		return
+	}
+	if info.Cluster == nil {
+		fmt.Printf("KV_UNIBUS_users: standalone (R1, no cluster replication); msgs=%d\n", info.State.Msgs)
+		return
+	}
+	current := 0
+	for _, r := range info.Cluster.Replicas {
+		if r.Current {
+			current++
+		}
+	}
+	fmt.Printf("KV_UNIBUS_users: leader=%s followers_current=%d/%d msgs=%d\n",
+		info.Cluster.Leader, current, len(info.Cluster.Replicas), info.State.Msgs)
+}

deploy/cluster/README.md

@@ -5,9 +5,12 @@ This directory holds the material to bring up unibus as a **3-node cluster**
 plane (rooms/members/keys/users on JetStream KV + the anti-replay nonce bucket)
 survives the loss of any one node (quorum 2/3).
 
-> **The agent that authored this never touched a VPS.** Every step that changes a
-> remote host is marked **HUMAN** and is executed by the operator. `deploy-cluster.sh`
-> defaults to a dry run.
+> **Status: this cluster is DEPLOYED in production** (magnus + homer + datardos,
+> R3, enforce+ACL+TLS) — see report 0011. The runbook below was authored before any
+> VPS existed and has since been **corrected against the real deploy** (report 0012):
+> the start ordering, the R1→R3 reality, and the live user-add path were all wrong
+> or missing. Steps that change a remote host are marked **HUMAN**; `deploy-cluster.sh`
+> still defaults to a dry run.
 
 ## Files
 
@@ -22,18 +25,22 @@ Generated keys/secrets (`out/`, `build/`, `secrets/`) are **gitignored** — the
 secret and never leave the operator's trusted machine except over the secure
 rsync channel.
 
-## Topology
+## Topology (as deployed, report 0011)
 
-| Node | SSH | Public IP | WireGuard IP | Role |
-|---|---|---|---|---|
-| magnus | `magnus` | `<MAGNUS_PUBLIC_IP>` | `<MAGNUS_WG_IP>` | seed (first up) |
-| homer | `homer` | `141.94.69.66` | `<HOMER_WG_IP>` | replica |
-| datardos | `dd` | `51.91.100.142` | `<DATARDOS_WG_IP>` (10.21.0.x) | replica |
+| Node | SSH | Public IP | Role |
+|---|---|---|---|
+| magnus | `magnus` (root) | `135.125.201.30` | node — **= organic-machine.com = `om`**, the critical host (caddy + gitea + registry-api + monitoring); the bus runs alongside, untouched |
+| homer | `homer` (ubuntu+sudo) | `141.94.69.66` | node |
+| datardos | `dd` (ubuntu+sudo) | `51.91.100.142` | node |
 
-The route layer (server-to-server) prefers the **WireGuard mesh**
-(`ROUTE_NETWORK=wg`); the client data plane and the HTTP control plane are reached
-over the public IPs. The route CA is **separate** from the client CA, so a client
-cert can never be presented to the route port.
+`ROUTE_NETWORK=public`, **not `wg`**: there is no WireGuard mesh between the three
+nodes (homer and datardos do not even have the `wg` binary; om's only WG peers are
+the operator's PCs). The server-to-server routes therefore travel over the public
+IPs, protected by the **separate cluster route CA** (mutual route TLS) — a client
+data-plane cert can never be presented to the route port. The client data plane and
+the HTTP control plane are also reached over the public IPs. There is no fixed
+"seed" node: with R3 the three are peers (see "Bring up" for why a lone node cannot
+self-serve).
 
 ## Prerequisites (HUMAN, once)
 
@@ -93,25 +100,48 @@ SEED
 
 > The KV written here lives in `./local_files/jetstream`, which the cluster unit
 > reuses (`--nats-store` default), so the admin is present when the enforce cluster
-> starts. Additional users are added the same loopback way until a
-> `user add --store kv` exists (see GAP in report 0009).
+> starts. This loopback bootstrap is needed ONLY for the very first admin (the
+> chicken-and-egg). **Every user after that is added with the cluster live** — no
+> stop-seed-restart — via `user add --store kv` (see "Add users to the live
+> cluster" below, report 0012).
 
-## Bring up (HUMAN — staggered)
+## Bring up (HUMAN)
 
-Bring up the seed first, then the replicas one at a time, checking each joins.
+> **CORRECTION (report 0012).** The original instruction — "start magnus alone and
+> verify healthz, then add the others" — is **WRONG and will look like a hung
+> deploy.** A 3-node JetStream cluster forms a RAFT meta-group that needs a quorum
+> (2 of 3) to elect a leader. A single started node has no quorum, so its JetStream
+> meta never becomes current: `--store kv` blocks creating the KV buckets and
+> **`/healthz` never returns ok** until a second node joins. Waiting for magnus to
+> "go green" before starting the others therefore deadlocks the rollout.
+
+Start the nodes so a quorum forms. On a **clean cluster** the simplest correct
+procedure is to start all three close together and let the meta-group converge:
 
 ```bash
-# 1. Seed node (after the seed step above).
-ssh root@magnus 'systemctl enable --now membershipd-cluster'
-ssh root@magnus 'curl -fsS https://127.0.0.1:8470/healthz --cacert /opt/unibus/tls/ca.crt'
+# Start all three (order does not matter); each blocks on the others until a
+# 2/3 quorum elects a JetStream meta leader, then the KV buckets are created.
+for h in magnus homer datardos; do ssh "$h" 'sudo systemctl enable --now membershipd-cluster'; done
 
-# 2. Replicas, one at a time.
-ssh root@homer    'systemctl enable --now membershipd-cluster'
-ssh root@datardos 'systemctl enable --now membershipd-cluster'
+# Only NOW does healthz return ok — once the meta-group has a leader (give it
+# ~10-30s on a cold start). Poll, do not assume the first node is broken.
+for h in magnus homer datardos; do
+  echo "== $h =="; ssh "$h" 'curl -fsS https://127.0.0.1:8470/healthz --cacert /opt/unibus/tls/ca.crt || echo "(not ready yet — needs quorum)"'
+done
 ```
 
-> Initial rollout runs at **R1** (`KV_REPLICAS=1` in `nodes.env`): the buckets live
-> on the seed only. This is NOT HA yet — see "Scale to R3".
+A **staggered** start also works, but only because `membershipd`'s KV open RETRIES
+the bucket creation for a 120s bootstrap budget (issue 0006g, fix #3): the first
+node sits in that retry loop — NOT serving healthz — until the second node makes a
+quorum, then both converge and the third catches up. Either way, a lone node never
+self-serves; do not gate the next node's start on the previous one's healthz.
+
+> A cold multi-node start only converges because of **three cold-start fixes**
+> (report 0011): route pooling off (`PoolSize=-1`), `NoAdvertise=true` (Docker
+> bridge IPs not gossiped), and the KV-open retry loop above. Without them the
+> meta-group re-elects leaders forever and bucket creation hangs. If a fresh
+> cluster will not form, confirm the running binary contains these fixes before
+> touching config.
 
 ## Promote an existing single-node (SQLite) deployment (HUMAN, optional)
 
@@ -137,11 +167,80 @@ ssh root@magnus 'nats --server nats://127.0.0.1:4250 server list'   # 3 servers,
 
 A healthy cluster shows 3 routed servers and a JetStream meta-group with a leader.
 
