fix(gateway): AUTO 范围冲突的提示要放在 checkQuota，否则是死代码

上一版把消息加在请求路径的范围检查上，实测才发现它永远走不到：
AUTO 请求先经过 checkQuota(ctx, "AUTO")（在链被读取之前），范围不含 AUTO
时它当场返回旧文案「model \"AUTO\" is not allowed for this key」——那句会
把运维引去查配额，而配额并不是原因。

改为在 checkQuota 内对 AUTO 分流，输出说明真实原因与修法。请求路径那处
换成注释说明为何不再重复。

判据 1 条，断言 checkQuota 函数体内必须同时出现 isAuto(model) 与新文案
——放在别处不算数。变异（删掉该分支）导致编译失败，被捕获。

生产实测（临时把某密钥范围改成不含 AUTO 再恢复）：
  HTTP 403  model_not_allowed
  model \"AUTO\" is not in this key's model scope, so it cannot use AUTO;
  add \"AUTO\" to the key's models or remove the scope restriction

Co-Authored-By: ModelRouter <noreply@modelrouter.dev>

cmd/gui/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "modelrouter-gui",
-  "version": "1.9.2",
+  "version": "1.9.3",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "modelrouter-gui",
-      "version": "1.9.2",
+      "version": "1.9.3",
       "license": "Proprietary",
       "devDependencies": {
         "electron": "^33.0.0",

cmd/gui/package.json

@@ -1,6 +1,6 @@
 {
   "name": "modelrouter-gui",
-  "version": "1.9.2",
+  "version": "1.9.3",
   "description": "ModelRouter Desktop — embedded ModelRouter core with tray",
   "author": "ModelRouter",
   "main": "main.js",

internal/gateway/chat.go

@@ -256,7 +256,17 @@ func (g *Gateway) checkQuota(ctx context.Context, model string) *quotaRejection
 		}
 		return nil
 	}
-	return &quotaRejection{msg: fmt.Sprintf("model %q is not allowed for this key", model)}
+// A scope that simply does not list the model is a different problem from
+// a quota being exhausted, and the message has to say which: telling an
+// operator "not allowed for this key" when they configured a quota sends
+// them to the wrong setting. AUTO is checked here before the chain is even
+// consulted, so this is the message that surfaces for AUTO.
+if isAuto(model) {
+return &quotaRejection{msg: fmt.Sprintf(
+"model %q is not in this key's model scope, so it cannot use AUTO; "+
+"add %q to the key's models or remove the scope restriction", model, model)}
+}
+return &quotaRejection{msg: fmt.Sprintf("model %q is not allowed for this key", model)}
 }
 
 // quotaWindowSuffix describes a quota's reset window for an error message, so
@@ -452,6 +462,11 @@ func (g *Gateway) handleChat(w http.ResponseWriter, r *http.Request) {
 			return
 		}
 	}
+	// Note: a key whose scope excludes AUTO is rejected earlier, by
+	// checkQuota(ctx, "AUTO") above, which produces the same message. That
+	// ordering matters — checkQuota runs before the chain is consulted, so it
+	// is the check an AUTO request actually hits. Duplicating it here would be
+	// dead code.
 	cands, effective := g.resolveCands(r.Context(), &req)
 	if len(cands) == 0 {
 		writeError(w, http.StatusNotFound, "model_not_found", fmt.Sprintf("model %q is not configured", model))

