mobile-mcp-ai 2.6.2__py3-none-any.whl → 2.6.4__py3-none-any.whl

mobile_mcp/core/basic_tools_lite.py

@@ -8,6 +8,7 @@
 - 核心功能精简
 - 保留 pytest 脚本生成
 - 支持操作历史记录
+- Token 优化模式（省钱）
 """
 
 import asyncio
@@ -17,6 +18,19 @@ from pathlib import Path
 from typing import Dict, List, Optional
 from datetime import datetime
 
+# Token 优化配置（只精简格式，不限制数量，确保准确度）
+try:
+    from mobile_mcp.config import Config
+    TOKEN_OPTIMIZATION = Config.TOKEN_OPTIMIZATION_ENABLED
+    MAX_ELEMENTS = Config.MAX_ELEMENTS_RETURN
+    MAX_SOM_ELEMENTS = Config.MAX_SOM_ELEMENTS_RETURN
+    COMPACT_RESPONSE = Config.COMPACT_RESPONSE
+except ImportError:
+    TOKEN_OPTIMIZATION = True
+    MAX_ELEMENTS = 0  # 0 = 不限制
+    MAX_SOM_ELEMENTS = 0  # 0 = 不限制
+    COMPACT_RESPONSE = True
+
 
 class BasicMobileToolsLite:
     """精简版移动端工具"""
@@ -48,7 +62,7 @@ class BasicMobileToolsLite:
         return None
     
     def _record_operation(self, action: str, **kwargs):
-        """记录操作到历史"""
+        """记录操作到历史（旧接口，保持兼容）"""
         record = {
             'action': action,
             'timestamp': datetime.now().isoformat(),
@@ -56,6 +70,81 @@ class BasicMobileToolsLite:
         }
         self.operation_history.append(record)
     
+    def _record_click(self, locator_type: str, locator_value: str, 
+                      x_percent: float = 0, y_percent: float = 0,
+                      element_desc: str = '', locator_attr: str = ''):
+        """记录点击操作（标准格式）
+        
+        Args:
+            locator_type: 定位类型 'text' | 'id' | 'percent' | 'coords'
+            locator_value: 定位值（文本内容、resource-id、或坐标描述）
+            x_percent: 百分比 X 坐标（兜底方案）
+            y_percent: 百分比 Y 坐标（兜底方案）
+            element_desc: 元素描述（用于脚本注释）
+            locator_attr: Android 选择器属性 'text'|'textContains'|'description'|'descriptionContains'
+        """
+        record = {
+            'action': 'click',
+            'timestamp': datetime.now().isoformat(),
+            'locator_type': locator_type,
+            'locator_value': locator_value,
+            'locator_attr': locator_attr or locator_type,  # 默认与 type 相同
+            'x_percent': x_percent,
+            'y_percent': y_percent,
+            'element_desc': element_desc or locator_value,
+        }
+        self.operation_history.append(record)
+    
+    def _record_long_press(self, locator_type: str, locator_value: str,
+                           duration: float = 1.0,
+                           x_percent: float = 0, y_percent: float = 0,
+                           element_desc: str = '', locator_attr: str = ''):
+        """记录长按操作（标准格式）"""
+        record = {
+            'action': 'long_press',
+            'timestamp': datetime.now().isoformat(),
+            'locator_type': locator_type,
+            'locator_value': locator_value,
+            'locator_attr': locator_attr or locator_type,
+            'duration': duration,
+            'x_percent': x_percent,
+            'y_percent': y_percent,
+            'element_desc': element_desc or locator_value,
+        }
+        self.operation_history.append(record)
+    
+    def _record_input(self, text: str, locator_type: str = '', locator_value: str = '',
+                      x_percent: float = 0, y_percent: float = 0):
+        """记录输入操作（标准格式）"""
+        record = {
+            'action': 'input',
+            'timestamp': datetime.now().isoformat(),
+            'text': text,
+            'locator_type': locator_type,
+            'locator_value': locator_value,
+            'x_percent': x_percent,
+            'y_percent': y_percent,
+        }
+        self.operation_history.append(record)
+    
+    def _record_swipe(self, direction: str):
+        """记录滑动操作"""
+        record = {
+            'action': 'swipe',
+            'timestamp': datetime.now().isoformat(),
+            'direction': direction,
+        }
+        self.operation_history.append(record)
+    
+    def _record_key(self, key: str):
+        """记录按键操作"""
+        record = {
+            'action': 'press_key',
+            'timestamp': datetime.now().isoformat(),
+            'key': key,
+        }
+        self.operation_history.append(record)
+    
     def _get_current_package(self) -> Optional[str]:
         """获取当前前台应用的包名/Bundle ID"""
         try:
@@ -260,7 +349,7 @@ class BasicMobileToolsLite:
                     size = ios_client.wda.window_size()
                     screen_width, screen_height = size[0], size[1]
                 else:
-                    return {"success": False, "message": "❌ iOS 客户端未初始化"}
+                    return {"success": False, "msg": "iOS未初始化"}
             else:
                 self.client.u2.screenshot(str(temp_path))
                 info = self.client.u2.info
@@ -311,22 +400,14 @@ class BasicMobileToolsLite:
                 
                 cropped_size = final_path.stat().st_size
                 
+                # 返回结果
                 return {
                     "success": True,
                     "screenshot_path": str(final_path),
-                    "screen_width": screen_width,
-                    "screen_height": screen_height,
                     "image_width": img.width,
                     "image_height": img.height,
                     "crop_offset_x": crop_offset_x,
-                    "crop_offset_y": crop_offset_y,
-                    "file_size": f"{cropped_size/1024:.1f}KB",
-                    "message": f"🔍 局部截图已保存: {final_path}\n"
-                              f"📐 裁剪区域: ({crop_offset_x}, {crop_offset_y}) 起，{img.width}x{img.height} 像素\n"
-                              f"📦 文件大小: {cropped_size/1024:.0f}KB\n"
-                              f"🎯 【坐标换算】AI 返回坐标 (x, y) 后：\n"
-                              f"   实际屏幕坐标 = ({crop_offset_x} + x, {crop_offset_y} + y)\n"
-                              f"   或直接调用 mobile_click_at_coords(x, y, crop_offset_x={crop_offset_x}, crop_offset_y={crop_offset_y})"
+                    "crop_offset_y": crop_offset_y
                 }
             
             # ========== 情况2：全屏压缩截图 ==========
@@ -379,24 +460,14 @@ class BasicMobileToolsLite:
                 compressed_size = final_path.stat().st_size
                 saved_percent = (1 - compressed_size / original_size) * 100
                 
+                # 返回结果
                 return {
                     "success": True,
                     "screenshot_path": str(final_path),
-                    "screen_width": screen_width,
-                    "screen_height": screen_height,
-                    "original_img_width": original_img_width,    # 截图原始宽度
-                    "original_img_height": original_img_height,  # 截图原始高度
-                    "image_width": image_width,                  # 压缩后宽度（AI 看到的）
-                    "image_height": image_height,                # 压缩后高度（AI 看到的）
-                    "original_size": f"{original_size/1024:.1f}KB",
-                    "compressed_size": f"{compressed_size/1024:.1f}KB",
-                    "saved_percent": f"{saved_percent:.0f}%",
-                    "message": f"📸 截图已保存: {final_path}\n"
-                              f"📐 原始尺寸: {original_img_width}x{original_img_height} → 压缩后: {image_width}x{image_height}\n"
-                              f"📦 已压缩: {original_size/1024:.0f}KB → {compressed_size/1024:.0f}KB (省 {saved_percent:.0f}%)\n"
-                              f"⚠️ 【坐标转换】AI 返回坐标后，请传入：\n"
-                              f"   image_width={image_width}, image_height={image_height},\n"
-                              f"   original_img_width={original_img_width}, original_img_height={original_img_height}"
+                    "image_width": image_width,
+                    "image_height": image_height,
+                    "original_img_width": original_img_width,
+                    "original_img_height": original_img_height
                 }
             
             # ========== 情况3：全屏不压缩截图 ==========
@@ -410,21 +481,12 @@ class BasicMobileToolsLite:
                 final_path = self.screenshot_dir / filename
                 temp_path.rename(final_path)
                 
-                # 不压缩时，用截图实际尺寸（可能和 screen_width 不同）
+                # 返回结果（不压缩时尺寸相同）
                 return {
                     "success": True,
                     "screenshot_path": str(final_path),
-                    "screen_width": screen_width,
-                    "screen_height": screen_height,
-                    "original_img_width": img.width,   # 截图实际尺寸
-                    "original_img_height": img.height,
-                    "image_width": img.width,          # 未压缩，和原图一样
-                    "image_height": img.height,
-                    "file_size": f"{original_size/1024:.1f}KB",
-                    "message": f"📸 截图已保存: {final_path}\n"
-                              f"📐 截图尺寸: {img.width}x{img.height}\n"
-                              f"📦 文件大小: {original_size/1024:.0f}KB（未压缩）\n"
-                              f"💡 未压缩，坐标可直接使用"
+                    "image_width": img.width,
+                    "image_height": img.height
                 }
         except ImportError:
             # 如果没有 PIL，回退到原始方式（不压缩）
@@ -432,7 +494,7 @@ class BasicMobileToolsLite:
         except Exception as e:
             return {"success": False, "message": f"❌ 截图失败: {e}"}
     