-## Scale to R3 (HUMAN — real HA)
+## Add users to the live cluster (HUMAN — `user add --store kv`)
 
-Once all three nodes are up and routed, raise the replication factor of every
-control-plane stream from 1 to 3 IN PLACE (no data loss), then flip `KV_REPLICAS=3`
-in `nodes.env` so future (re)deploys keep it:
+With the cluster up, add (and revoke) bus users **without stopping anything**,
+directly against the replicated KV allowlist. This replaces the stop-seed-restart
+procedure the original runbook implied for every user beyond the first admin.
+
+The mechanism is the cluster's own **privileged internal connection**: under
+`enforce` every bus user is confined by the per-subject ACL to its own rooms, so no
+ordinary identity may write the control-plane buckets. The only identity the
+authenticator grants full JetStream permissions is `membershipd`'s internal service
+identity. The unit persists that identity to `${INTERNAL_ID_FILE}`
+(`/opt/unibus/secrets/internal.id`, 0600) via `--internal-id-file`, so the same key
+is available to the CLI. Run the CLI **on a node, over loopback** (the data-plane
+TLS cert SAN covers `127.0.0.1`); reading the identity file requires root on that
+node, which already implies full control of it, so this adds no practical exposure.
+
+```bash
+# Add a member to the live cluster's replicated allowlist (run on any node).
+ssh root@magnus 'sudo /opt/unibus/membershipd user add --store kv \
+  --handle alice --role member --sign-pub <64-hex-ed25519-pub>'
+#   -> added user "alice" (...) role=member
+#   -> KV_UNIBUS_users: leader=<node> followers_current=2/2 msgs=N   (replicated, HA)
+
+# List / revoke against the same live KV:
+ssh root@magnus 'sudo /opt/unibus/membershipd user list   --store kv'
+ssh root@magnus 'sudo /opt/unibus/membershipd user revoke --store kv <64-hex-ed25519-pub>'
+```
+
+Defaults assume an on-node invocation (`--nats-url nats://127.0.0.1:4250`,
+`--internal-id-file /opt/unibus/secrets/internal.id`, `--ca /opt/unibus/tls/ca.crt`,
+`--kv-replicas 3`). Semantics:
+
+- **Idempotent / non-destructive**: re-adding the same key is an explicit
+  `already registered` error (exit 1), never a silent overwrite — a re-add cannot
+  flip a member to admin. To replace a user, `revoke` then add.
+- **HA**: the write commits through the JetStream quorum, so it succeeds even with
+  one node down (2/3); the printed `followers_current` shows replication.
+- **No hard delete**: `revoke` flips status to `revoked` (denied on both planes,
+  auditable); the KV has no row deletion, matching the SQLite store.
+
+> **Rollout note (report 0012):** the live verification deployed this binary +
+> `--internal-id-file` to **datardos only** (the non-critical node). magnus and
+> homer still run the 0011 binary. To make the capability available (and the unit)
+> on all three — recommended, the posture is identical so there is no urgency — roll
+> the new binary with backups, one node at a time, verifying healthz between each:
+> ```bash
+> for h in homer magnus; do
+>   ssh "$h" 'sudo cp -a /opt/unibus/membershipd /opt/unibus/membershipd.bak'   # backup
+>   scp build/membershipd "$h:/tmp/m" && ssh "$h" 'sudo install -o ubuntu -g ubuntu -m0775 /tmp/m /opt/unibus/membershipd'
+>   # add INTERNAL_ID_FILE=/opt/unibus/secrets/internal.id to /opt/unibus/cluster.env
+>   # add `--internal-id-file ${INTERNAL_ID_FILE} \` to the unit before `--store kv`
+>   ssh "$h" 'sudo systemctl daemon-reload && sudo systemctl restart membershipd-cluster'
+>   ssh "$h" 'curl -fsS https://127.0.0.1:8470/healthz --cacert /opt/unibus/tls/ca.crt'  # green before next
+> done
+> ```
+> (`deploy-cluster.sh` + the unit template already emit `INTERNAL_ID_FILE` and the
+> flag, so a fresh `./deploy-cluster.sh --yes` is correct for all three.)
+
+## Replication: go straight to R3 (HUMAN — real HA)
+
+> **CORRECTION (report 0012).** The original "start at R1, then scale to R3" plan
+> assumed R1 is a usable interim state. **It is not, in this cluster.** At R1 all six
+> control-plane buckets (`KV_UNIBUS_users/rooms/members/room_keys/rooms_by_member`
+> + `KV_UNIBUS_nonces`) live on a SINGLE node — a hard **SPOF for authentication**:
+> if that node dies, the nonce/KV control plane is unreachable and EVERY
+> authenticated request fails closed (auth DoS). Worse, the cold multi-node start
+> only converges at all because of the three cold-start fixes (see "Bring up"); the
+> real deploy never ran a healthy R1 and **jumped straight to R3 once the cluster
+> formed.** Treat R1 as a transient artifact of bucket creation, not a milestone.
+
+The deployed config already sets `KV_REPLICAS=3` in `nodes.env`. If buckets were
+created at R1 (e.g. only one node was up when `--store kv` first opened them), raise
+every control-plane stream to R3 IN PLACE (no data loss) once all three nodes are
+routed:
 
 ```bash
 for s in KV_UNIBUS_users KV_UNIBUS_rooms KV_UNIBUS_members KV_UNIBUS_room_keys \
@@ -151,27 +250,32 @@ done
 # (also OBJ_UNIBUS_blobs if the object store is in use)
 ```
 
-Until this is done, R1 means the seed node is a **single point of failure for
-authentication**: if it dies, the nonce/KV control plane is unreachable and every
-authenticated request fails closed (auth DoS). R1 is a rollout step, not HA.
+After this each bucket shows `followers_current=2/2` (quorum 2/3). The
+`user add --store kv` command prints that figure for `KV_UNIBUS_users` on every add,
+which is a cheap live HA check.
 
-## Chaos test (HUMAN — requires the 3 live VPS; NOT run here)
+## Chaos test (HUMAN — requires the 3 live VPS)
 
 Validate quorum tolerance after R3:
 
 ```bash
-# Kill one node; the cluster keeps serving (quorum 2/3).
-ssh root@datardos 'systemctl stop membershipd-cluster'
+# Kill one node; the cluster keeps serving (quorum 2/3). On ubuntu nodes use sudo.
+ssh dd 'sudo systemctl stop membershipd-cluster'
 #   -> clients fail over (multiple seed URLs); reads/writes still succeed.
-ssh root@datardos 'systemctl start membershipd-cluster'   # rejoins, catches up
+ssh dd 'sudo systemctl start membershipd-cluster'   # rejoins, catches up
 
 # Kill two nodes; quorum is LOST — the control plane should fail CLOSED (deny),
 # never fail open. Verify a request is rejected, not silently served.
 ```
 
-This network-level chaos test (kill 1/3, kill 2/3, partition/split-brain) is part
-of the deploy validation (issue 0003f) and runs against the real VPS — it is
-deliberately out of scope for the authoring agent.
+> **Validated (report 0012).** The 0011 chaos run checked only the control plane
+> (healthz + meta/stream-leader failover + KV readable with 2/3). Report 0012 added
+> the missing data-plane proofs against the live cluster: a real authenticated
+> client (`cmd/clientcheck`, operator identity, nkey+TLS) creating an E2E room and
+> publishing/subscribing — including a node stopped mid-stream, where the client
+> failed over to a survivor and kept receiving with zero loss (quorum 2/3) — and
+> `user add --store kv` committing with one node (the KV leader) down. The kill-2/3
+> fail-closed case remains a documented manual step.
 
