Refactor AssistantPage and DebugVoicePanel for improved view management

- Remove the SegmentedIconGroup for view selection and streamline view handling in DebugDrawer.
- Introduce DebugVisionWorkspace component to manage video and chat views more effectively.
- Update DebugVoicePanel to accept onViewChange prop for better view state management.
- Enhance video handling with dynamic positioning and drag functionality for the floating video panel.

frontend/src/components/pages/AssistantPage.tsx

@@ -2243,22 +2243,6 @@ function DebugDrawer({
               </SegmentedIconGroup>
             </>
           )}
-          <SegmentedIconGroup label="预览视图">
-            <SegmentedIconButton
-              selected={view === "chat"}
-              label="聊天记录"
-              onClick={() => setView("chat")}
-            >
-              <MessageSquareText size={14} />
-            </SegmentedIconButton>
-            <SegmentedIconButton
-              selected={view === "video"}
-              label="视频流"
-              onClick={() => setView("video")}
-            >
-              <Video size={14} />
-            </SegmentedIconButton>
-          </SegmentedIconGroup>
         </div>
       </div>
       <div className="shrink-0 border-b border-hairline px-5 py-2.5">
@@ -2268,6 +2252,7 @@ function DebugDrawer({
       </div>
       <DebugVoicePanel
         view={view}
+        onViewChange={setView}
         showTranscript={showTranscript}
         vizStyle={vizStyle}
         assistantId={assistantId}
@@ -2596,6 +2581,7 @@ function DebugInputModeButton({
 
 function DebugVoicePanel({
   view,
+  onViewChange,
   showTranscript,
   vizStyle,
   assistantId,
@@ -2607,6 +2593,7 @@ function DebugVoicePanel({
   dynamicVariablesError,
 }: {
   view: DebugView;
+  onViewChange: (view: DebugView) => void;
   showTranscript: boolean;
   vizStyle: VizStyle;
   assistantId: string | null;
@@ -2672,7 +2659,16 @@ function DebugVoicePanel({
     <div className="flex min-h-0 flex-1 flex-col">
       {/* 后端 TTS 音频经 WebRTC 媒体流过来,挂这里播放 */}
       <audio ref={audioRef} autoPlay playsInline className="hidden" />
-      {view === "video" ? (
+      {vision && !showIdleHub ? (
+        <DebugVisionWorkspace
+          view={view}
+          onViewChange={onViewChange}
+          messages={messages}
+          recording={recording}
+          camera={camera}
+          videoStream={preview.videoStream}
+        />
+      ) : view === "video" ? (
         showIdleHub ? (
           <DebugIdleHub
             status={status}
@@ -2941,12 +2937,133 @@ function DebugIdleHub({
   );
 }
 