-    def take_screenshot_with_grid(self, grid_size: int = 100, show_popup_hints: bool = True) -> Dict:
+    def take_screenshot_with_grid(self, grid_size: int = 100, show_popup_hints: bool = False) -> Dict:
         """截图并添加网格坐标标注（用于精确定位元素）
         
         在截图上绘制网格线和坐标刻度，帮助快速定位元素位置。
@@ -464,7 +526,7 @@ class BasicMobileToolsLite:
                     size = ios_client.wda.window_size()
                     screen_width, screen_height = size[0], size[1]
                 else:
-                    return {"success": False, "message": "❌ iOS 客户端未初始化"}
+                    return {"success": False, "msg": "iOS未初始化"}
             else:
                 self.client.u2.screenshot(str(temp_path))
                 info = self.client.u2.info
@@ -500,7 +562,7 @@ class BasicMobileToolsLite:
                 # 左侧标注 Y 坐标
                 draw.text((2, y + 2), str(y), fill=text_color, font=font_small)
             
-            # 第3步：检测弹窗并标注
+            # 第3步：检测弹窗并标注（使用严格的置信度检测，避免误识别）
             popup_info = None
             close_positions = []
             
@@ -510,35 +572,12 @@ class BasicMobileToolsLite:
                     xml_string = self.client.u2.dump_hierarchy(compressed=False)
                     root = ET.fromstring(xml_string)
                     
-                    # 检测弹窗区域
-                    popup_bounds = None
-                    for elem in root.iter():
-                        bounds_str = elem.attrib.get('bounds', '')
-                        class_name = elem.attrib.get('class', '')
-                        
-                        if not bounds_str:
-                            continue
-                        
-                        match = re.match(r'\[(\d+),(\d+)\]\[(\d+),(\d+)\]', bounds_str)
-                        if not match:
-                            continue
-                        
-                        x1, y1, x2, y2 = map(int, match.groups())
-                        width = x2 - x1
-                        height = y2 - y1
-                        area = width * height
-                        screen_area = screen_width * screen_height
-                        
-                        is_container = any(kw in class_name for kw in ['Layout', 'View', 'Dialog', 'Card'])
-                        area_ratio = area / screen_area if screen_area > 0 else 0
-                        is_not_fullscreen = (width < screen_width * 0.98 or height < screen_height * 0.98)
-                        is_reasonable_size = 0.08 < area_ratio < 0.85
-                        
-                        if is_container and is_not_fullscreen and is_reasonable_size and y1 > 50:
-                            if popup_bounds is None or area > (popup_bounds[2] - popup_bounds[0]) * (popup_bounds[3] - popup_bounds[1]):
-                                popup_bounds = (x1, y1, x2, y2)
+                    # 使用严格的弹窗检测（置信度 >= 0.6 才认为是弹窗）
+                    popup_bounds, popup_confidence = self._detect_popup_with_confidence(
+                        root, screen_width, screen_height
+                    )
                     
-                    if popup_bounds:
+                    if popup_bounds and popup_confidence >= 0.6:
                         px1, py1, px2, py2 = popup_bounds
                         popup_width = px2 - px1
                         popup_height = py2 - py1
@@ -601,26 +640,16 @@ class BasicMobileToolsLite:
             result = {
                 "success": True,
                 "screenshot_path": str(final_path),
-                "screen_width": screen_width,
-                "screen_height": screen_height,
                 "image_width": img_width,
                 "image_height": img_height,
-                "grid_size": grid_size,
-                "message": f"📸 网格截图已保存: {final_path}\n"
-                          f"📐 尺寸: {img_width}x{img_height}\n"
-                          f"📏 网格间距: {grid_size}px"
+                "grid_size": grid_size
             }
             
             if popup_info:
-                result["popup_detected"] = True
-                result["popup_bounds"] = popup_info["bounds"]
-                result["close_button_hints"] = close_positions
-                result["message"] += f"\n🎯 检测到弹窗: {popup_info['bounds']}"
-                result["message"] += f"\n💡 可能的关闭按钮位置（绿色圆圈标注）："
-                for pos in close_positions:
-                    result["message"] += f"\n   {pos['priority']}. {pos['name']}: ({pos['x']}, {pos['y']})"
-            else:
-                result["popup_detected"] = False
+                result["popup"] = popup_info["bounds"]
+                # 只返回前3个最可能的关闭按钮位置
+                if close_positions:
+                    result["close_hints"] = [(p['x'], p['y']) for p in close_positions[:3]]
             
             return result
             
@@ -657,7 +686,7 @@ class BasicMobileToolsLite:
                     size = ios_client.wda.window_size()
                     screen_width, screen_height = size[0], size[1]
                 else:
-                    return {"success": False, "message": "❌ iOS 客户端未初始化"}
+                    return {"success": False, "msg": "iOS未初始化"}
             else:
                 self.client.u2.screenshot(str(temp_path))
                 info = self.client.u2.info
@@ -766,52 +795,24 @@ class BasicMobileToolsLite:
                     'index': i + 1,
                     'center': (cx, cy),
                     'bounds': f"[{x1},{y1}][{x2},{y2}]",
-                    'desc': elem['desc']
+                    'desc': elem['desc'],
+                    'text': elem.get('text', ''),
+                    'resource_id': elem.get('resource_id', '')
                 })
             
-            # 第3.5步：检测弹窗区域（用于标注）
+            # 第3.5步：检测弹窗区域（使用严格的置信度检测，避免误识别普通页面）
             popup_bounds = None
+            popup_confidence = 0
             
             if not self._is_ios():
                 try:
-                    # 检测弹窗区域
-                    for elem in root.iter():
-                        bounds_str = elem.attrib.get('bounds', '')
-                        class_name = elem.attrib.get('class', '')
-                        
-                        if not bounds_str:
-                            continue
-                        
-                        match = re.match(r'\[(\d+),(\d+)\]\[(\d+),(\d+)\]', bounds_str)
-                        if not match:
-                            continue
-                        
-                        px1, py1, px2, py2 = map(int, match.groups())
-                        p_width = px2 - px1
-                        p_height = py2 - py1
-                        p_area = p_width * p_height
-                        screen_area = screen_width * screen_height
-                        
-                        is_container = any(kw in class_name for kw in ['Layout', 'View', 'Dialog', 'Card', 'Frame'])
-                        area_ratio = p_area / screen_area if screen_area > 0 else 0
-                        
-                        # 弹窗特征判断（更严格，排除主要内容区域）：
-                        # 1. 不是全屏（宽度和高度都要小于屏幕的95%）
-                        is_not_fullscreen = (p_width < screen_width * 0.95 and p_height < screen_height * 0.95)
-                        # 2. 面积范围：10% - 70%（排除主要内容区域，通常占80%+）
-                        is_reasonable_size = 0.10 < area_ratio < 0.70
-                        # 3. 不在屏幕左边缘（排除从x=0开始的主要内容容器）
-                        is_not_at_left_edge = px1 > screen_width * 0.05
-                        # 4. 高度不能占据屏幕的大部分（排除主要内容区域）
-                        height_ratio = p_height / screen_height if screen_height > 0 else 0
-                        is_not_main_content = height_ratio < 0.85
-                        
-                        if is_container and is_not_fullscreen and is_reasonable_size and is_not_at_left_edge and is_not_main_content and py1 > 30:
-                            if popup_bounds is None or p_area > (popup_bounds[2] - popup_bounds[0]) * (popup_bounds[3] - popup_bounds[1]):
-                                popup_bounds = (px1, py1, px2, py2)
+                    # 使用严格的弹窗检测（置信度 >= 0.6 才认为是弹窗）
+                    popup_bounds, popup_confidence = self._detect_popup_with_confidence(
+                        root, screen_width, screen_height
+                    )
                     
                     # 如果检测到弹窗，标注弹窗边界（不再猜测X按钮位置）
-                    if popup_bounds:
+                    if popup_bounds and popup_confidence >= 0.6:
                         px1, py1, px2, py2 = popup_bounds
                         
                         # 只画弹窗边框（蓝色），不再猜测X按钮位置
@@ -845,38 +846,15 @@ class BasicMobileToolsLite:
             img.save(str(final_path), "JPEG", quality=85)
             temp_path.unlink()
             
