hermes: preserve SSO and image continuations

services/hermes/chat-configmap.yaml

@@ -126,6 +126,13 @@ data:
     the default posture for that request. Never exceed xhigh reasoning.
 
     When a user asks to create or edit an image, use an image generation tool.
+    Treat natural follow-ups such as "edit this", "turn it into", "make this
+    image", or a reference to the subject in the most recent generated image
+    as image-edit requests. For those turns, reuse the newest `MEDIA:` image
+    path in the conversation as `image_url`; do not answer with instructions,
+    route the request as ordinary text, or require the user to upload the image
+    again. Preserve the most recently selected image lane for an edit unless
+    the user explicitly requests local, OpenAI/hosted, or AUTO instead.
     Use `image_generate_local` when the request says local, private, on my
     hardware, or FLUX. Use `image_generate_hosted` when the request says
     OpenAI, hosted, GPT Image, or highest hosted quality. Otherwise use the
@@ -156,5 +163,7 @@ data:
     present a single final answer.
     Use the image generation tool for natural-language image creation and
     editing requests; generated images remain in this tenant's private cache.
+    Natural follow-ups that refer to the latest generated image must edit its
+    newest `MEDIA:` path rather than starting an unrelated text-only answer.
     Do not claim access to Kubernetes, Vault, Gitea, Brad's projects, other
     users, the agent coordinator, or automated triage.

services/hermes/chat-router.yaml

@@ -20,7 +20,7 @@ spec:
         app: hermes-chat-router
       annotations:
         ai.bstein.dev/role: privacy-preserving-chat-tenant-router
-        ai.bstein.dev/config-rev: "20260809-private-file-browser"
+        ai.bstein.dev/config-rev: "20260812-keycloak-image-continuation"
         vault.hashicorp.com/agent-inject: "true"
         vault.hashicorp.com/agent-pre-populate-only: "true"
         vault.hashicorp.com/agent-init-first: "true"
@@ -62,7 +62,7 @@ spec:
                     values: [rpi5]
       containers:
         - name: router
-          image: registry.bstein.dev/bstein/hermes-chat-router@sha256:882f9c3af7268618a19424765536f4cf6c37e8d380e515780baf4104e5b2931e
+          image: registry.bstein.dev/bstein/hermes-chat-router@sha256:72fbba10d108b1086e5620137f7549a838438ba510abb8cc8884bc592e7fd6bf
           imagePullPolicy: IfNotPresent
           ports:
             - {name: http, containerPort: 8080, protocol: TCP}

services/hermes/chat-statefulset.yaml

@@ -29,7 +29,7 @@ spec:
         ai.bstein.dev/router-wire-contract: ollama-numeric-keepalive
         ai.bstein.dev/isolation: one Hermes process and PVC per Keycloak subject
         ai.bstein.dev/model-policy: uniform automatic policy with per-user overrides
-        ai.bstein.dev/config-rev: "20260811-explicit-image-route-tools"
+        ai.bstein.dev/config-rev: "20260812-keycloak-image-continuation"
         vault.hashicorp.com/agent-inject: "true"
         vault.hashicorp.com/role: hermes-chat
         vault.hashicorp.com/agent-inject-secret-anthropic-token: kv/data/atlas/hermes/agent-tokens
@@ -265,6 +265,12 @@ spec:
             - {name: HERMES_WEBUI_GATEWAY_USE_RUNS_API, value: "true"}
             - {name: HERMES_WEBUI_SKIP_ONBOARDING, value: "1"}
             - {name: HERMES_WEBUI_SECURE, value: "1"}
+            - {name: HERMES_WEBUI_COOKIE_NAME, value: hermes_chat_session}
+            - {name: HERMES_WEBUI_PROFILE_COOKIE_NAME, value: hermes_chat_profile}
+            - {name: HERMES_WEBUI_TRUSTED_AUTH_HEADER, value: X-Hermes-Tenant-Identity}
+            # NetworkPolicy admits this port only from hermes-chat-router; the
+            # CIDR lets the WebUI validate that router's changing pod address.
+            - {name: HERMES_WEBUI_TRUSTED_PROXY_CIDRS, value: 10.42.0.0/16}
             - {name: HERMES_WEBUI_ALLOWED_ORIGINS, value: https://chat.hermes.bstein.dev}
             - {name: HERMES_WEBUI_TRUST_FORWARDED_HOST, value: "1"}
             - {name: HERMES_WEBUI_TRUST_FORWARDED_PROTO, value: "1"}