 ## Rollback
 

deploy/cluster/deploy-cluster.sh

@@ -97,6 +97,7 @@ TLS_KEY=${REMOTE_DIR}/tls/server-${name}.key
 ROUTE_TLS_CERT=${REMOTE_DIR}/tls/route-${name}.crt
 ROUTE_TLS_KEY=${REMOTE_DIR}/tls/route-${name}.key
 ROUTE_TLS_CA=${REMOTE_DIR}/tls/cluster-ca.crt
+INTERNAL_ID_FILE=${REMOTE_DIR}/secrets/internal.id
 EOF
 
   run ssh "$target" "mkdir -p ${REMOTE_DIR}/tls ${REMOTE_DIR}/secrets"
@@ -114,13 +115,16 @@ if [[ $APPLY -eq 0 ]]; then
 fi
 cat <<'NEXT'
 
-HUMAN — staggered start (do NOT enable all at once; see README "Bring up"):
-  1. Seed node first (e.g. magnus):
-       ssh root@magnus 'systemctl enable --now membershipd-cluster'
-       ssh root@magnus '/opt/unibus/membershipd user add --admin ...'   # seed admin
-  2. Then the other two, one at a time, checking quorum after each:
-       ssh root@homer    'systemctl enable --now membershipd-cluster'
-       ssh root@datardos 'systemctl enable --now membershipd-cluster'
+HUMAN — bring up (see README "Bring up" — a LONE node has no quorum and never
+serves healthz, so do NOT gate the next node on the previous one going green):
+  1. Seed the FIRST admin into the KV via the loopback bootstrap (README
+     "Seed the first admin"); this is needed only for the chicken-and-egg admin.
+  2. Start all three so a 2/3 quorum forms (order does not matter); healthz
+     turns ok only once the meta-group elects a leader (~10-30s cold):
+       for h in magnus homer datardos; do ssh "$h" 'sudo systemctl enable --now membershipd-cluster'; done
   3. Verify posture + quorum (README "Verify").
-  4. Scale replicas 1 -> 3 once all three are up (README "Scale to R3").
+  4. Ensure R3 on every control-plane stream (README "Replication: go straight to
+     R3"); R1 is a SPOF, not a milestone.
+  5. Add further users with the cluster LIVE — no restart — via
+     `membershipd user add --store kv` (README "Add users to the live cluster").
 NEXT

deploy/cluster/membershipd-cluster.service

@@ -33,6 +33,7 @@ ExecStart=/opt/unibus/membershipd \
   --route-tls-cert ${ROUTE_TLS_CERT} \
   --route-tls-key ${ROUTE_TLS_KEY} \
   --route-tls-ca ${ROUTE_TLS_CA} \
+  --internal-id-file ${INTERNAL_ID_FILE} \
   --store kv \
   --kv-replicas ${KV_REPLICAS}
 # Restart=always (NOT on-failure): a clean SIGTERM exits success, and on-failure

deploy/cluster/nodes.env

@@ -2,10 +2,10 @@
 #
 # This file is SOURCED by generate-cluster-certs.sh and deploy-cluster.sh.
 #
-# HUMAN: fill in every <PLACEHOLDER> with the real value before running the
+# HUMAN: fill in every placeholder with the real value before running the
 # scripts. The public IPs known at authoring time are pre-filled; the WireGuard
 # mesh IPs and magnus's public IP must be supplied. The scripts refuse to run
-# while any <PLACEHOLDER> remains.
+# while any unfilled placeholder remains.
 
 # Cluster identity (must be identical on every node).
 CLUSTER_NAME="unibus"
@@ -16,7 +16,7 @@ CLUSTER_USER="unibus-cluster"
 # KV/nonce replication factor. START AT 1 for the initial 1->3 rollout, then raise
 # to 3 IN PLACE (see README "Scale to R3") once all three nodes have joined. Only
 # set this to 3 here after the third node is up and you re-run the KV update.
-KV_REPLICAS=1
+KV_REPLICAS=3
 
 # Ports (same on every node; the route port is server-to-server only).
 NATS_CLIENT_PORT=4250
@@ -30,15 +30,28 @@ SSH_USER="root"
 # Which address family the inter-node routes use. "wg" builds --routes from the
 # WireGuard mesh IPs (private server-to-server links, preferred); "public" uses
 # the public IPs. The route layer is always mutual-TLS regardless.
-ROUTE_NETWORK="wg"
+#
+# DEPLOY DECISION (2026-06-07): set to "public". No WireGuard mesh exists between
+# the three cluster nodes — homer and datardos do not even have the `wg` binary
+# installed, and om's only WG peers are the operator's personal PCs, not the VPS.
+# Rather than stand up a fresh mesh blindly, the routes go over the public IPs,
+# still protected by the separate cluster route CA (mutual-TLS). On magnus (the
+# only node with ufw active) the route port 6250 is restricted to the homer and
+# datardos public IPs; homer/datardos run ufw inactive (Docker hosts) and rely on
+# the route mutual-TLS for 6250.
+ROUTE_NETWORK="public"
 
 # One row per node: NAME  SSH_HOST  PUBLIC_IP  WG_IP
 #   NAME      -> --server-name and the per-node cert filenames (unique).
-#   SSH_HOST  -> the `ssh <SSH_HOST>` alias (see ~/.ssh/config).
+#   SSH_HOST  -> the `ssh ALIAS` alias (see ~/.ssh/config).
 #   PUBLIC_IP -> public address; goes in the cert SANs (client-facing data plane).
 #   WG_IP     -> WireGuard mesh address; cert SAN + route target when ROUTE_NETWORK=wg.
+# NOTE: with ROUTE_NETWORK=public and no WireGuard mesh, the WG_IP column is set to
+# each node's public IP so the cert SAN covers the address actually used by the
+# public routes and no unfilled placeholder remains (scripts refuse to run otherwise).
+# magnus == organic-machine.com == om (135.125.201.30); SSH alias `magnus` enters as root.
 CLUSTER_NODES=(
-  "magnus    magnus  <MAGNUS_PUBLIC_IP>  <MAGNUS_WG_IP>"
-  "homer     homer   141.94.69.66        <HOMER_WG_IP>"
-  "datardos  dd      51.91.100.142       <DATARDOS_WG_IP>"
+  "magnus    magnus  135.125.201.30  135.125.201.30"
+  "homer     homer   141.94.69.66    141.94.69.66"
+  "datardos  dd      51.91.100.142   51.91.100.142"
 )