-            # 构建元素列表文字
-            elements_text = "\n".join([
-                f"  [{e['index']}] {e['desc']} → ({e['center'][0]}, {e['center'][1]})"
-                for e in som_elements[:15]  # 只显示前15个
-            ])
-            if len(som_elements) > 15:
-                elements_text += f"\n  ... 还有 {len(som_elements) - 15} 个元素"
-            
-            # 构建弹窗提示文字
-            hints_text = ""
-            if popup_bounds:
-                hints_text = f"\n🎯 检测到弹窗区域（蓝色边框）\n"
-                hints_text += f"   如需关闭弹窗，请观察图片中的 X 按钮位置\n"
-                hints_text += f"   然后使用 mobile_click_by_percent(x%, y%) 点击"
-            
+            # 返回结果（Token 优化：不返回 elements 列表，已存储在 self._som_elements）
             return {
                 "success": True,
                 "screenshot_path": str(final_path),
                 "screen_width": screen_width,
                 "screen_height": screen_height,
-                "image_width": img_width,
-                "image_height": img_height,
                 "element_count": len(som_elements),
-                "elements": som_elements,
                 "popup_detected": popup_bounds is not None,
-                "popup_bounds": f"[{popup_bounds[0]},{popup_bounds[1]}][{popup_bounds[2]},{popup_bounds[3]}]" if popup_bounds else None,
-                "message": f"📸 SoM 截图已保存: {final_path}\n"
-                          f"🏷️ 已标注 {len(som_elements)} 个可点击元素\n"
-                          f"📋 元素列表：\n{elements_text}{hints_text}\n\n"
-                          f"💡 使用方法：\n"
-                          f"   - 点击标注元素：mobile_click_by_som(编号)\n"
-                          f"   - 点击任意位置：mobile_click_by_percent(x%, y%)"
+                "hint": "查看截图上的编号，用 click_by_som(编号) 点击"
             }
             
         except ImportError:
@@ -922,14 +900,41 @@ class BasicMobileToolsLite:
                 ios_client = self._get_ios_client()
                 if ios_client and hasattr(ios_client, 'wda'):
                     ios_client.wda.click(cx, cy)
+                    size = ios_client.wda.window_size()
+                    screen_width, screen_height = size[0], size[1]
             else:
                 self.client.u2.click(cx, cy)
-            
+                info = self.client.u2.info
+                screen_width = info.get('displayWidth', 0)
+                screen_height = info.get('displayHeight', 0)
+
             time.sleep(0.3)
             
+            # 计算百分比坐标用于跨设备兼容
+            x_percent = round(cx / screen_width * 100, 1) if screen_width > 0 else 0
+            y_percent = round(cy / screen_height * 100, 1) if screen_height > 0 else 0
+            
+            # 使用标准记录格式
+            # 优先使用元素的文本/描述信息，这样生成脚本时可以用文本定位
+            elem_text = target.get('text', '')
+            elem_id = target.get('resource_id', '')
+            elem_desc = target.get('desc', '')
+            
+            if elem_text and not elem_text.startswith('['):  # 排除类似 "[可点击]" 的描述
+                # 有文本，使用文本定位
+                self._record_click('text', elem_text, x_percent, y_percent,
+                                  element_desc=f"[{index}]{elem_desc}", locator_attr='text')
+            elif elem_id:
+                # 有 resource-id，使用 ID 定位
+                self._record_click('id', elem_id, x_percent, y_percent,
+                                  element_desc=f"[{index}]{elem_desc}")
+            else:
+                # 都没有，使用百分比定位
+                self._record_click('percent', f"{x_percent}%,{y_percent}%", x_percent, y_percent,
+                                  element_desc=f"[{index}]{elem_desc}")
+
             return {
                 "success": True,
-                "message": f"✅ 已点击 [{index}] {target['desc']} → ({cx}, {cy})\n💡 建议：再次截图确认操作是否成功",
                 "clicked": {
                     "index": index,
                     "desc": target['desc'],
@@ -963,7 +968,7 @@ class BasicMobileToolsLite:
                     size = ios_client.wda.window_size()
                     width, height = size[0], size[1]
                 else:
-                    return {"success": False, "message": "❌ iOS 客户端未初始化"}
+                    return {"success": False, "msg": "iOS未初始化"}
             else:
                 self.client.u2.screenshot(str(screenshot_path))
                 info = self.client.u2.info
@@ -1041,7 +1046,7 @@ class BasicMobileToolsLite:
                     size = ios_client.wda.window_size()
                     screen_width, screen_height = size[0], size[1]
                 else:
-                    return {"success": False, "message": "❌ iOS 客户端未初始化"}
+                    return {"success": False, "msg": "iOS未初始化"}
             else:
                 info = self.client.u2.info
                 screen_width = info.get('displayWidth', 0)
@@ -1085,17 +1090,9 @@ class BasicMobileToolsLite:
             x_percent = round(x / screen_width * 100, 1) if screen_width > 0 else 0
             y_percent = round(y / screen_height * 100, 1) if screen_height > 0 else 0
             
-            # 记录操作（包含屏幕尺寸和百分比，便于脚本生成时转换）
-            self._record_operation(
-                'click', 
-                x=x, 
-                y=y, 
-                x_percent=x_percent,
-                y_percent=y_percent,
-                screen_width=screen_width,
-                screen_height=screen_height,
-                ref=f"coords_{x}_{y}"
-            )
+            # 使用标准记录格式：坐标点击用百分比作为定位方式（跨分辨率兼容）
+            self._record_click('percent', f"{x_percent}%,{y_percent}%", x_percent, y_percent,
+                              element_desc=f"坐标({x},{y})")
             
             # 🎯 关键步骤：检查应用是否跳转，如果跳转则自动返回目标应用
             app_check = self._check_app_switched()
@@ -1164,14 +1161,14 @@ class BasicMobileToolsLite:
                     size = ios_client.wda.window_size()
                     width, height = size[0], size[1]
                 else:
-                    return {"success": False, "message": "❌ iOS 客户端未初始化"}
+                    return {"success": False, "msg": "iOS未初始化"}
             else:
                 info = self.client.u2.info
                 width = info.get('displayWidth', 0)
                 height = info.get('displayHeight', 0)
             
             if width == 0 or height == 0:
-                return {"success": False, "message": "❌ 无法获取屏幕尺寸"}
+                return {"success": False, "msg": "无法获取屏幕尺寸"}
             
             # 第2步：百分比转像素坐标
             # 公式：像素 = 屏幕尺寸 × (百分比 / 100)
@@ -1186,23 +1183,12 @@ class BasicMobileToolsLite:
             
             time.sleep(0.3)
             
-            # 第4步：记录操作（同时记录百分比和像素）
-            self._record_operation(
-                'click',
-                x=x,
-                y=y,
-                x_percent=x_percent,
-                y_percent=y_percent,
-                screen_width=width,
-                screen_height=height,
-                ref=f"percent_{x_percent}_{y_percent}"
-            )
+            # 第4步：使用标准记录格式
+            self._record_click('percent', f"{x_percent}%,{y_percent}%", x_percent, y_percent,
+                              element_desc=f"百分比({x_percent}%,{y_percent}%)")
             
             return {
                 "success": True,
-                "message": f"✅ 百分比点击成功: ({x_percent}%, {y_percent}%) → 像素({x}, {y})",
-                "screen_size": {"width": width, "height": height},
-                "percent": {"x": x_percent, "y": y_percent},
                 "pixel": {"x": x, "y": y}
             }
         except Exception as e:
@@ -1228,10 +1214,16 @@ class BasicMobileToolsLite:
                     if elem.exists:
                         elem.click()
                         time.sleep(0.3)
-                        self._record_operation('click', element=text, ref=text)
-                        return {"success": True, "message": f"✅ 点击成功: '{text}'"}
-                    return {"success": False, "message": f"❌ 文本不存在: {text}"}
+                        self._record_click('text', text, element_desc=text, locator_attr='text')
+                        return {"success": True}
+                    # 控件树找不到，提示用视觉识别
+                    return {"success": False, "fallback": "vision", "msg": f"未找到'{text}'，用截图点击"}
+                else:
+                    return {"success": False, "msg": "iOS未初始化"}
             else:
+                # 获取屏幕尺寸用于计算百分比
+                screen_width, screen_height = self.client.u2.window_size()
+                
                 # 🔍 先查 XML 树，找到元素及其属性
                 found_elem = self._find_element_in_tree(text, position=position)
                 
@@ -1240,15 +1232,23 @@ class BasicMobileToolsLite:
                     attr_value = found_elem['attr_value']
                     bounds = found_elem.get('bounds')
                     
-                    # 如果有位置参数，直接使用坐标点击（避免 u2 选择器匹配到错误的元素）
+                    # 计算百分比坐标作为兜底
+                    x_pct, y_pct = 0, 0
+                    if bounds:
+                        cx = (bounds[0] + bounds[2]) // 2
+                        cy = (bounds[1] + bounds[3]) // 2
+                        x_pct = round(cx / screen_width * 100, 1)
+                        y_pct = round(cy / screen_height * 100, 1)
+                    
+                    # 如果有位置参数，直接使用坐标点击
                     if position and bounds:
                         x = (bounds[0] + bounds[2]) // 2
                         y = (bounds[1] + bounds[3]) // 2
                         self.client.u2.click(x, y)
                         time.sleep(0.3)
-                        position_info = f" ({position})" if position else ""
-                        self._record_operation('click', element=text, x=x, y=y, ref=f"coords:{x},{y}")
-                        return {"success": True, "message": f"✅ 点击成功(坐标定位): '{text}'{position_info} @ ({x},{y})"}
+                        self._record_click('text', attr_value, x_pct, y_pct, 
+                                          element_desc=f"{text}({position})", locator_attr=attr_type)
+                        return {"success": True}
                     
                     # 没有位置参数时，使用选择器定位
                     if attr_type == 'text':
@@ -1265,23 +1265,24 @@ class BasicMobileToolsLite:
                     if elem and elem.exists(timeout=1):
                         elem.click()
                         time.sleep(0.3)
-                        position_info = f" ({position})" if position else ""
-                        self._record_operation('click', element=text, ref=f"{attr_type}:{attr_value}")
-                        return {"success": True, "message": f"✅ 点击成功({attr_type}): '{text}'{position_info}"}
+                        self._record_click('text', attr_value, x_pct, y_pct,
+                                          element_desc=text, locator_attr=attr_type)
+                        return {"success": True}
                     