services/hermes/plugins/image-gen-broker/__init__.py

@@ -82,7 +82,10 @@ LOCAL_IMAGE_SCHEMA = {
         "FLUX, on my hardware, or otherwise explicitly rejects a hosted image "
         "provider. Never substitute the generic image_generate tool for an "
         "explicit local request. The backend is already provisioned; never ask "
-        "for an endpoint or tell the user to install Diffusers or ComfyUI."
+        "for an endpoint or tell the user to install Diffusers or ComfyUI. "
+        "For an edit follow-up, pass the newest MEDIA: path from the conversation "
+        "as image_url, even when the user refers to it only as this, it, the "
+        "image, or its pictured subject."
     ),
     "parameters": IMAGE_GENERATE_PARAMETERS,
 }
@@ -93,7 +96,10 @@ HOSTED_IMAGE_SCHEMA = {
         "Generate or edit an image only with hosted OpenAI GPT Image at the "
         "highest configured quality. Use this tool when the user explicitly "
         "asks for OpenAI, GPT Image, or hosted image generation. Do not use it "
-        "when the user explicitly requests local or private generation."
+        "when the user explicitly requests local or private generation. For an "
+        "edit follow-up, pass the newest MEDIA: path from the conversation as "
+        "image_url, even when the user refers to it only as this, it, the image, "
+        "or its pictured subject."
     ),
     "parameters": IMAGE_GENERATE_PARAMETERS,
 }

services/hermes/router/main.go

@@ -59,6 +59,12 @@ var deniedPrefixes = []string{
 
 const retiredServiceWorker = `self.addEventListener("install",()=>self.skipWaiting());self.addEventListener("activate",event=>event.waitUntil((async()=>{for(const key of await caches.keys())await caches.delete(key);await self.registration.unregister();for(const client of await self.clients.matchAll({type:"window"}))await client.navigate(client.url)})()));`
 
+const (
+	trustedTenantHeader = "X-Hermes-Tenant-Identity"
+	tenantSessionCookie = "hermes_chat_session"
+	tenantProfileCookie = "hermes_chat_profile"
+)
+
 func newTenantRouter(statePath string, slots int, backendURL func(int) string) (*tenantRouter, error) {
 	if slots < 1 {
 		return nil, fmt.Errorf("tenant slots must be positive")
@@ -247,6 +253,29 @@ func authenticatedSubject(request *http.Request) string {
 	return ""
 }
 
+func safeLoginNext(request *http.Request) string {
+	next := strings.TrimSpace(request.URL.Query().Get("next"))
+	if next == "" || len(next) > 2048 || !strings.HasPrefix(next, "/") ||
+		strings.HasPrefix(next, "//") || strings.ContainsAny(next, "\\\r\n") {
+		return "/"
+	}
+	parsed, err := url.ParseRequestURI(next)
+	if err != nil || parsed.IsAbs() || parsed.Host != "" || parsed.Path == "/login" || strings.HasPrefix(parsed.Path, "/login/") {
+		return "/"
+	}
+	return next
+}
+
+func preserveTenantCookies(request *http.Request) []*http.Cookie {
+	var preserved []*http.Cookie
+	for _, cookie := range request.Cookies() {
+		if cookie.Name == tenantSessionCookie || cookie.Name == tenantProfileCookie {
+			preserved = append(preserved, cookie)
+		}
+	}
+	return preserved
+}
+
 func privateWorkspaceReadAllowed(method, path string) bool {
 	if method != http.MethodGet {
 		return false
@@ -295,6 +324,13 @@ func (router *tenantRouter) ServeHTTP(writer http.ResponseWriter, request *http.
 		http.Error(writer, "authenticated identity required", http.StatusUnauthorized)
 		return
 	}
+	// Keycloak is the public login authority. Never expose the tenant WebUI's
+	// redundant password screen after a successful SSO handoff or when a user
+	// follows a stale /login bookmark.
+	if request.Method == http.MethodGet && request.URL.Path == "/login" {
+		http.Redirect(writer, request, safeLoginNext(request), http.StatusFound)
+		return
+	}
 	if router.serveTelegramWeb(writer, request, subject) {
 		return
 	}
@@ -321,6 +357,7 @@ func (router *tenantRouter) ServeHTTP(writer http.ResponseWriter, request *http.
 	originalDirector := proxy.Director
 	proxy.Director = func(outbound *http.Request) {
 		originalDirector(outbound)
+		cookies := preserveTenantCookies(outbound)
 		for _, header := range []string{
 			"Authorization", "Cookie", "X-Auth-Request-Access-Token",
 			"X-Auth-Request-Email", "X-Auth-Request-Groups", "X-Auth-Request-User",
@@ -329,6 +366,13 @@ func (router *tenantRouter) ServeHTTP(writer http.ResponseWriter, request *http.
 		} {
 			outbound.Header.Del(header)
 		}
+		// Each Keycloak subject is already mapped to exactly one isolated pod.
+		// Give that pod a non-sensitive, router-asserted identity while retaining
+		// only its own WebUI cookies; OAuth credentials never reach the backend.
+		outbound.Header.Set(trustedTenantHeader, fmt.Sprintf("slot-%d", slot))
+		for _, cookie := range cookies {
+			outbound.AddCookie(cookie)
+		}
 		outbound.Header.Del("Accept-Encoding")
 	}
 	proxy.ServeHTTP(writer, request)

services/hermes/router/main_test.go

@@ -96,8 +96,15 @@ func TestRouterRetiresStaleServiceWorkerWithoutIdentity(t *testing.T) {
 