+function DebugVisionWorkspace({
+  view,
+  onViewChange,
+  messages,
+  recording,
+  camera,
+  videoStream,
+}: {
+  view: DebugView;
+  onViewChange: (view: DebugView) => void;
+  messages: ChatMessage[];
+  recording: boolean;
+  camera: CameraPreview;
+  videoStream: MediaStream | null;
+}) {
+  const containerRef = useRef<HTMLDivElement>(null);
+  const [position, setPosition] = useState({ x: 16, y: 16 });
+  const positionedRef = useRef(false);
+  const dragRef = useRef<{
+    pointerId: number;
+    startX: number;
+    startY: number;
+    originX: number;
+    originY: number;
+    moved: boolean;
+  } | null>(null);
+  const latestMessage = [...messages]
+    .reverse()
+    .find((message) => message.content.trim());
+
+  useEffect(() => {
+    const container = containerRef.current;
+    if (!container) return;
+    const rect = container.getBoundingClientRect();
+    if (!positionedRef.current) {
+      positionedRef.current = true;
+      setPosition({ x: Math.max(16, rect.width - 192), y: 16 });
+      return;
+    }
+    setPosition((current) => ({
+      x: Math.min(Math.max(16, current.x), Math.max(16, rect.width - 192)),
+      y: Math.min(Math.max(16, current.y), Math.max(16, rect.height - 116)),
+    }));
+  }, [view]);
+
+  const moveFloatingVideo = useCallback((event: React.PointerEvent<HTMLButtonElement>) => {
+    const drag = dragRef.current;
+    const container = containerRef.current;
+    if (!drag || !container || drag.pointerId !== event.pointerId) return;
+    const dx = event.clientX - drag.startX;
+    const dy = event.clientY - drag.startY;
+    if (Math.abs(dx) + Math.abs(dy) > 4) drag.moved = true;
+    const rect = container.getBoundingClientRect();
+    setPosition({
+      x: Math.min(Math.max(16, drag.originX + dx), Math.max(16, rect.width - 192)),
+      y: Math.min(Math.max(16, drag.originY + dy), Math.max(16, rect.height - 116)),
+    });
+  }, []);
+
+  if (view === "video") {
+    return (
+      <div ref={containerRef} className="relative flex min-h-0 flex-1 overflow-hidden bg-black">
+        <DebugVideoPanel camera={camera} streamOverride={videoStream} />
+        <button
+          type="button"
+          onClick={() => onViewChange("chat")}
+          className="absolute left-4 right-4 top-4 z-10 flex h-16 items-center gap-3 overflow-hidden rounded-2xl border border-white/15 bg-[#07101a]/55 px-4 py-2 text-left text-white shadow-lg backdrop-blur-md transition-colors hover:bg-[#07101a]/70 focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-white/40"
+          aria-label="切换到聊天记录"
+        >
+          <MessageSquareText size={17} className="shrink-0 text-white/70" />
+          <span className="line-clamp-2 min-w-0 flex-1 text-xs leading-5 text-white/90">
+            <span className="mr-1.5 font-medium text-white/55">
+              {latestMessage?.role === "user" ? "我：" : "助手："}
+            </span>
+            <span>
+              {latestMessage?.content || "暂无消息，点击返回聊天记录"}
+            </span>
+          </span>
+        </button>
+      </div>
+    );
+  }
+
+  return (
+    <div ref={containerRef} className="relative flex min-h-0 flex-1 overflow-hidden">
+      <DebugTranscriptPanel messages={messages} recording={recording} />
+      <button
+        type="button"
+        style={{ left: position.x, top: position.y }}
+        className="absolute z-10 h-[100px] w-44 touch-none overflow-hidden rounded-2xl border border-white/20 bg-black text-left shadow-xl ring-1 ring-black/10 transition-shadow hover:shadow-2xl focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-primary"
+        aria-label="拖动摄像头浮窗，点击切换到视频流"
+        onPointerDown={(event) => {
+          event.currentTarget.setPointerCapture(event.pointerId);
+          dragRef.current = {
+            pointerId: event.pointerId,
+            startX: event.clientX,
+            startY: event.clientY,
+            originX: position.x,
+            originY: position.y,
+            moved: false,
+          };
+        }}
+        onPointerMove={moveFloatingVideo}
+        onPointerUp={(event) => {
+          const drag = dragRef.current;
+          event.currentTarget.releasePointerCapture(event.pointerId);
+          dragRef.current = null;
+          if (drag && !drag.moved) onViewChange("video");
+        }}
+        onPointerCancel={() => {
+          dragRef.current = null;
+        }}
+      >
+        <DebugVideoPanel camera={camera} streamOverride={videoStream} compact />
+      </button>
+    </div>
+  );
+}
+
 function DebugVideoPanel({
   camera,
   streamOverride,
+  compact = false,
 }: {
   camera: CameraPreview;
   streamOverride?: MediaStream | null;
+  compact?: boolean;
 }) {
   const { stream, error, starting, active } = camera;
   const effectiveStream = streamOverride ?? stream;
@@ -2961,6 +3078,7 @@ function DebugVideoPanel({
     <div
       className={[
         "relative flex min-h-0 flex-1 items-center justify-center overflow-hidden",
+        compact ? "h-full w-full" : "",
         effectiveActive ? "bg-black" : "bg-canvas-soft",
       ].join(" ")}
     >
@@ -2970,19 +3088,12 @@ function DebugVideoPanel({
         playsInline
         muted
         className={[
-          "h-full w-full -scale-x-100 object-contain",
+          "h-full w-full -scale-x-100",
+          compact ? "object-cover" : "object-contain",
           effectiveActive ? "" : "hidden",
         ].join(" ")}
       />
-      {effectiveActive ? (
-        <Badge
-          variant="secondary"
-          className="absolute left-3 top-3 gap-1.5 rounded-full border border-hairline bg-card/80 px-2.5 py-1 text-[11px] font-medium text-muted-foreground shadow-none backdrop-blur"
-        >
-          <span className="h-1.5 w-1.5 animate-pulse rounded-full bg-success" />
-          预览中
-        </Badge>
-      ) : (
+      {!effectiveActive ? (
         <div className="flex max-w-xs flex-col items-center gap-2 px-6 text-center">
           <Video size={28} className="text-muted-soft" />
           <div className="text-sm font-medium text-foreground">
@@ -2992,7 +3103,7 @@ function DebugVideoPanel({
             <p className="text-xs leading-5 text-muted-foreground">{error}</p>
           )}
         </div>
-      )}
+      ) : null}
     </div>
   );
 }