-                    # 如果选择器失败，用坐标兜底
+                    # 选择器失败，用坐标兜底
                     if bounds:
                         x = (bounds[0] + bounds[2]) // 2
                         y = (bounds[1] + bounds[3]) // 2
                         self.client.u2.click(x, y)
                         time.sleep(0.3)
-                        position_info = f" ({position})" if position else ""
-                        self._record_operation('click', element=text, x=x, y=y, ref=f"coords:{x},{y}")
-                        return {"success": True, "message": f"✅ 点击成功(坐标兜底): '{text}'{position_info} @ ({x},{y})"}
+                        self._record_click('percent', f"{x_pct}%,{y_pct}%", x_pct, y_pct,
+                                          element_desc=text)
+                        return {"success": True}
                 
-                return {"success": False, "message": f"❌ 文本不存在: {text}"}
+                # 控件树找不到，提示用视觉识别
+                return {"success": False, "fallback": "vision", "msg": f"未找到'{text}'，用截图点击"}
         except Exception as e:
-            return {"success": False, "message": f"❌ 点击失败: {e}"}
+            return {"success": False, "msg": str(e)}
     
     def _find_element_in_tree(self, text: str, position: Optional[str] = None) -> Optional[Dict]:
         """在 XML 树中查找包含指定文本的元素，优先返回可点击的元素
@@ -1420,15 +1421,8 @@ class BasicMobileToolsLite:
             return None
     
     def click_by_id(self, resource_id: str, index: int = 0) -> Dict:
-        """通过 resource-id 点击（支持点击第 N 个元素）
-        
-        Args:
-            resource_id: 元素的 resource-id
-            index: 第几个元素（从 0 开始），默认 0 表示第一个
-        """
+        """通过 resource-id 点击"""
         try:
-            index_desc = f"[{index}]" if index > 0 else ""
-            
             if self._is_ios():
                 ios_client = self._get_ios_client()
                 if ios_client and hasattr(ios_client, 'wda'):
@@ -1436,31 +1430,31 @@ class BasicMobileToolsLite:
                     if not elem.exists:
                         elem = ios_client.wda(name=resource_id)
                     if elem.exists:
-                        # 获取所有匹配的元素
                         elements = elem.find_elements()
                         if index < len(elements):
                             elements[index].click()
                             time.sleep(0.3)
-                            self._record_operation('click', element=f"{resource_id}{index_desc}", ref=resource_id, index=index)
-                            return {"success": True, "message": f"✅ 点击成功: {resource_id}{index_desc}"}
+                            self._record_click('id', resource_id, element_desc=resource_id)
+                            return {"success": True}
                         else:
-                            return {"success": False, "message": f"❌ 索引超出范围: 找到 {len(elements)} 个元素，但请求索引 {index}"}
-                    return {"success": False, "message": f"❌ 元素不存在: {resource_id}"}
+                            return {"success": False, "msg": f"索引{index}超出范围(共{len(elements)}个)"}
+                    return {"success": False, "fallback": "vision", "msg": f"未找到ID'{resource_id}'"}
+                else:
+                    return {"success": False, "msg": "iOS未初始化"}
             else:
                 elem = self.client.u2(resourceId=resource_id)
                 if elem.exists(timeout=0.5):
-                    # 获取匹配元素数量
                     count = elem.count
                     if index < count:
                         elem[index].click()
                         time.sleep(0.3)
-                        self._record_operation('click', element=f"{resource_id}{index_desc}", ref=resource_id, index=index)
-                        return {"success": True, "message": f"✅ 点击成功: {resource_id}{index_desc}" + (f" (共 {count} 个)" if count > 1 else "")}
+                        self._record_click('id', resource_id, element_desc=resource_id)
+                        return {"success": True}
                     else:
-                        return {"success": False, "message": f"❌ 索引超出范围: 找到 {count} 个元素，但请求索引 {index}"}
-                return {"success": False, "message": f"❌ 元素不存在: {resource_id}"}
+                        return {"success": False, "msg": f"索引{index}超出范围(共{count}个)"}
+                return {"success": False, "fallback": "vision", "msg": f"未找到ID'{resource_id}'"}
         except Exception as e:
-            return {"success": False, "message": f"❌ 点击失败: {e}"}
+            return {"success": False, "msg": str(e)}
     
     # ==================== 长按操作 ====================
     
@@ -1494,7 +1488,7 @@ class BasicMobileToolsLite:
                     size = ios_client.wda.window_size()
                     screen_width, screen_height = size[0], size[1]
                 else:
-                    return {"success": False, "message": "❌ iOS 客户端未初始化"}
+                    return {"success": False, "msg": "iOS未初始化"}
             else:
                 info = self.client.u2.info
                 screen_width = info.get('displayWidth', 0)
@@ -1541,38 +1535,17 @@ class BasicMobileToolsLite:
             x_percent = round(x / screen_width * 100, 1) if screen_width > 0 else 0
             y_percent = round(y / screen_height * 100, 1) if screen_height > 0 else 0
             
-            # 记录操作
-            self._record_operation(
-                'long_press', 
-                x=x, 
-                y=y, 
-                x_percent=x_percent,
-                y_percent=y_percent,
-                duration=duration,
-                screen_width=screen_width,
-                screen_height=screen_height,
-                ref=f"coords_{x}_{y}"
-            )
+            # 使用标准记录格式
+            self._record_long_press('percent', f"{x_percent}%,{y_percent}%", duration,
+                                   x_percent, y_percent, element_desc=f"坐标({x},{y})")
             
             if converted:
                 if conversion_type == "crop_offset":
-                    return {
-                        "success": True,
-                        "message": f"✅ 长按成功: ({x}, {y}) 持续 {duration}s\n"
-                                  f"   🔍 局部截图坐标转换: ({original_x},{original_y}) + 偏移({crop_offset_x},{crop_offset_y}) → ({x},{y})"
-                    }
+                    return {"success": True}
                 else:
-                    return {
-                        "success": True,
-                        "message": f"✅ 长按成功: ({x}, {y}) 持续 {duration}s\n"
-                                  f"   📐 坐标已转换: ({original_x},{original_y}) → ({x},{y})\n"
-                                  f"   🖼️ 图片尺寸: {image_width}x{image_height} → 屏幕: {screen_width}x{screen_height}"
-                    }
+                    return {"success": True}
             else:
-                return {
-                    "success": True,
-                    "message": f"✅ 长按成功: ({x}, {y}) 持续 {duration}s [相对位置: {x_percent}%, {y_percent}%]"
-                }
+                return {"success": True}
         except Exception as e:
             return {"success": False, "message": f"❌ 长按失败: {e}"}
     
@@ -1601,14 +1574,14 @@ class BasicMobileToolsLite:
                     size = ios_client.wda.window_size()
                     width, height = size[0], size[1]
                 else:
-                    return {"success": False, "message": "❌ iOS 客户端未初始化"}
+                    return {"success": False, "msg": "iOS未初始化"}
             else:
                 info = self.client.u2.info
                 width = info.get('displayWidth', 0)
                 height = info.get('displayHeight', 0)
             
             if width == 0 or height == 0:
-                return {"success": False, "message": "❌ 无法获取屏幕尺寸"}
+                return {"success": False, "msg": "无法获取屏幕尺寸"}
             
             # 第2步：百分比转像素坐标
             x = int(width * x_percent / 100)
@@ -1626,26 +1599,11 @@ class BasicMobileToolsLite:
             
             time.sleep(0.3)
             
-            # 第4步：记录操作
-            self._record_operation(
-                'long_press',
-                x=x,
-                y=y,
-                x_percent=x_percent,
-                y_percent=y_percent,
-                duration=duration,
-                screen_width=width,
-                screen_height=height,
-                ref=f"percent_{x_percent}_{y_percent}"
-            )
+            # 第4步：使用标准记录格式
+            self._record_long_press('percent', f"{x_percent}%,{y_percent}%", duration,
+                                   x_percent, y_percent, element_desc=f"百分比({x_percent}%,{y_percent}%)")
             
-            return {
-                "success": True,
-                "message": f"✅ 百分比长按成功: ({x_percent}%, {y_percent}%) → 像素({x}, {y}) 持续 {duration}s",
-                "screen_size": {"width": width, "height": height},
-                "percent": {"x": x_percent, "y": y_percent},
-                "pixel": {"x": x, "y": y},
-                "duration": duration
+            return {"success": True
             }
         except Exception as e:
             return {"success": False, "message": f"❌ 百分比长按失败: {e}"}
@@ -1674,10 +1632,13 @@ class BasicMobileToolsLite:
                         else:
                             ios_client.wda.swipe(x, y, x, y, duration=duration)
                         time.sleep(0.3)
-                        self._record_operation('long_press', element=text, duration=duration, ref=text)
-                        return {"success": True, "message": f"✅ 长按成功: '{text}' 持续 {duration}s"}
-                    return {"success": False, "message": f"❌ 文本不存在: {text}"}
+                        self._record_long_press('text', text, duration, element_desc=text, locator_attr='text')
+                        return {"success": True}
+                    return {"success": False, "msg": f"未找到'{text}'"}
             else:
+                # 获取屏幕尺寸用于计算百分比
+                screen_width, screen_height = self.client.u2.window_size()
+                
                 # 先查 XML 树，找到元素
                 found_elem = self._find_element_in_tree(text)
                 