 func TestRouterProxiesWebUIAndAddsTelegramShortcut(t *testing.T) {
 	backend := httptest.NewServer(http.HandlerFunc(func(writer http.ResponseWriter, request *http.Request) {
-		if request.Header.Get("X-Forwarded-User") != "" || request.Header.Get("Cookie") != "" {
-			t.Fatal("identity or session cookie leaked to tenant backend")
+		if request.Header.Get("X-Forwarded-User") != "" || request.Header.Get("X-Auth-Request-User") != "" {
+			t.Fatal("external identity header leaked to tenant backend")
+		}
+		if request.Header.Get(trustedTenantHeader) != "slot-0" {
+			t.Fatalf("trusted tenant identity was not asserted: %q", request.Header.Get(trustedTenantHeader))
+		}
+		cookies := request.Cookies()
+		if len(cookies) != 2 || cookies[0].Name != tenantSessionCookie || cookies[1].Name != tenantProfileCookie {
+			t.Fatalf("unexpected backend cookies: %#v", cookies)
 		}
 		writer.Header().Set("Content-Type", "text/html; charset=utf-8")
 		_, _ = io.WriteString(writer, "<html><head></head><body>Hermes WebUI</body></html>")
@@ -109,7 +116,8 @@ func TestRouterProxiesWebUIAndAddsTelegramShortcut(t *testing.T) {
 	}
 	request := httptest.NewRequest(http.MethodGet, "/", nil)
 	request.Header.Set("X-Forwarded-User", "subject")
-	request.Header.Set("Cookie", "oauth-cookie")
+	request.Header.Set("X-Auth-Request-User", "subject")
+	request.Header.Set("Cookie", "_oauth2_proxy=secret; "+tenantSessionCookie+"=session; "+tenantProfileCookie+"=default")
 	response := httptest.NewRecorder()
 	router.ServeHTTP(response, request)
 	if response.Code != http.StatusOK {
@@ -120,6 +128,35 @@ func TestRouterProxiesWebUIAndAddsTelegramShortcut(t *testing.T) {
 	}
 }
 
+func TestRouterRedirectsNativeLoginToSafeChatDestination(t *testing.T) {
+	router, err := newTenantRouter(filepath.Join(t.TempDir(), "state.json"), 1, func(slot int) string { return "" })
+	if err != nil {
+		t.Fatal(err)
+	}
+	for _, test := range []struct {
+		path string
+		want string
+	}{
+		{"/login?next=%2Fsession%2F9048e2a574d1", "/session/9048e2a574d1"},
+		{"/login?next=%2Fchat%3Fresume%3Dabc", "/chat?resume=abc"},
+		{"/login?next=https%3A%2F%2Fevil.example", "/"},
+		{"/login?next=%2F%2Fevil.example", "/"},
+		{"/login?next=%2Flogin", "/"},
+		{"/login?next=%2Flogin%2Fagain", "/"},
+	} {
+		request := httptest.NewRequest(http.MethodGet, test.path, nil)
+		request.Header.Set("X-Forwarded-User", "subject")
+		response := httptest.NewRecorder()
+		router.ServeHTTP(response, request)
+		if response.Code != http.StatusFound {
+			t.Fatalf("%s: got status %d", test.path, response.Code)
+		}
+		if location := response.Header().Get("Location"); location != test.want {
+			t.Fatalf("%s: got redirect %q, want %q", test.path, location, test.want)
+		}
+	}
+}
+
 func TestRouterAllowsTenantScopedPersonalization(t *testing.T) {
 	backend := httptest.NewServer(http.HandlerFunc(func(writer http.ResponseWriter, request *http.Request) {
 		writer.WriteHeader(http.StatusNoContent)

testing/tests/test_hermes_chat_quality.py

@@ -352,6 +352,8 @@ def test_chat_image_generation_uses_private_owner_broker():
     assert "Use `image_generate_local`" in configmap["data"]["SOUL.md"]
     assert "Use `image_generate_hosted`" in configmap["data"]["SOUL.md"]
     assert "ComfyUI endpoint" in configmap["data"]["SOUL.md"]
+    assert "newest `MEDIA:` image" in configmap["data"]["SOUL.md"]
+    assert "Preserve the most recently selected image lane" in configmap["data"]["SOUL.md"]
     config = yaml.safe_load(configmap["data"]["config.yaml"])
     assert config["image_gen"] == {
         "provider": "atlas-broker",
@@ -376,6 +378,7 @@ def test_chat_image_generation_uses_private_owner_broker():
     assert '"hosted": "gpt-image-2-high"' in plugin
     assert 'name="image_generate_local"' in plugin
     assert 'name="image_generate_hosted"' in plugin
+    assert "newest MEDIA: path from the conversation" in plugin
     assert "override=True" not in plugin
 
     agent = _documents(HERMES / "agent-deployment.yaml")[0]
@@ -481,7 +484,7 @@ def test_chat_reasoning_uses_switchyard_without_owner_credentials():
     statefulset = _documents(HERMES / "chat-statefulset.yaml")[0]
     assert statefulset["spec"]["template"]["metadata"]["annotations"][
         "ai.bstein.dev/config-rev"
-    ] == "20260811-explicit-image-route-tools"
+    ] == "20260812-keycloak-image-continuation"
     hermes = next(
         item
         for item in statefulset["spec"]["template"]["spec"]["containers"]