dev/issues/0007-jetstream-encryption-at-rest.md

@@ -0,0 +1,78 @@
+---
+issue: 0007
+title: Cifrado at-rest del control plane (JetStream KV / SQLite en disco)
+status: spec
+created: 2026-06-07
+domain: security
+scope: unibus (pkg/embeddednats, cmd/membershipd, deploy/cluster) + procedimiento de migración del store existente
+---
+
+# Objetivo
+
+Cifrar en reposo el almacenamiento del plano de control para que un nodo comprometido
+(root en el VPS) o un disco robado no exponga los metadatos de control en claro.
+
+Estado actual (auditado el 07/06/2026, report 0012 y siguientes):
+
+- **Contenido de los mensajes**: cifrado E2E por room (megolm/olm). El servidor nunca ve el
+  plaintext; no vive en el plano de control. **No es el objeto de este issue.**
+- **Claves de room** (`UNIBUS_room_keys`): guardadas **selladas** (sealed box X25519, cifradas
+  para cada miembro). El servidor las almacena y reparte pero no puede abrirlas. **Ya protegidas.**
+- **Metadatos de control** (`UNIBUS_rooms`, `UNIBUS_members`, `UNIBUS_rooms_by_member`,
+  `UNIBUS_users`): se serializan con `json.Marshal` y se escriben **en claro** en el store. En
+  cluster ese store es el directorio `local_files/jetstream/` de cada nodo; en single-node es el
+  archivo SQLite `local_files/unibus.db`. Hoy **no hay cifrado at-rest**: con root en un nodo se
+  pueden leer subjects de salas, la pertenencia (quién está en qué sala con qué rol), los handles
+  y roles de los usuarios, y las claves públicas (signPub/kexPub). No se exponen mensajes (E2E) ni
+  se pueden descifrar salas (claves selladas), pero sí toda la topología.
+
+Tras este issue, los buckets/archivos del control plane quedan cifrados en disco con una clave por
+nodo gestionada fuera de git. El modelo de amenaza pasa de "root del nodo ve la topología" a "root
+del nodo necesita además la clave at-rest (que puede vivir en un secreto separado / TPM / variable
+de entorno inyectada) para leer cualquier cosa".
+
+# Contexto técnico
+
+- NATS Server / JetStream soporta **encryption at-rest** nativo: se configura una cifra
+  (`aes` o `chacha20`) y una clave; JetStream cifra los ficheros de los streams/KV en disco. El
+  bus usa un NATS **embebido** (`pkg/embeddednats`), así que la activación es por opciones del
+  servidor embebido, no por un `nats-server.conf` externo.
+- Para el backend SQLite (single-node) el equivalente sería SQLCipher o cifrado a nivel de
+  archivo/FS; queda como sub-tarea de menor prioridad porque el despliegue real es cluster (KV).
+
+# Tareas
+
+1. Confirmar la API de encryption-at-rest del NATS embebido en la versión usada (opción de
+   servidor para cipher + clave; cómo se pasa la clave de forma que no quede en argv ni en git).
+2. Activar el cifrado en `pkg/embeddednats` detrás de una opción de configuración. La clave se
+   inyecta por archivo (`--jetstream-encryption-key-file`, 0600, junto a las claves TLS del nodo)
+   o variable de entorno desde el unit systemd; nunca en argv ni commiteada.
+3. `cmd/membershipd`: flag/env para la clave + reflejar el estado en la posture publicada en
+   `/healthz` (p.ej. `"at_rest":true`) para que el monitor lo verifique.
+4. `deploy/cluster`: provisionar la clave at-rest por nodo (generación + `pass`/secrets gitignored)
+   y cablearla en `cluster.env` + el unit. Documentar en el runbook.
+5. **Migración del store existente** (gotcha crítico): JetStream no re-cifra retroactivamente los
+   datos ya escritos en claro. Diseñar y documentar el procedimiento seguro para el cluster en
+   producción (probable: backup → exportar snapshot del control plane → parar nodo → recrear el
+   store con la clave activa → re-importar; o rotación nodo a nodo aprovechando la replicación R3).
+   Respetar la regla de migraciones (aditivo, sin pérdida de datos).
+6. Tests: arrancar un nodo con clave at-rest, escribir un user/room, y verificar que el fichero en
+   disco **no** contiene en claro un subject/handle conocido (grep negativo), y que el nodo sigue
+   leyéndolos con la clave. Verificar que sin la clave el store no se abre.
+
+# Definition of Done
+
+- Cifrado at-rest activo en los 3 nodos del cluster; `/healthz` lo refleja en la posture.
+- Evidencia ejecutable: un valor conocido (subject de sala / handle de usuario) **no** aparece en
+  claro al hacer `grep` sobre `local_files/jetstream/`; el nodo lo sigue sirviendo con la clave.
+- Procedimiento de migración probado sobre datos reales sin pérdida (snapshot/restore verificado).
+- La clave at-rest nunca está en git ni en argv; vive en archivo 0600 / secreto inyectado.
+- No baja ninguna otra capa de seguridad (enforce + ACL + TLS + E2E + sealed keys intactas).
+
+# Notas
+
+Aditivo y ortogonal al resto de la seguridad: TLS protege en tránsito, E2E el contenido, las claves
+de room van selladas; este issue cierra el último hueco (metadatos de control en claro en disco)
+para el modelo de amenaza "VPS comprometido / disco robado". Prioridad media: el despliegue ya es
+seguro frente a ataques de red (enforce+TLS+ACL); esto endurece frente a compromiso físico/root del
+host. Relacionado con el endurecimiento de los issues 0004/0005/0006.

pkg/client/client.go

@@ -456,6 +456,23 @@ type memberRoomJSON struct {
 	Role    string     `json:"role"`
 }
 
+// userJSON mirrors the server's wire type on the admin user-management endpoints.
+type userJSON struct {
+	SignPub   string `json:"sign_pub"`
+	Handle    string `json:"handle"`
+	Role      string `json:"role"`
+	Status    string `json:"status"`
+	CreatedAt string `json:"created_at"`
+	RevokedAt string `json:"revoked_at,omitempty"`
+}
+
+// addUserReq is the POST /users body (mirror of the server type).
+type addUserReq struct {
+	SignPub string `json:"sign_pub"`
+	Handle  string `json:"handle"`
+	Role    string `json:"role"`
+}
+
 // ---- room operations ------------------------------------------------------
 
 // RoomRef is a room this peer belongs to, returned by ListMyRooms. It is the
@@ -490,6 +507,59 @@ func (c *Client) ListMyRooms() ([]RoomRef, error) {
 	return out, nil
 }
 
+// ---- user administration (admin-only) ------------------------------------
+
+// UserInfo is a bus user as returned by the admin user-management endpoints. It
+// is a flat view (no nested types) for the admin panel: the signing key
+// (lowercase hex), handle, role ("admin"|"member"), status ("active"|"revoked"),
+// and timestamps. RevokedAt is empty for an active user.
+type UserInfo struct {
+	SignPub   string
+	Handle    string
+	Role      string
+	Status    string
+	CreatedAt string
+	RevokedAt string
+}
+
+// ListUsers returns the full bus allowlist, including revoked users. The caller
+// must be signing as an admin: a non-admin signer is rejected by the server with
+// 403, surfaced here as an error.
+func (c *Client) ListUsers() ([]UserInfo, error) {
+	var resp []userJSON
+	if err := c.doJSON("GET", "/users", nil, &resp); err != nil {
+		return nil, err
+	}
+	out := make([]UserInfo, 0, len(resp))
+	for _, u := range resp {
+		out = append(out, UserInfo{
+			SignPub:   u.SignPub,
+			Handle:    u.Handle,
+			Role:      u.Role,
+			Status:    u.Status,
+			CreatedAt: u.CreatedAt,
+			RevokedAt: u.RevokedAt,
+		})
+	}
+	return out, nil
+}
+
+// AddUser registers a bus user from their Ed25519 signing public key (64-hex).
+// role is "admin" or "member" (empty defaults to member, matching the server).
+// The caller must be signing as an admin. Re-adding an already-registered key
+// returns an error (the server replies 409 and leaves the existing row
+// untouched — no silent role/status change).
+func (c *Client) AddUser(signPub, handle, role string) error {
+	return c.doJSON("POST", "/users", addUserReq{SignPub: signPub, Handle: handle, Role: role}, nil)
+}
+
+// RevokeUser revokes a bus user by their signing public key (64-hex). Revocation
+// is a status flip (no hard delete): the identity stays auditable and is denied
+// on both planes immediately. The caller must be signing as an admin.
+func (c *Client) RevokeUser(signPub string) error {
+	return c.doJSON("POST", "/users/"+signPub+"/revoke", nil, nil)
+}
+
 // newRoomKey returns 32 random bytes for a symmetric room key.
 func newRoomKey() ([]byte, error) {
 	k := make([]byte, 32)

pkg/client/identity.go

@@ -33,20 +33,36 @@ type identityFile struct {
 	KexPriv  string `json:"kex_priv"`
 }
 