@@ -1686,6 +1647,14 @@ class BasicMobileToolsLite:
                     attr_value = found_elem['attr_value']
                     bounds = found_elem.get('bounds')
                     
+                    # 计算百分比坐标作为兜底
+                    x_pct, y_pct = 0, 0
+                    if bounds:
+                        cx = (bounds[0] + bounds[2]) // 2
+                        cy = (bounds[1] + bounds[3]) // 2
+                        x_pct = round(cx / screen_width * 100, 1)
+                        y_pct = round(cy / screen_height * 100, 1)
+                    
                     # 根据找到的属性类型，使用对应的选择器
                     if attr_type == 'text':
                         elem = self.client.u2(text=attr_value)
@@ -1701,8 +1670,9 @@ class BasicMobileToolsLite:
                     if elem and elem.exists(timeout=1):
                         elem.long_click(duration=duration)
                         time.sleep(0.3)
-                        self._record_operation('long_press', element=text, duration=duration, ref=f"{attr_type}:{attr_value}")
-                        return {"success": True, "message": f"✅ 长按成功({attr_type}): '{text}' 持续 {duration}s"}
+                        self._record_long_press('text', attr_value, duration, x_pct, y_pct,
+                                               element_desc=text, locator_attr=attr_type)
+                        return {"success": True}
                     
                     # 如果选择器失败，用坐标兜底
                     if bounds:
@@ -1710,10 +1680,11 @@ class BasicMobileToolsLite:
                         y = (bounds[1] + bounds[3]) // 2
                         self.client.u2.long_click(x, y, duration=duration)
                         time.sleep(0.3)
-                        self._record_operation('long_press', element=text, x=x, y=y, duration=duration, ref=f"coords:{x},{y}")
-                        return {"success": True, "message": f"✅ 长按成功(坐标兜底): '{text}' @ ({x},{y}) 持续 {duration}s"}
+                        self._record_long_press('percent', f"{x_pct}%,{y_pct}%", duration, x_pct, y_pct,
+                                               element_desc=text)
+                        return {"success": True}
                 
-                return {"success": False, "message": f"❌ 文本不存在: {text}"}
+                return {"success": False, "msg": f"未找到'{text}'"}
         except Exception as e:
             return {"success": False, "message": f"❌ 长按失败: {e}"}
     
@@ -1740,17 +1711,17 @@ class BasicMobileToolsLite:
                         else:
                             ios_client.wda.swipe(x, y, x, y, duration=duration)
                         time.sleep(0.3)
-                        self._record_operation('long_press', element=resource_id, duration=duration, ref=resource_id)
-                        return {"success": True, "message": f"✅ 长按成功: {resource_id} 持续 {duration}s"}
-                    return {"success": False, "message": f"❌ 元素不存在: {resource_id}"}
+                        self._record_long_press('id', resource_id, duration, element_desc=resource_id)
+                        return {"success": True}
+                    return {"success": False, "msg": f"未找到'{resource_id}'"}
             else:
                 elem = self.client.u2(resourceId=resource_id)
                 if elem.exists(timeout=0.5):
                     elem.long_click(duration=duration)
                     time.sleep(0.3)
-                    self._record_operation('long_press', element=resource_id, duration=duration, ref=resource_id)
+                    self._record_long_press('id', resource_id, duration, element_desc=resource_id)
                     return {"success": True, "message": f"✅ 长按成功: {resource_id} 持续 {duration}s"}
-                return {"success": False, "message": f"❌ 元素不存在: {resource_id}"}
+                return {"success": False, "msg": f"未找到'{resource_id}'"}
         except Exception as e:
             return {"success": False, "message": f"❌ 长按失败: {e}"}
     
@@ -1775,7 +1746,7 @@ class BasicMobileToolsLite:
                     if elem.exists:
                         elem.set_text(text)
                         time.sleep(0.3)
-                        self._record_operation('input', element=resource_id, ref=resource_id, text=text)
+                        self._record_input(text, 'id', resource_id)
                         
                         # 🎯 关键步骤：检查应用是否跳转，如果跳转则自动返回目标应用
                         app_check = self._check_app_switched()
@@ -1810,7 +1781,7 @@ class BasicMobileToolsLite:
                     if count == 1:
                         elements.set_text(text)
                         time.sleep(0.3)
-                        self._record_operation('input', element=resource_id, ref=resource_id, text=text)
+                        self._record_input(text, 'id', resource_id)
                         
                         # 🎯 关键步骤：检查应用是否跳转，如果跳转则自动返回目标应用
                         app_check = self._check_app_switched()
@@ -1844,7 +1815,7 @@ class BasicMobileToolsLite:
                                 if info.get('editable') or info.get('focusable'):
                                     elem.set_text(text)
                                     time.sleep(0.3)
-                                    self._record_operation('input', element=resource_id, ref=resource_id, text=text)
+                                    self._record_input(text, 'id', resource_id)
                                     
                                     # 🎯 关键步骤：检查应用是否跳转，如果跳转则自动返回目标应用
                                     app_check = self._check_app_switched()
@@ -1872,7 +1843,7 @@ class BasicMobileToolsLite:
                         # 没找到可编辑的，用第一个
                         elements[0].set_text(text)
                         time.sleep(0.3)
-                        self._record_operation('input', element=resource_id, ref=resource_id, text=text)
+                        self._record_input(text, 'id', resource_id)
                         
                         # 🎯 关键步骤：检查应用是否跳转，如果跳转则自动返回目标应用
                         app_check = self._check_app_switched()
@@ -1903,7 +1874,7 @@ class BasicMobileToolsLite:
                     if et_count == 1:
                         edit_texts.set_text(text)
                         time.sleep(0.3)
-                        self._record_operation('input', element='EditText', ref='EditText', text=text)
+                        self._record_input(text, 'class', 'EditText')
                         
                         # 🎯 关键步骤：检查应用是否跳转，如果跳转则自动返回目标应用
                         app_check = self._check_app_switched()
@@ -1943,7 +1914,7 @@ class BasicMobileToolsLite:
                     if best_elem:
                         best_elem.set_text(text)
                         time.sleep(0.3)
-                        self._record_operation('input', element='EditText', ref='EditText', text=text)
+                        self._record_input(text, 'class', 'EditText')
                         
                         # 🎯 关键步骤：检查应用是否跳转，如果跳转则自动返回目标应用
                         app_check = self._check_app_switched()
@@ -2007,15 +1978,8 @@ class BasicMobileToolsLite:
             x_percent = round(x / screen_width * 100, 1) if screen_width > 0 else 0
             y_percent = round(y / screen_height * 100, 1) if screen_height > 0 else 0
             
-            self._record_operation(
-                'input', 
-                x=x, 
-                y=y, 
-                x_percent=x_percent,
-                y_percent=y_percent,
-                ref=f"coords_{x}_{y}", 
-                text=text
-            )
+            # 使用标准记录格式
+            self._record_input(text, 'percent', f"{x_percent}%,{y_percent}%", x_percent, y_percent)
             
             # 🎯 关键步骤：检查应用是否跳转，如果跳转则自动返回目标应用
             app_check = self._check_app_switched()
@@ -2060,7 +2024,7 @@ class BasicMobileToolsLite:
                     size = ios_client.wda.window_size()
                     width, height = size[0], size[1]
                 else:
-                    return {"success": False, "message": "❌ iOS 客户端未初始化"}
+                    return {"success": False, "msg": "iOS未初始化"}
             else:
                 width, height = self.client.u2.window_size()
             
@@ -2098,13 +2062,8 @@ class BasicMobileToolsLite:
             else:
                 self.client.u2.swipe(x1, y1, x2, y2, duration=0.5)
             
-            # 记录操作信息
-            record_info = {'direction': direction}
-            if y is not None:
-                record_info['y'] = y
-            if y_percent is not None:
-                record_info['y_percent'] = y_percent
-            self._record_operation('swipe', **record_info)
+            # 使用标准记录格式
+            self._record_swipe(direction)
             
             # 🎯 关键步骤：检查应用是否跳转，如果跳转则自动返回目标应用
             app_check = self._check_app_switched()
@@ -2161,22 +2120,22 @@ class BasicMobileToolsLite:
                             ios_client.wda.send_keys('\n')
                         elif ios_key == 'home':
                             ios_client.wda.home()
-                        return {"success": True, "message": f"✅ 按键成功: {key}"}
-                return {"success": False, "message": f"❌ iOS 不支持: {key}"}
+                        return {"success": True}
+                return {"success": False, "msg": f"iOS不支持{key}"}
             else:
                 keycode = key_map.get(key.lower())
                 if keycode:
                     self.client.u2.shell(f'input keyevent {keycode}')
-                    self._record_operation('press_key', key=key)
-                    return {"success": True, "message": f"✅ 按键成功: {key}"}
-                return {"success": False, "message": f"❌ 不支持的按键: {key}"}
+                    self._record_key(key)
+                    return {"success": True}
+                return {"success": False, "msg": f"不支持按键{key}"}
         except Exception as e:
             return {"success": False, "message": f"❌ 按键失败: {e}"}
     
     def wait(self, seconds: float) -> Dict:
         """等待指定时间"""
         time.sleep(seconds)
-        return {"success": True, "message": f"✅ 已等待 {seconds} 秒"}
+        return {"success": True}
     