internal/gateway/ui/index.html

@@ -517,7 +517,20 @@
         #toast{left:12px;right:12px;bottom:12px;text-align:center}
         th,td{padding:8px 10px}
       }
-    </style>
+    
+    /* Shown when a key's model scope blocks AUTO: the chain editor can look
+       fine while every request 403s, so the conflict must be visible here. */
+    .warn-box {
+      margin: 0 0 10px;
+      padding: 8px 10px;
+      border-radius: 8px;
+      border: 1px solid rgba(224, 108, 51, 0.45);
+      background: rgba(224, 108, 51, 0.1);
+      color: #b45309;
+      font-size: 12.5px;
+      line-height: 1.5;
+    }
+</style>
   </head>
   <body>
     <div id="bgfx" aria-hidden="true">
@@ -835,6 +848,7 @@
           kAutoChainSeeded: "已复制全局 AUTO 链作为编辑起点，保存后该密钥将使用这条独立链",
           kAutoChainNewOwn: "该密钥当前跟随全局链，保存后改用这里配置的独立链",
           kAutoChainWillInherit: "保存后该密钥将恢复跟随全局 AUTO 链",
+          kAutoChainScopeWarn: "警告：该密钥的模型范围不包含 AUTO，配了独立 AUTO 链也用不了。当前范围：",
           kAutoChainBad: "已保存，但槽位无法解析（模型与源的组合不存在），AUTO 请求会失败",
           kName: "名称",
           kRole: "角色",
@@ -1109,6 +1123,7 @@
           kAutoChainSeeded: "Copied the global AUTO chain as a starting point — saving gives this key its own chain",
           kAutoChainNewOwn: "This key currently follows the global chain; saving switches it to the chain configured here",
           kAutoChainWillInherit: "Saving makes this key follow the global AUTO chain again",
+          kAutoChainScopeWarn: "Warning: this key\u2019s model scope does not include AUTO, so a per-key AUTO chain will still be rejected. Current scope: ",
           kAutoChainBad: "Saved, but the slot does not resolve (no source serves that model); AUTO requests will fail",
           kName: "Name",
           kRole: "Role",
@@ -4811,8 +4826,11 @@ function afterChainEdit() {
       async function loadKeys() {
         const el = $("#k-list");
         if (!el) return;
-        const j = await api("/api/keys");
+const j = await api("/api/keys");
         const ks = j.keys || [];
+        // Cached so the per-key AUTO editor can read a key's model scope
+        // without a second request (keyAutoScopeWarning).
+        window._keyRows = ks;
         el.innerHTML = ks.length
           ? ks.map((k) => keyCanvasHtml(k)).join("")
           : `<div class="muted">${t("kEmpty")}</div>`;
@@ -5268,6 +5286,7 @@ function afterChainEdit() {
         wrap.innerHTML = `<div class="card" style="width:900px;max-width:100%">
     <h2>${esc(t("kAutoChainTitle"))} \u00b7 ${esc(name || key)}</h2>
     <div class="muted" style="margin-bottom:10px">${esc(t("kAutoChainHint"))}</div>
+    <div id="ka-scope-warn"></div>
     <div id="ka-canvas-host"></div>
     <p><button class="ghost small" onclick="scrAddModal()">+ ${esc(t("sortAdd"))}</button>
        <button class="ghost small" onclick="keyAutoInherit()">${esc(t("kAutoChainInherit"))}</button>
@@ -5309,6 +5328,7 @@ try {
             renderSortEditor($("#ka-canvas-host"), { canvasId: "key-auto-canvas" });
             keyAutoState.seeded = seededFromGlobal;
             keyAutoInheritNotice();
+            keyAutoScopeWarning(key);
           } catch (e) {
             toast(String(e));
           }
@@ -5364,6 +5384,31 @@ try {
         paintSort();
         keyAutoInheritNotice();
       }
+      // keyAutoScopeWarning tells the admin when the key's model scope blocks
+      // AUTO outright. Configuring a per-key AUTO chain for such a key looks
+      // like it works, yet every request 403s — the scope filter runs before
+      // the chain is consulted. We deliberately do NOT widen the scope
+      // automatically: that would silently grant access nobody asked for.
+      function keyAutoScopeWarning(key) {
+        const host = $("#ka-scope-warn");
+        if (!host) return;
+        const rec = (window._keyRows || []).find((k) => k.key === key);
+        const models = (rec && rec.models) || [];
+        if (!models.length) {
+          host.innerHTML = "";
+          return;
+        }
+        const hasAuto = models.some(
+          (m) => (m.model || "").toUpperCase() === "AUTO",
+        );
+        host.innerHTML = hasAuto
+          ? ""
+          : '<div class="warn-box">' +
+            esc(t("kAutoChainScopeWarn")) +
+            " " +
+            esc(models.map((m) => m.model).join(", ")) +
+            "</div>";
+      }
       function keyAutoClose() {
         const m = document.getElementById("key-auto-modal");
         if (m) m.remove();