+// LoadIdentity loads an existing identity from path. Unlike LoadOrCreateIdentity
+// it NEVER creates one: a missing or unreadable file is an error. It is for
+// callers that must consume a specific, pre-provisioned identity rather than mint
+// a fresh one — for example membershipd's persisted internal service identity,
+// which `membershipd user add --store kv` reads to present the privileged nkey
+// the cluster authenticator recognizes.
+func LoadIdentity(path string) (cs.Identity, error) {
+	data, err := os.ReadFile(path)
+	if err != nil {
+		return cs.Identity{}, fmt.Errorf("client: read identity %q: %w", path, err)
+	}
+	var f identityFile
+	if err := json.Unmarshal(data, &f); err != nil {
+		return cs.Identity{}, fmt.Errorf("client: parse identity %q: %w", path, err)
+	}
+	id, err := f.toIdentity()
+	if err != nil {
+		return cs.Identity{}, fmt.Errorf("client: decode identity %q: %w", path, err)
+	}
+	return id, nil
+}
+
 // LoadOrCreateIdentity loads the identity at path, or generates and persists a
 // new one if the file does not exist. The file is written with 0600
-// permissions because it holds private keys.
+// permissions because it holds private keys. A file that exists but is
+// unreadable or corrupt is an error (NOT silently regenerated), so a damaged
+// identity surfaces instead of minting a new key that cannot decrypt old data.
 func LoadOrCreateIdentity(path string) (cs.Identity, error) {
-	if data, err := os.ReadFile(path); err == nil {
-		var f identityFile
-		if err := json.Unmarshal(data, &f); err != nil {
-			return cs.Identity{}, fmt.Errorf("client: parse identity %q: %w", path, err)
-		}
-		id, err := f.toIdentity()
-		if err != nil {
-			return cs.Identity{}, fmt.Errorf("client: decode identity %q: %w", path, err)
-		}
-		return id, nil
+	if _, statErr := os.Stat(path); statErr == nil {
+		return LoadIdentity(path)
 	}
 
 	id, err := cs.GenerateIdentity()

pkg/client/users_test.go

@@ -0,0 +1,99 @@
+package client_test
+
+import (
+	"encoding/hex"
+	"strings"
+	"testing"
+
+	"github.com/enmanuel/unibus/pkg/client"
+	"github.com/enmanuel/unibus/pkg/membership"
+)
+
+// findUserInfo returns the row with the given signing key (case-insensitive).
+func findUserInfo(users []client.UserInfo, signPub string) (client.UserInfo, bool) {
+	want := strings.ToLower(signPub)
+	for _, u := range users {
+		if strings.ToLower(u.SignPub) == want {
+			return u, true
+		}
+	}
+	return client.UserInfo{}, false
+}
+
+// TestClientUsersAdminAPI drives the admin user-management API through the real
+// pkg/client methods against an in-process membershipd under enforce: an admin
+// client adds a user, lists it, revokes it, and sees the status flip — and a
+// non-admin client is denied. This is the path the admin panel uses, so it locks
+// the client/server contract the panel depends on.
+func TestClientUsersAdminAPI(t *testing.T) {
+	h := newHarnessMode(t, membership.AuthEnforce)
+	waitHealth(t, h.ctrlURL)
+
+	admin, err := client.New(h.natsURL, h.ctrlURL, mustIdentity(t))
+	if err != nil {
+		t.Fatalf("connect admin: %v", err)
+	}
+	defer admin.Close()
+	registerClient(t, h, admin, "admin", membership.RoleAdmin)
+
+	member, err := client.New(h.natsURL, h.ctrlURL, mustIdentity(t))
+	if err != nil {
+		t.Fatalf("connect member: %v", err)
+	}
+	defer member.Close()
+	registerClient(t, h, member, "member", membership.RoleMember)
+
+	// A brand-new identity the admin will register over HTTP.
+	carol := mustIdentity(t)
+	carolPub := hex.EncodeToString(carol.SignPub)
+
+	// Admin adds carol as a member.
+	if err := admin.AddUser(carolPub, "carol", membership.RoleMember); err != nil {
+		t.Fatalf("admin AddUser: %v", err)
+	}
+
+	// Admin lists: carol present and active.
+	users, err := admin.ListUsers()
+	if err != nil {
+		t.Fatalf("admin ListUsers: %v", err)
+	}
+	row, ok := findUserInfo(users, carolPub)
+	if !ok {
+		t.Fatalf("carol missing from list after add: %+v", users)
+	}
+	if row.Status != membership.StatusActive || row.Role != membership.RoleMember {
+		t.Fatalf("carol row wrong after add: %+v", row)
+	}
+
+	// Re-adding the same key is a conflict surfaced as an error (no silent upsert).
+	if err := admin.AddUser(carolPub, "carol-again", membership.RoleAdmin); err == nil {
+		t.Fatalf("re-adding carol should error (409), got nil")
+	}
+
+	// Admin revokes carol; list shows the status flip (no hard delete).
+	if err := admin.RevokeUser(carolPub); err != nil {
+		t.Fatalf("admin RevokeUser: %v", err)
+	}
+	users, err = admin.ListUsers()
+	if err != nil {
+		t.Fatalf("admin ListUsers after revoke: %v", err)
+	}
+	row, ok = findUserInfo(users, carolPub)
+	if !ok {
+		t.Fatalf("carol vanished after revoke (should be a status flip): %+v", users)
+	}
+	if row.Status != membership.StatusRevoked {
+		t.Fatalf("carol should be revoked, got status %q", row.Status)
+	}
+
+	// A non-admin (member) is denied on every user-management method.
+	if _, err := member.ListUsers(); err == nil {
+		t.Fatalf("non-admin ListUsers should error (403), got nil")
+	}
+	if err := member.AddUser(carolPub, "x", membership.RoleMember); err == nil {
+		t.Fatalf("non-admin AddUser should error (403), got nil")
+	}
+	if err := member.RevokeUser(carolPub); err == nil {
+		t.Fatalf("non-admin RevokeUser should error (403), got nil")
+	}
+}

pkg/embeddednats/embeddednats.go

@@ -9,6 +9,7 @@ import (
 	"crypto/tls"
 	"fmt"
 	"net/url"
+	"os"
 	"time"
 
 	server "github.com/nats-io/nats-server/v2/server"
@@ -106,6 +107,13 @@ func StartHostAuth(storeDir, host string, port int, auth server.Authentication)
 // blocks until the server is ready to accept connections (up to 5s) and returns
 // the running server; the caller must Shutdown it.
 func StartServer(cfg ServerConfig) (*server.Server, error) {
+	// Diagnostic toggle: UNIBUS_NATS_DEBUG=1 enables the embedded nats-server's own
+	// logger (route/RAFT/JetStream errors), which is otherwise silenced. Off by
+	// default so production behavior is unchanged; only set it when debugging the
+	// cluster route layer.
+	debugLevel := os.Getenv("UNIBUS_NATS_DEBUG")
+	debugNATS := debugLevel == "1" || debugLevel == "2"
+	traceNATS := debugLevel == "2"
 	opts := &server.Options{
 		JetStream:  true,
 		StoreDir:   cfg.StoreDir,
@@ -114,8 +122,17 @@ func StartServer(cfg ServerConfig) (*server.Server, error) {
 		ServerName: cfg.ServerName,
 		DontListen: false,
 		// Keep the embedded server quiet by default; the host app logs the URLs.
-		NoLog:  true,
-		NoSigs: true,
+		NoLog:   !debugNATS,
+		Debug:   debugNATS,
+		Trace:   traceNATS,
+		Logtime: true,
+		NoSigs:  true,
+	}
+	if debugNATS {
+		// Expose the nats-server monitoring endpoint (loopback) so the operator can
+		// inspect /jsz, /routez, /varz while debugging the cluster meta-group.
+		opts.HTTPHost = "127.0.0.1"
+		opts.HTTPPort = 8222
 	}
 	if cfg.Auth != nil {
 		opts.CustomClientAuthentication = cfg.Auth
@@ -141,6 +158,10 @@ func StartServer(cfg ServerConfig) (*server.Server, error) {
 		return nil, fmt.Errorf("embeddednats: new server: %w", err)
 	}
 