     # ==================== 应用管理 ====================
     
@@ -2205,10 +2164,7 @@ class BasicMobileToolsLite:
             
             self._record_operation('launch_app', package_name=package_name)
             
-            return {
-                "success": True,
-                "message": f"✅ 已启动: {package_name}\n💡 建议等待 2-3 秒让页面加载\n📱 已设置应用状态监测"
-            }
+            return {"success": True}
         except Exception as e:
             return {"success": False, "message": f"❌ 启动失败: {e}"}
     
@@ -2221,9 +2177,9 @@ class BasicMobileToolsLite:
                     ios_client.wda.app_terminate(package_name)
             else:
                 self.client.u2.app_stop(package_name)
-            return {"success": True, "message": f"✅ 已终止: {package_name}"}
+            return {"success": True}
         except Exception as e:
-            return {"success": False, "message": f"❌ 终止失败: {e}"}
+            return {"success": False, "msg": str(e)}
     
     def list_apps(self, filter_keyword: str = "") -> Dict:
         """列出已安装应用"""
@@ -2335,6 +2291,26 @@ class BasicMobileToolsLite:
                     '_shadow', 'shadow_', '_divider', 'divider_', '_line', 'line_'
                 }
                 
+                # Token 优化：构建精简元素（只返回非空字段）
+                def build_compact_element(resource_id, text, content_desc, bounds, likely_click, class_name):
+                    """只返回有值的字段，节省 token"""
+                    item = {}
+                    if resource_id:
+                        # 精简 resource_id，只保留最后一段
+                        item['id'] = resource_id.split('/')[-1] if '/' in resource_id else resource_id
+                    if text:
+                        item['text'] = text
+                    if content_desc:
+                        item['desc'] = content_desc
+                    if bounds:
+                        item['bounds'] = bounds
+                    if likely_click:
+                        item['click'] = True  # 启发式判断可点击
+                    # class 精简：只保留关键类型
+                    if class_name in ('EditText', 'TextInput', 'Button', 'ImageButton', 'CheckBox', 'Switch'):
+                        item['type'] = class_name
+                    return item
+                
                 result = []
                 for elem in elements:
                     # 获取元素属性
@@ -2354,14 +2330,11 @@ class BasicMobileToolsLite:
                     
                     # 2. 检查是否是功能控件（直接保留）
                     if class_name in FUNCTIONAL_WIDGETS:
-                        result.append({
-                            'resource_id': resource_id,
-                            'text': text,
-                            'content_desc': content_desc,
-                            'bounds': bounds,
-                            'clickable': clickable,
-                            'class': class_name
-                        })
+                        # 使用启发式判断可点击性（替代不准确的 clickable 属性）
+                        likely_click = self._is_likely_clickable(class_name, resource_id, text, content_desc, clickable, bounds)
+                        item = build_compact_element(resource_id, text, content_desc, bounds, likely_click, class_name)
+                        if item:
+                            result.append(item)
                         continue
                     
                     # 3. 检查是否是容器控件
@@ -2374,14 +2347,10 @@ class BasicMobileToolsLite:
                                 # 所有属性都是默认值，过滤掉
                                 continue
                         # 有业务ID或其他有意义属性，保留
-                        result.append({
-                            'resource_id': resource_id,
-                            'text': text,
-                            'content_desc': content_desc,
-                            'bounds': bounds,
-                            'clickable': clickable,
-                            'class': class_name
-                        })
+                        likely_click = self._is_likely_clickable(class_name, resource_id, text, content_desc, clickable, bounds)
+                        item = build_compact_element(resource_id, text, content_desc, bounds, likely_click, class_name)
+                        if item:
+                            result.append(item)
                         continue
                     
                     # 4. 检查是否是装饰类控件
@@ -2398,14 +2367,21 @@ class BasicMobileToolsLite:
                         continue
                     
                     # 6. 其他情况：有意义的元素保留
-                    result.append({
-                        'resource_id': resource_id,
-                        'text': text,
-                        'content_desc': content_desc,
-                        'bounds': bounds,
-                        'clickable': clickable,
-                        'class': class_name
+                    likely_click = self._is_likely_clickable(class_name, resource_id, text, content_desc, clickable, bounds)
+                    item = build_compact_element(resource_id, text, content_desc, bounds, likely_click, class_name)
+                    if item:
+                        result.append(item)
+                
+                # Token 优化：可选限制返回元素数量（默认不限制，确保准确度）
+                if TOKEN_OPTIMIZATION and MAX_ELEMENTS > 0 and len(result) > MAX_ELEMENTS:
+                    # 仅在用户明确设置 MAX_ELEMENTS_RETURN 时才截断
+                    truncated = result[:MAX_ELEMENTS]
+                    truncated.append({
+                        '_truncated': True,
+                        '_total': len(result),
+                        '_shown': MAX_ELEMENTS
                     })
+                    return truncated
                 
                 return result
         except Exception as e:
@@ -2442,6 +2418,68 @@ class BasicMobileToolsLite:
         
         return True
     
+    def _is_likely_clickable(self, class_name: str, resource_id: str, text: str,
+                             content_desc: str, clickable: bool, bounds: str) -> bool:
+        """
+        启发式判断元素是否可能可点击
+        
+        Android 的 clickable 属性经常不准确，因为：
+        1. 点击事件可能设置在父容器上
+        2. 使用 onTouchListener 而不是 onClick
+        3. RecyclerView item 通过 ItemClickListener 处理
+        
+        此方法通过多种规则推断元素的真实可点击性
+        """
+        # 规则1：clickable=true 肯定可点击
+        if clickable:
+            return True
+        
+        # 规则2：特定类型的控件通常可点击
+        TYPICALLY_CLICKABLE = {
+            'Button', 'ImageButton', 'CheckBox', 'RadioButton', 'Switch',
+            'ToggleButton', 'FloatingActionButton', 'Chip', 'TabView',
+            'EditText', 'TextInput',  # 输入框可点击获取焦点
+        }
+        if class_name in TYPICALLY_CLICKABLE:
+            return True
+        
+        # 规则3：resource_id 包含可点击关键词
+        if resource_id:
+            id_lower = resource_id.lower()
+            CLICK_KEYWORDS = [
+                'btn', 'button', 'click', 'tap', 'submit', 'confirm',
+                'cancel', 'close', 'back', 'next', 'prev', 'more',
+                'action', 'link', 'menu', 'tab', 'item', 'cell',
+                'card', 'avatar', 'icon', 'entry', 'option', 'arrow'
+            ]
+            for kw in CLICK_KEYWORDS:
+                if kw in id_lower:
+                    return True
+        
+        # 规则4：content_desc 包含可点击暗示
+        if content_desc:
+            desc_lower = content_desc.lower()
+            CLICK_HINTS = ['点击', '按钮', '关闭', '返回', '更多', 'click', 'tap', 'button', 'close']
+            for hint in CLICK_HINTS:
+                if hint in desc_lower:
+                    return True
+        
+        # 规则5：有 resource_id 或 content_desc 的小图标可能可点击
+        # （纯 ImageView 不加判断，误判率太高）
+        if class_name in ('ImageView', 'Image') and (resource_id or content_desc) and bounds:
+            match = re.match(r'\[(\d+),(\d+)\]\[(\d+),(\d+)\]', bounds)
+            if match:
+                x1, y1, x2, y2 = map(int, match.groups())
+                w, h = x2 - x1, y2 - y1
+                # 小图标（20-100px）更可能是按钮
+                if 20 <= w <= 100 and 20 <= h <= 100:
+                    return True
+        
+        # 规则6：移除（TextView 误判率太高，只依赖上面的规则）
+        # 如果有 clickable=true 或 ID/desc 中有关键词，前面的规则已经覆盖
+        
+        return False
+    
     def find_close_button(self) -> Dict:
         """智能查找关闭按钮（不点击，只返回位置）
         
@@ -2455,7 +2493,7 @@ class BasicMobileToolsLite:
             import re
             
             if self._is_ios():
-                return {"success": False, "message": "iOS 暂不支持，请使用截图+坐标点击"}
+                return {"success": False, "msg": "iOS暂不支持"}
             
             # 获取屏幕尺寸
             screen_width = self.client.u2.info.get('displayWidth', 720)
@@ -2466,6 +2504,14 @@ class BasicMobileToolsLite:
             import xml.etree.ElementTree as ET
             root = ET.fromstring(xml_string)
             
+            # 🔴 先检测是否有弹窗，避免误识别普通页面的按钮
+            popup_bounds, popup_confidence = self._detect_popup_with_confidence(
+                root, screen_width, screen_height
+            )
+            
+            if popup_bounds is None or popup_confidence < 0.5:
+                return {"success": True, "popup": False}
+            
             # 关闭按钮特征
             close_texts = ['×', 'X', 'x', '关闭', '取消', 'close', 'Close', '跳过', '知道了', '我知道了']
             candidates = []
@@ -2567,27 +2613,16 @@ class BasicMobileToolsLite:
             candidates.sort(key=lambda x: x['score'], reverse=True)
             best = candidates[0]
             