internal/gateway/ui_key_auto_test.go

@@ -1,6 +1,7 @@
 package gateway
 
 import (
+	"os"
 	"strings"
 	"testing"
 )
@@ -329,3 +330,66 @@ func TestUIKeyAutoNoticeDistinguishesSavedFromCopied(t *testing.T) {
 		}
 	}
 }
+
+// A key whose model scope excludes AUTO cannot use AUTO at all — the scope
+// filter runs before the chain — yet the editor happily lets an admin
+// configure a per-key chain. That combination is easy to create by accident
+// and produces a 403 with no visible cause, so the dialog must say so.
+//
+// We deliberately do NOT widen the scope automatically: silently granting a
+// model the operator did not ask for is worse than a loud warning.
+func TestUIKeyAutoWarnsOnScopeConflict(t *testing.T) {
+	src := uiSource(t)
+	if !strings.Contains(src, "function keyAutoScopeWarning(") {
+		t.Fatal("no scope-conflict check in the per-key editor")
+	}
+	body := readFuncBody(t, src, "keyAutoScopeWarning")
+	if !strings.Contains(body, "AUTO") {
+		t.Fatal("keyAutoScopeWarning must test whether the scope lists AUTO")
+	}
+	// It must not write anything back to the key: reporting only.
+	for _, forbidden := range []string{"PUT", "POST", "fetch(", "api("} {
+		if strings.Contains(body, forbidden) {
+			t.Fatalf("keyAutoScopeWarning calls %s — it must only report the "+
+				"conflict, never widen the scope itself", forbidden)
+		}
+	}
+	// Empty scope means unrestricted, so no warning is correct there.
+	if !strings.Contains(body, "if (!models.length)") {
+		t.Fatal("an unrestricted key (no models) must not be warned")
+	}
+	// Both languages.
+	if n := strings.Count(src, "kAutoChainScopeWarn"); n < 2 {
+		t.Fatalf("i18n key kAutoChainScopeWarn appears %d time(s), want zh+en", n)
+	}
+}
+
+// The scope conflict for AUTO must be reported by checkQuota, because that is
+// the check an AUTO request actually reaches first (it runs before the chain is
+// consulted). Putting the message on the later path made it dead code — the
+// request was rejected one check earlier with the generic "not allowed for this
+// key" wording, which sends an operator looking for a quota that isn't the
+// problem.
+func TestAutoScopeConflictMessageComesFromCheckQuota(t *testing.T) {
+	b, err := os.ReadFile("chat.go")
+	if err != nil {
+		t.Fatalf("read chat.go: %v", err)
+	}
+	src := string(b)
+	i := strings.Index(src, "func (g *Gateway) checkQuota(")
+	if i < 0 {
+		t.Fatal("checkQuota not found")
+	}
+	// Body = up to the next top-level func.
+	rest := src[i:]
+	if j := strings.Index(rest[1:], "\nfunc "); j > 0 {
+		rest = rest[:j+1]
+	}
+	if !strings.Contains(rest, "isAuto(model)") {
+		t.Fatal("checkQuota must special-case AUTO: a scope that omits AUTO is " +
+			"rejected there before anything else looks at the chain")
+	}
+	if !strings.Contains(rest, "not in this key's model scope") {
+		t.Fatal("checkQuota's AUTO branch must name the scope as the cause")
+	}
+}