+	if debugNATS {
+		ns.ConfigureLogger()
+	}
+
 	go ns.Start()
 
 	if !ns.ReadyForConnections(5 * time.Second) {
@@ -162,6 +183,21 @@ func applyClusterOpts(opts *server.Options, c *ClusterConfig) error {
 		Port:     c.Port,
 		Username: c.Username,
 		Password: c.Password,
+		// Disable route connection pooling (nats-server 2.10+ defaults to a pool of
+		// 3 connections per peer). On a small cluster the pool churns with
+		// "duplicate route"/"client closed" reconnects that interrupt the meta-group
+		// RAFT heartbeats, causing perpetual leader re-elections so the JetStream
+		// meta never becomes current and stream/KV creation hangs (issue 0006g).
+		// PoolSize=-1 forces the classic single route per peer, which is stable for
+		// the 3-node unibus cluster.
+		PoolSize: -1,
+		// NoAdvertise stops the server from gossiping its locally-discovered IPs to
+		// peers. The cluster nodes are Docker hosts, so without this NATS advertises
+		// the docker bridge addresses (172.x / 10.0.x) as reachable routes; peers
+		// then try to dial those private, mutually-unreachable IPs, churning the
+		// route layer and destabilizing the JetStream meta-group. With NoAdvertise
+		// the nodes use ONLY the explicit public-IP routes we configure (issue 0006g).
+		NoAdvertise: true,
 	}
 	if c.TLS != nil {
 		opts.Cluster.TLSConfig = c.TLS

pkg/membership/jetstream_store.go

@@ -85,8 +85,18 @@ func OpenJetStream(js jetstream.JetStream, cfg JetStreamConfig) (Store, error) {
 	if opTimeout <= 0 {
 		opTimeout = defaultKVOpTime
 	}
-	ctx, cancel := context.WithTimeout(context.Background(), 15*time.Second)
-	defer cancel()
+	// Bootstrap budget for creating/opening the buckets. On a single node JetStream
+	// is ready the instant the server starts, so the first attempt succeeds. On a
+	// COLD multi-node cluster the JetStream meta-group must first elect a leader and
+	// each node must establish contact with it before its $JS.API responds. A KV
+	// op is a NATS request/reply: if it is published before the node's JetStream is
+	// ready the request is dropped (not queued), and a single long-context call then
+	// just blocks until it times out (issue 0006g). So we RETRY each bucket op with
+	// short per-attempt contexts until it succeeds or the overall bootstrap budget
+	// is exhausted; once the cluster is ready the next retry lands and the buckets
+	// are created, after which they persist and every node opens them quickly.
+	bootstrapBudget := 120 * time.Second
+	deadline := time.Now().Add(bootstrapBudget)
 
 	s := &jetstreamStore{opTimeout: opTimeout}
 	for _, b := range []struct {
@@ -99,14 +109,27 @@ func OpenJetStream(js jetstream.JetStream, cfg JetStreamConfig) (Store, error) {
 		{bucketRoomKeys, &s.keys},
 		{bucketUsers, &s.users},
 	} {
-		kv, err := js.CreateOrUpdateKeyValue(ctx, jetstream.KeyValueConfig{
-			Bucket:   b.name,
-			Replicas: cfg.Replicas,
-			History:  1,
-			Storage:  jetstream.FileStorage,
-		})
-		if err != nil {
-			return nil, fmt.Errorf("membership: open KV bucket %q (replicas=%d): %w", b.name, cfg.Replicas, err)
+		var kv jetstream.KeyValue
+		var lastErr error
+		for {
+			opCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+			kv, lastErr = js.CreateOrUpdateKeyValue(opCtx, jetstream.KeyValueConfig{
+				Bucket:   b.name,
+				Replicas: cfg.Replicas,
+				History:  1,
+				Storage:  jetstream.FileStorage,
+			})
+			cancel()
+			if lastErr == nil {
+				break
+			}
+			if time.Now().After(deadline) {
+				return nil, fmt.Errorf("membership: open KV bucket %q (replicas=%d) after %s: %w", b.name, cfg.Replicas, bootstrapBudget, lastErr)
+			}
+			// JetStream not ready yet (no meta leader / request dropped). Wait and
+			// re-publish the op; in a cluster cold start this lands once the meta
+			// group settles.
+			time.Sleep(1 * time.Second)
 		}
 		*b.dst = kv
 	}

pkg/membership/server.go

@@ -213,9 +213,12 @@ func (s *Server) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 		writeErr(w, http.StatusUnauthorized, "unauthorized: "+err.Error())
 		return
 	}
-	// Carry the authenticated signer's endpoint into the handler so room handlers
-	// can authorize by membership (audit H3). Only set on a verified identity.
-	s.mux.ServeHTTP(w, r.WithContext(withSigner(r.Context(), res.endpoint)))
+	// Carry the authenticated signer's endpoint AND signing key into the handler.
+	// Room handlers authorize by membership via the endpoint (audit H3); the
+	// user-management handlers authorize by role via the signing key (the endpoint
+	// id is a one-way hash of the key, so it cannot be reversed to look the signer
+	// up in the user allowlist). Both are set only on a verified identity.
+	s.mux.ServeHTTP(w, r.WithContext(withSigner(r.Context(), res.endpoint, res.pubHex)))
 }
 
 // isBodyTooLarge reports whether err is the sentinel returned by MaxBytesReader
@@ -229,11 +232,19 @@ func isBodyTooLarge(err error) bool {
 // values cannot collide with keys set by other packages.
 type ctxKey int
 
-const ctxSignerEndpoint ctxKey = iota
+const (
+	ctxSignerEndpoint ctxKey = iota
+	ctxSignerPub
+)
 
-// withSigner returns a context carrying the authenticated signer's endpoint id.
-func withSigner(ctx context.Context, endpoint string) context.Context {
-	return context.WithValue(ctx, ctxSignerEndpoint, endpoint)
+// withSigner returns a context carrying the authenticated signer's endpoint id
+// and signing public key (lowercase hex). The endpoint authorizes room
+// membership; the signing key authorizes user-management by role, because the
+// endpoint id is a one-way hash of the key (base64url(sha256(signPub))) and so
+// cannot be reversed to look the signer up in the user allowlist.
+func withSigner(ctx context.Context, endpoint, pubHex string) context.Context {
+	ctx = context.WithValue(ctx, ctxSignerEndpoint, endpoint)
+	return context.WithValue(ctx, ctxSignerPub, pubHex)
 }
 
 // signerEndpoint returns the authenticated signer's endpoint id and whether one
@@ -245,6 +256,16 @@ func signerEndpoint(r *http.Request) (string, bool) {
 	return v, ok && v != ""
 }
 
+// signerPubHex returns the authenticated signer's signing public key (lowercase
+// hex) and whether one is present. Like signerEndpoint it is absent under
+// AuthOff and on a soft-mode pass-through; the user-management handlers treat
+// that absence as "no admin identity" and deny (default-deny), since a
+// privilege-granting operation must never run without a verified admin.
+func signerPubHex(r *http.Request) (string, bool) {
+	v, ok := r.Context().Value(ctxSignerPub).(string)
+	return v, ok && v != ""
+}
+
 // requireMember authorizes a room request by membership (audit H3): it returns
 // the signer endpoint and true when the request may proceed, or writes 403 and
 // returns false when an authenticated signer is not a member of roomID. When no
@@ -262,6 +283,31 @@ func (s *Server) requireMember(w http.ResponseWriter, r *http.Request, roomID st
 	return signer, true
 }
 