+            # Token 优化：只返回最必要的信息
             return {
                 "success": True,
-                "message": f"✅ 找到可能的关闭按钮",
-                "best_candidate": {
-                    "reason": best['reason'],
-                    "center": {"x": best['center_x'], "y": best['center_y']},
-                    "percent": {"x": best['x_percent'], "y": best['y_percent']},
-                    "bounds": best['bounds'],
-                    "size": best['size'],
-                    "score": best['score']
-                },
-                "click_command": f"mobile_click_by_percent({best['x_percent']}, {best['y_percent']})",
-                "other_candidates": [
-                    {"reason": c['reason'], "percent": f"({c['x_percent']}%, {c['y_percent']}%)", "score": c['score']}
-                    for c in candidates[1:4]
-                ] if len(candidates) > 1 else [],
-                "screen_size": {"width": screen_width, "height": screen_height}
+                "popup": True,
+                "close": {"x": best['x_percent'], "y": best['y_percent']},
+                "cmd": f"click_by_percent({best['x_percent']},{best['y_percent']})"
             }
             
         except Exception as e:
-            return {"success": False, "message": f"❌ 查找关闭按钮失败: {e}"}
+            return {"success": False, "msg": str(e)}
     
     def close_popup(self) -> Dict:
         """智能关闭弹窗（改进版）
@@ -2610,7 +2645,7 @@ class BasicMobileToolsLite:
             
             # 获取屏幕尺寸
             if self._is_ios():
-                return {"success": False, "message": "iOS 暂不支持，请使用截图+坐标点击"}
+                return {"success": False, "msg": "iOS暂不支持"}
             
             screen_width = self.client.u2.info.get('displayWidth', 720)
             screen_height = self.client.u2.info.get('displayHeight', 1280)
@@ -2630,58 +2665,18 @@ class BasicMobileToolsLite:
                 root = ET.fromstring(xml_string)
                 all_elements = list(root.iter())
                 
-                # ===== 第一步：检测弹窗区域 =====
-                # 弹窗特征：非全屏、面积较大、通常在屏幕中央的容器
-                popup_containers = []
-                for idx, elem in enumerate(all_elements):
-                    bounds_str = elem.attrib.get('bounds', '')
-                    class_name = elem.attrib.get('class', '')
-                    
-                    if not bounds_str:
-                        continue
-                    
-                    match = re.match(r'\[(\d+),(\d+)\]\[(\d+),(\d+)\]', bounds_str)
-                    if not match:
-                        continue
-                    
-                    x1, y1, x2, y2 = map(int, match.groups())
-                    width = x2 - x1
-                    height = y2 - y1
-                    area = width * height
-                    screen_area = screen_width * screen_height
-                    
-                    # 弹窗容器特征（更严格，排除主要内容区域）：
-                    # 1. 面积在屏幕的 10%-70% 之间（排除主要内容区域，通常占80%+）
-                    # 2. 宽度和高度都要小于屏幕的95%（不是全屏）
-                    # 3. 是容器类型（Layout/View/Dialog）
-                    # 4. 不在屏幕左边缘（排除从x=0开始的主要内容容器）
-                    # 5. 高度不能占据屏幕的大部分（排除主要内容区域）
-                    is_container = any(kw in class_name for kw in ['Layout', 'View', 'Dialog', 'Card', 'Container'])
-                    area_ratio = area / screen_area
-                    is_not_fullscreen = (width < screen_width * 0.95 and height < screen_height * 0.95)
-                    is_reasonable_size = 0.10 < area_ratio < 0.70
-                    is_not_at_left_edge = x1 > screen_width * 0.05
-                    height_ratio = height / screen_height if screen_height > 0 else 0
-                    is_not_main_content = height_ratio < 0.85
-                    
-                    # 排除状态栏区域（y1 通常很小）
-                    is_below_statusbar = y1 > 50
-                    
-                    if is_container and is_not_fullscreen and is_reasonable_size and is_not_at_left_edge and is_not_main_content and is_below_statusbar:
-                        popup_containers.append({
-                            'bounds': (x1, y1, x2, y2),
-                            'bounds_str': bounds_str,
-                            'area': area,
-                            'area_ratio': area_ratio,
-                            'idx': idx,  # 元素在 XML 中的顺序（越后越上层）
-                            'class': class_name
-                        })
+                # ===== 第一步：使用严格的置信度检测弹窗区域 =====
+                popup_bounds, popup_confidence = self._detect_popup_with_confidence(
+                    root, screen_width, screen_height
+                )
                 
-                # 选择最可能的弹窗容器（优先选择：XML 顺序靠后 + 面积适中）
-                if popup_containers:
-                    # 按 XML 顺序倒序（后出现的在上层），然后按面积适中程度排序
-                    popup_containers.sort(key=lambda x: (x['idx'], -abs(x['area_ratio'] - 0.3)), reverse=True)
-                    popup_bounds = popup_containers[0]['bounds']
+                # 如果置信度不够高，记录但继续尝试查找关闭按钮
+                popup_detected = popup_bounds is not None and popup_confidence >= 0.6
+                
+                # 🔴 关键检查：如果没有检测到弹窗区域，直接返回"无弹窗"
+                # 避免误点击普通页面上的"关闭"、"取消"等按钮
+                if not popup_detected:
+                    return {"success": True, "popup": False}
                 
                 # ===== 第二步：在弹窗范围内查找关闭按钮 =====
                 for idx, elem in enumerate(all_elements):
@@ -2813,93 +2808,16 @@ class BasicMobileToolsLite:
                             'content_desc': content_desc,
                             'x_percent': round(rel_x * 100, 1),
                             'y_percent': round(rel_y * 100, 1),
-                            'in_popup': popup_bounds is not None
+                            'in_popup': popup_detected
                         })
                         
             except ET.ParseError:
                 pass
             
             if not close_candidates:
-                # 如果检测到弹窗区域，先尝试点击常见的关闭按钮位置
-                if popup_bounds:
-                    px1, py1, px2, py2 = popup_bounds
-                    popup_width = px2 - px1
-                    popup_height = py2 - py1
-                    
-                    # 【优化】X按钮有三种常见位置：
-                    # 1. 弹窗内靠近顶部边界（内嵌X按钮）- 最常见
-                    # 2. 弹窗边界上方（浮动X按钮）
-                    # 3. 弹窗正下方（底部关闭按钮）
-                    offset_x = max(60, int(popup_width * 0.07))   # 宽度7%
-                    offset_y_above = max(35, int(popup_height * 0.025))  # 高度2.5%，在边界之上
-                    offset_y_near = max(45, int(popup_height * 0.03))    # 高度3%，紧贴顶边界内侧
-                    
-                    try_positions = [
-                        # 【最高优先级】弹窗内紧贴顶部边界
-                        (px2 - offset_x, py1 + offset_y_near, "弹窗右上角"),
-                        # 弹窗边界上方（浮动X按钮）
-                        (px2 - offset_x, py1 - offset_y_above, "弹窗右上浮"),
-                        # 弹窗正下方中间（底部关闭按钮）
-                        ((px1 + px2) // 2, py2 + max(50, int(popup_height * 0.04)), "弹窗下方中间"),
-                        # 弹窗正上方中间
-                        ((px1 + px2) // 2, py1 - 40, "弹窗正上方"),
-                    ]
-                    
-                    for try_x, try_y, position_name in try_positions:
-                        if 0 <= try_x <= screen_width and 0 <= try_y <= screen_height:
-                            self.client.u2.click(try_x, try_y)
-                            time.sleep(0.3)
-                    
-                    # 🎯 关键步骤：检查应用是否跳转，如果跳转说明弹窗去除失败，需要返回目标应用
-                    app_check = self._check_app_switched()
-                    return_result = None
-                    
-                    if app_check['switched']:
-                        # 应用已跳转，说明弹窗去除失败，尝试返回目标应用
-                        return_result = self._return_to_target_app()
-                    
-                    # 尝试后截图，让 AI 判断是否成功
-                    screenshot_result = self.take_screenshot("尝试关闭后")
-                    
-                    msg = f"✅ 已尝试点击常见关闭按钮位置"
-                    if app_check['switched']:
-                        msg += f"\n⚠️ 应用已跳转，说明弹窗去除失败"
-                        if return_result:
-                            if return_result['success']:
-                                msg += f"\n{return_result['message']}"
-                            else:
-                                msg += f"\n❌ 自动返回失败: {return_result['message']}"
-                    
-                    return {
-                        "success": True,
-                        "message": msg,
-                        "tried_positions": [p[2] for p in try_positions],
-                        "screenshot": screenshot_result.get("screenshot_path", ""),
-                        "app_check": app_check,
-                        "return_to_app": return_result,
-                        "tip": "请查看截图确认弹窗是否已关闭。如果还在，可手动分析截图找到关闭按钮位置。"
-                    }
-                
-                # 没有检测到弹窗区域，截图让 AI 分析
-                screenshot_result = self.take_screenshot(description="页面截图", compress=True)
-                
-                return {
-                    "success": False,
-                    "message": "❌ 未检测到弹窗区域，已截图供 AI 分析",
-                    "action_required": "请查看截图找到关闭按钮，调用 mobile_click_at_coords 点击",
-                    "screenshot": screenshot_result.get("screenshot_path", ""),
-                    "screen_size": {"width": screen_width, "height": screen_height},
-                    "image_size": {
-                        "width": screenshot_result.get("image_width", screen_width),
-                        "height": screenshot_result.get("image_height", screen_height)
-                    },
-                    "original_size": {
-                        "width": screenshot_result.get("original_img_width", screen_width),
-                        "height": screenshot_result.get("original_img_height", screen_height)
-                    },
-                    "search_areas": ["弹窗右上角", "弹窗正上方", "弹窗下方中间", "屏幕右上角"],
-                    "time_warning": "⚠️ 截图分析期间弹窗可能自动消失。如果是定时弹窗，建议等待其自动消失。"
-                }
+                if popup_detected and popup_bounds:
+                    return {"success": False, "fallback": "vision", "popup": True}
+                return {"success": True, "popup": False}
             