+// requireAdmin authorizes a user-management request: it returns the signer's
+// signing-key hex and true ONLY when the authenticated signer is a user with
+// role admin and active status; otherwise it writes 403 and returns false.
+//
+// Default-deny, with no dev relaxation: unlike requireMember (which allows a
+// request when no authenticated signer is present, preserving AuthOff/dev
+// behavior for room reads), this denies whenever the signer is absent or is not
+// a verified active admin. The user-management endpoints grant and revoke bus
+// access, so they must never be reachable without a verified admin identity —
+// the store is consulted on every call so a just-revoked admin is denied
+// immediately, and any store error fails closed.
+func (s *Server) requireAdmin(w http.ResponseWriter, r *http.Request) (string, bool) {
+	pubHex, ok := signerPubHex(r)
+	if !ok {
+		writeErr(w, http.StatusForbidden, "forbidden: admin role required")
+		return "", false
+	}
+	u, err := s.store.GetUser(pubHex)
+	if err != nil || u.Role != RoleAdmin || u.Status != StatusActive {
+		writeErr(w, http.StatusForbidden, "forbidden: admin role required")
+		return "", false
+	}
+	return pubHex, true
+}
+
 // isAuthExempt lists requests that bypass control-plane auth even under enforce.
 // Only the unauthenticated health probe qualifies: it carries no data and is
 // needed by load balancers / smoke checks / systemd before any identity exists.
@@ -280,6 +326,13 @@ func (s *Server) routes() {
 	s.mux.HandleFunc("GET /rooms/{id}", s.handleGetRoom)
 	s.mux.HandleFunc("POST /blobs", s.handlePutBlob)
 	s.mux.HandleFunc("GET /blobs/{hash}", s.handleGetBlob)
+	// User-management (admin-only) — the HTTP-signed equivalent of the local
+	// `membershipd user` CLI, so the admin panel manages the bus allowlist by
+	// signing as an admin instead of needing direct store/KV access. All three
+	// pass through requireAdmin; they hit the same store the room handlers do.
+	s.mux.HandleFunc("GET /users", s.handleListUsers)
+	s.mux.HandleFunc("POST /users", s.handleAddUser)
+	s.mux.HandleFunc("POST /users/{signpub}/revoke", s.handleRevokeUser)
 }
 
 // ---- wire types -----------------------------------------------------------
@@ -357,6 +410,27 @@ type blobResp struct {
 	Hash string `json:"hash"`
 }
 
+// userJSON is the wire representation of a bus user on the admin endpoints. It
+// carries the full record the panel needs to render the allowlist, including
+// status (so revoked users are visible) and the timestamps. revoked_at is
+// omitted for an active user.
+type userJSON struct {
+	SignPub   string `json:"sign_pub"`
+	Handle    string `json:"handle"`
+	Role      string `json:"role"`
+	Status    string `json:"status"`
+	CreatedAt string `json:"created_at"`
+	RevokedAt string `json:"revoked_at,omitempty"`
+}
+
+// addUserReq is the POST /users body: the new user's Ed25519 signing key
+// (64-hex), human handle, and role. role is optional and defaults to member.
+type addUserReq struct {
+	SignPub string `json:"sign_pub"`
+	Handle  string `json:"handle"`
+	Role    string `json:"role"`
+}
+
 // ---- helpers --------------------------------------------------------------
 
 func writeJSON(w http.ResponseWriter, code int, v any) {
@@ -674,3 +748,95 @@ func (s *Server) handleGetBlob(w http.ResponseWriter, r *http.Request) {
 	w.WriteHeader(http.StatusOK)
 	_, _ = w.Write(data)
 }
+
+// ---- user-management handlers (admin-only) --------------------------------
+
+// handleListUsers returns the full bus allowlist, including revoked users, so an
+// admin sees the complete picture (a revoked identity stays auditable). Admin-only.
+func (s *Server) handleListUsers(w http.ResponseWriter, r *http.Request) {
+	if _, ok := s.requireAdmin(w, r); !ok {
+		return
+	}
+	users, err := s.store.ListUsers()
+	if err != nil {
+		writeServerErr(w, r, http.StatusInternalServerError, "internal error", err)
+		return
+	}
+	out := make([]userJSON, 0, len(users))
+	for _, u := range users {
+		out = append(out, userJSON{
+			SignPub:   u.SignPub,
+			Handle:    u.Handle,
+			Role:      u.Role,
+			Status:    u.Status,
+			CreatedAt: u.CreatedAt,
+			RevokedAt: u.RevokedAt,
+		})
+	}
+	writeJSON(w, http.StatusOK, out)
+}
+
+// handleAddUser registers a new bus user from an admin-supplied Ed25519 signing
+// key. It mirrors the `membershipd user add` CLI: the key must be 64-hex, the
+// role must be admin or member (empty defaults to member), and re-adding an
+// already-registered key is a 409 that leaves the existing row untouched — no
+// silent upsert that could flip a role or clobber status. Admin-only.
+func (s *Server) handleAddUser(w http.ResponseWriter, r *http.Request) {
+	if _, ok := s.requireAdmin(w, r); !ok {
+		return
+	}
+	var req addUserReq
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		writeErr(w, http.StatusBadRequest, "bad json: "+err.Error())
+		return
+	}
+	if req.SignPub == "" || req.Handle == "" {
+		writeErr(w, http.StatusBadRequest, "sign_pub and handle required")
+		return
+	}
+	if err := ValidateSignPubHex(req.SignPub); err != nil {
+		writeErr(w, http.StatusBadRequest, err.Error())
+		return
+	}
+	role := req.Role
+	if role == "" {
+		role = RoleMember
+	}
+	if role != RoleAdmin && role != RoleMember {
+		writeErr(w, http.StatusBadRequest,
+			fmt.Sprintf("invalid role %q (want %q or %q)", role, RoleAdmin, RoleMember))
+		return
+	}
+	if err := s.store.AddUser(req.SignPub, req.Handle, role); err != nil {
+		if errors.Is(err, ErrUserExists) {
+			// Idempotency contract (mirrors the CLI): re-adding a key is an explicit,
+			// non-destructive conflict. To replace a user, revoke then add again.
+			writeErr(w, http.StatusConflict,
+				"user already registered (unchanged); revoke it first to replace")
+			return
+		}
+		writeServerErr(w, r, http.StatusInternalServerError, "internal error", err)
+		return
+	}
+	writeJSON(w, http.StatusCreated, map[string]string{"status": "added"})
+}
+
+// handleRevokeUser revokes a bus user by signing key. Revocation is a status
+// flip (no hard delete) so the identity stays auditable and IsAuthorized denies
+// it on both planes immediately. Revoking an unknown or already-revoked key is a
+// 404. Admin-only.
+func (s *Server) handleRevokeUser(w http.ResponseWriter, r *http.Request) {
+	if _, ok := s.requireAdmin(w, r); !ok {
+		return
+	}
+	signPub := r.PathValue("signpub")
+	if err := ValidateSignPubHex(signPub); err != nil {
+		writeErr(w, http.StatusBadRequest, err.Error())
+		return
+	}
+	if err := s.store.RevokeUser(signPub); err != nil {
+		writeServerErr(w, r, http.StatusNotFound, "no active user with that key", err)
+		return
+	}
+	writeJSON(w, http.StatusOK, map[string]string{"status": "revoked"})
+}

pkg/membership/users.go

@@ -2,6 +2,7 @@ package membership
 
 import (
 	"database/sql"
+	"encoding/hex"
 	"errors"
 	"fmt"
 	"strings"
@@ -35,6 +36,23 @@ type User struct {
 	RevokedAt string // empty unless revoked
 }
 
+// ValidateSignPubHex ensures signPub is exactly a 32-byte Ed25519 public key in
+// hex (64 hex chars). It is the single source of truth for that check, shared by
+// the local admin CLI (which validates before seeding the first admin) and the
+// HTTP user-management handlers (which validate an admin-supplied key before it
+// reaches the store). Catching a malformed key here turns a silent "authorized
+// nobody" into an explicit error at the boundary.
+func ValidateSignPubHex(signPub string) error {
+	b, err := hex.DecodeString(signPub)
+	if err != nil {
+		return fmt.Errorf("sign-pub is not valid hex: %w", err)
+	}
+	if len(b) != 32 {
+		return fmt.Errorf("sign-pub must be a 32-byte Ed25519 public key (64 hex chars), got %d bytes", len(b))
+	}
+	return nil
+}
+
 // normalizeSignPub lowercases the hex key so lookups are case-insensitive: the
 // primary key is stored lowercase and every query normalizes its input the same
 // way, so a caller passing uppercase hex still matches.

pkg/membership/users_http_test.go

@@ -0,0 +1,164 @@
+package membership
+
+import (
+	"encoding/hex"
+	"encoding/json"
+	"net/http"
+	"testing"
+	"time"
+
+	cs "fn-registry/functions/cybersecurity"
+)
+
+// signedJSON is signedReq for a JSON body: it marshals v and signs the request
+// as id with a distinct nonce. It returns the response status and body, reusing
+// the auth_test harness so these tests exercise the real signed wire contract.
+func signedJSON(t *testing.T, h *authHarness, method, path string, v any, id cs.Identity, n int) (int, string) {
+	t.Helper()
+	var body []byte
+	if v != nil {
+		b, err := json.Marshal(v)
+		if err != nil {
+			t.Fatalf("marshal body: %v", err)
+		}
+		body = b
+	}
+	return do(t, signedReq(t, h.ts.URL, method, path, body, id, time.Now().Unix(), nonceN(n)))
+}
+
+// TestUsersHTTP_NonAdminForbidden is the security spine: a REGISTERED but
+// non-admin signer (bob, role member) is denied on every user-management
+// endpoint. His signature clears auth (he is in the allowlist), so each request
+// reaches the handler, where requireAdmin returns 403 — default-deny by role.
+func TestUsersHTTP_NonAdminForbidden(t *testing.T) {
+	h := newAuthHarness(t, AuthEnforce)
+
+	bob, _ := cs.GenerateIdentity()
+	register(t, h, bob, "bob") // role member (see register in authz_test.go)
+	bobPub := hex.EncodeToString(bob.SignPub)
+
+	victim, _ := cs.GenerateIdentity()
+	victimPub := hex.EncodeToString(victim.SignPub)
+
+	checks := []struct {
+		name   string
+		method string
+		path   string
+		body   any
+	}{
+		{"list users", "GET", "/users", nil},
+		{"add user", "POST", "/users", addUserReq{SignPub: victimPub, Handle: "mallory", Role: RoleMember}},
+		{"revoke user", "POST", "/users/" + bobPub + "/revoke", nil},
+	}
+	for i, c := range checks {
+		code, body := signedJSON(t, h, c.method, c.path, c.body, bob, i+1)
+		if code != http.StatusForbidden {
+			t.Fatalf("non-admin %s should be 403, got %d (%s)", c.name, code, body)
+		}
+	}
+}
+
+// TestUsersHTTP_AdminRoundtrip exercises the golden path end to end: alice (the
+// seeded admin) adds carol, sees her in the list as active, revokes her, then
+// sees her status flip to revoked (no hard delete — she stays in the list).
+func TestUsersHTTP_AdminRoundtrip(t *testing.T) {
+	h := newAuthHarness(t, AuthEnforce)
+
+	carol, _ := cs.GenerateIdentity()
+	carolPub := hex.EncodeToString(carol.SignPub)
+
+	// Add carol as a member.
+	if code, body := signedJSON(t, h, "POST", "/users",
+		addUserReq{SignPub: carolPub, Handle: "carol", Role: RoleMember}, h.alice, 1); code != http.StatusCreated {
+		t.Fatalf("admin add carol should be 201, got %d (%s)", code, body)
+	}
+
+	// List: carol present and active; alice (the seed admin) also present.
+	users := listUsers(t, h, 2)
+	carolRow, ok := findUser(users, carolPub)
+	if !ok {
+		t.Fatalf("carol missing from list after add: %+v", users)
+	}
+	if carolRow.Status != StatusActive || carolRow.Role != RoleMember || carolRow.Handle != "carol" {
+		t.Fatalf("carol row wrong after add: %+v", carolRow)
+	}
+	if _, ok := findUser(users, h.alicePub); !ok {
+		t.Fatalf("seeded admin alice missing from list: %+v", users)
+	}
+
+	// Revoke carol.
+	if code, body := signedJSON(t, h, "POST", "/users/"+carolPub+"/revoke", nil, h.alice, 3); code != http.StatusOK {
+		t.Fatalf("admin revoke carol should be 200, got %d (%s)", code, body)
+	}
+
+	// List again: carol still present, now revoked (status flip, not delete).
+	users = listUsers(t, h, 4)
+	carolRow, ok = findUser(users, carolPub)
+	if !ok {
+		t.Fatalf("carol vanished from list after revoke (should be a status flip): %+v", users)
+	}
+	if carolRow.Status != StatusRevoked {
+		t.Fatalf("carol should be revoked, got status %q", carolRow.Status)
+	}
+}
+
+// TestUsersHTTP_Validation covers the input-validation contract: a malformed hex
+// key is 400, an unknown role is 400, and re-adding an already-registered key is
+// 409 (the existing row is left untouched — no silent upsert).
+func TestUsersHTTP_Validation(t *testing.T) {
+	h := newAuthHarness(t, AuthEnforce)
+
+	good, _ := cs.GenerateIdentity()
+	goodPub := hex.EncodeToString(good.SignPub)
+
+	// Invalid hex (too short) -> 400.
+	if code, body := signedJSON(t, h, "POST", "/users",
+		addUserReq{SignPub: "abcd", Handle: "shorty", Role: RoleMember}, h.alice, 1); code != http.StatusBadRequest {
+		t.Fatalf("malformed sign_pub should be 400, got %d (%s)", code, body)
+	}
+
+	// Invalid role -> 400.
+	if code, body := signedJSON(t, h, "POST", "/users",
+		addUserReq{SignPub: goodPub, Handle: "weirdrole", Role: "superuser"}, h.alice, 2); code != http.StatusBadRequest {
+		t.Fatalf("invalid role should be 400, got %d (%s)", code, body)
+	}
+
+	// Re-adding the seeded admin's own key -> 409 (idempotency, no overwrite).
+	if code, body := signedJSON(t, h, "POST", "/users",
+		addUserReq{SignPub: h.alicePub, Handle: "alice-again", Role: RoleMember}, h.alice, 3); code != http.StatusConflict {
+		t.Fatalf("re-adding an existing key should be 409, got %d (%s)", code, body)
+	}
+	// And the existing row is untouched: alice is still an active admin.
+	u, err := h.store.GetUser(h.alicePub)
+	if err != nil {
+		t.Fatalf("get alice after conflicting re-add: %v", err)
+	}
+	if u.Role != RoleAdmin || u.Status != StatusActive || u.Handle != "alice" {
+		t.Fatalf("conflicting re-add mutated the existing row: %+v", u)
+	}
+}
+
+// listUsers signs a GET /users as alice and decodes the response.
+func listUsers(t *testing.T, h *authHarness, n int) []userJSON {
+	t.Helper()
+	code, body := signedJSON(t, h, "GET", "/users", nil, h.alice, n)
+	if code != http.StatusOK {
+		t.Fatalf("admin list users should be 200, got %d (%s)", code, body)
+	}
+	var users []userJSON
+	if err := json.Unmarshal([]byte(body), &users); err != nil {
+		t.Fatalf("decode users: %v (%s)", err, body)
+	}
+	return users
+}
+
+// findUser returns the row with the given signing key (case-insensitive).
+func findUser(users []userJSON, signPub string) (userJSON, bool) {
+	want := normalizeSignPub(signPub)
+	for _, u := range users {
+		if normalizeSignPub(u.SignPub) == want {
+			return u, true
+		}
+	}
+	return userJSON{}, false
+}