             # 按得分排序，取最可能的
             close_candidates.sort(key=lambda x: x['score'], reverse=True)
@@ -2917,61 +2835,22 @@ class BasicMobileToolsLite:
                 # 应用已跳转，说明弹窗去除失败，尝试返回目标应用
                 return_result = self._return_to_target_app()
             
-            # 点击后截图，让 AI 判断是否成功
-            screenshot_result = self.take_screenshot("关闭弹窗后")
-            
-            # 记录操作（使用百分比，跨设备兼容）
-            self._record_operation(
-                'click',
-                x=best['center_x'],
-                y=best['center_y'],
-                x_percent=best['x_percent'],
-                y_percent=best['y_percent'],
-                screen_width=screen_width,
-                screen_height=screen_height,
-                ref=f"close_popup_{best['position']}"
-            )
+            # 记录操作
+            self._record_click('percent', f"{best['x_percent']}%,{best['y_percent']}%", 
+                              best['x_percent'], best['y_percent'],
+                              element_desc=f"关闭按钮({best['position']})")
             
-            # 构建返回消息
-            msg = f"✅ 已点击关闭按钮 ({best['position']}): ({best['center_x']}, {best['center_y']})"
+            # Token 优化：精简返回值
+            result = {"success": True, "clicked": True}
             if app_check['switched']:
-                msg += f"\n⚠️ 应用已跳转，说明弹窗去除失败"
+                result["switched"] = True
                 if return_result:
-                    if return_result['success']:
-                        msg += f"\n{return_result['message']}"
-                    else:
-                        msg += f"\n❌ 自动返回失败: {return_result['message']}"
+                    result["returned"] = return_result['success']
             
-            # 返回候选按钮列表，让 AI 看截图判断
-            # 如果弹窗还在，AI 可以选择点击其他候选按钮
-            return {
-                "success": True,
-                "message": msg,
-                "clicked": {
-                    "position": best['position'],
-                    "match_type": best['match_type'],
-                    "coords": (best['center_x'], best['center_y']),
-                    "percent": (best['x_percent'], best['y_percent'])
-                },
-                "screenshot": screenshot_result.get("screenshot_path", ""),
-                "popup_detected": popup_bounds is not None,
-                "popup_bounds": f"[{popup_bounds[0]},{popup_bounds[1]}][{popup_bounds[2]},{popup_bounds[3]}]" if popup_bounds else None,
-                "app_check": app_check,
-                "return_to_app": return_result,
-                "other_candidates": [
-                    {
-                        "position": c['position'], 
-                        "type": c['match_type'], 
-                        "coords": (c['center_x'], c['center_y']),
-                        "percent": (c['x_percent'], c['y_percent'])
-                    }
-                    for c in close_candidates[1:4]  # 返回其他3个候选，AI 可以选择
-                ],
-                "tip": "请查看截图判断弹窗是否已关闭。如果弹窗还在，可以尝试点击 other_candidates 中的其他位置"
-            }
+            return result
             
         except Exception as e:
-            return {"success": False, "message": f"❌ 关闭弹窗失败: {e}"}
+            return {"success": False, "msg": str(e)}
     
     def _get_position_name(self, rel_x: float, rel_y: float) -> str:
         """根据相对坐标获取位置名称"""
@@ -3014,6 +2893,308 @@ class BasicMobileToolsLite:
                 return 0.8
         else:  # 中间区域
             return 0.5
+
+    def _detect_popup_with_confidence(self, root, screen_width: int, screen_height: int) -> tuple:
+        """严格的弹窗检测 - 使用置信度评分，避免误识别普通页面
+        
+        真正的弹窗特征：
+        1. class 名称包含 Dialog/Popup/Alert/Modal/BottomSheet（强特征）
+        2. resource-id 包含 dialog/popup/alert/modal（强特征）
+        3. 有遮罩层（大面积半透明 View 在弹窗之前）
+        4. 居中显示且非全屏
+        5. XML 层级靠后且包含可交互元素
+        
+        Returns:
+            (popup_bounds, confidence) 或 (None, 0)
+            confidence >= 0.6 才认为是弹窗
+        """
+        import re
+        
+        screen_area = screen_width * screen_height
+        
+        # 收集所有元素信息
+        all_elements = []
+        for idx, elem in enumerate(root.iter()):
+            bounds_str = elem.attrib.get('bounds', '')
+            if not bounds_str:
+                continue
+            
+            match = re.match(r'\[(\d+),(\d+)\]\[(\d+),(\d+)\]', bounds_str)
+            if not match:
+                continue
+            
+            x1, y1, x2, y2 = map(int, match.groups())
+            width = x2 - x1
+            height = y2 - y1
+            area = width * height
+            
+            class_name = elem.attrib.get('class', '')
+            resource_id = elem.attrib.get('resource-id', '')
+            clickable = elem.attrib.get('clickable', 'false') == 'true'
+            
+            all_elements.append({
+                'idx': idx,
+                'bounds': (x1, y1, x2, y2),
+                'width': width,
+                'height': height,
+                'area': area,
+                'area_ratio': area / screen_area if screen_area > 0 else 0,
+                'class': class_name,
+                'resource_id': resource_id,
+                'clickable': clickable,
+                'center_x': (x1 + x2) // 2,
+                'center_y': (y1 + y2) // 2,
+            })
+        
+        if not all_elements:
+            return None, 0
+        
+        # 弹窗检测关键词
+        dialog_class_keywords = ['Dialog', 'Popup', 'Alert', 'Modal', 'BottomSheet', 'PopupWindow']
+        dialog_id_keywords = ['dialog', 'popup', 'alert', 'modal', 'bottom_sheet', 'overlay', 'mask']
+        
+        popup_candidates = []
+        has_mask_layer = False
+        mask_idx = -1
+        
+        for elem in all_elements:
+            x1, y1, x2, y2 = elem['bounds']
+            class_name = elem['class']
+            resource_id = elem['resource_id']
+            area_ratio = elem['area_ratio']
+            
+            # 检测遮罩层（大面积、几乎全屏、通常是 FrameLayout/View）
+            if area_ratio > 0.85 and elem['width'] >= screen_width * 0.95:
+                # 可能是遮罩层，记录位置
+                if 'FrameLayout' in class_name or 'View' in class_name:
+                    has_mask_layer = True
+                    mask_idx = elem['idx']
+            
+            # 跳过全屏元素
+            if area_ratio > 0.9:
+                continue
+            
+            # 跳过太小的元素
+            if area_ratio < 0.05:
+                continue
+            
+            # 跳过状态栏区域
+            if y1 < 50:
+                continue
+            
+            confidence = 0.0
+            
+            # 【强特征】class 名称包含弹窗关键词 (+0.5)
+            if any(kw in class_name for kw in dialog_class_keywords):
+                confidence += 0.5
+            
+            # 【强特征】resource-id 包含弹窗关键词 (+0.4)
+            if any(kw in resource_id.lower() for kw in dialog_id_keywords):
+                confidence += 0.4
+            
+            # 【中等特征】居中显示 (+0.2)
+            center_x = elem['center_x']
+            center_y = elem['center_y']
+            is_centered_x = abs(center_x - screen_width / 2) < screen_width * 0.15
+            is_centered_y = abs(center_y - screen_height / 2) < screen_height * 0.25
+            if is_centered_x and is_centered_y:
+                confidence += 0.2
+            elif is_centered_x:
+                confidence += 0.1
+            
+            # 【中等特征】非全屏但有一定大小 (+0.15)
+            if 0.15 < area_ratio < 0.75:
+                confidence += 0.15
+            
+            # 【弱特征】XML 顺序靠后（在视图层级上层）(+0.1)
+            if elem['idx'] > len(all_elements) * 0.5:
+                confidence += 0.1
+            
+            # 【弱特征】有遮罩层且在遮罩层之后 (+0.15)
+            if has_mask_layer and elem['idx'] > mask_idx:
+                confidence += 0.15
+            
+            # 只有达到阈值才加入候选
+            if confidence >= 0.3:
+                popup_candidates.append({
+                    'bounds': elem['bounds'],
+                    'confidence': confidence,
+                    'class': class_name,
+                    'resource_id': resource_id,
+                    'idx': elem['idx']
+                })
+        
+        if not popup_candidates:
+            return None, 0
+        
+        # 选择置信度最高的
+        popup_candidates.sort(key=lambda x: (x['confidence'], x['idx']), reverse=True)
+        best = popup_candidates[0]
+        
+        # 只有置信度 >= 0.6 才返回弹窗
+        if best['confidence'] >= 0.6:
+            return best['bounds'], best['confidence']
+        
+        return None, best['confidence']
+    
+    def start_toast_watch(self) -> Dict:
+        """开始监听 Toast（仅 Android）
+        
+        ⚠️ 必须在执行操作之前调用！
+        
+        正确流程：
+        1. 调用 mobile_start_toast_watch() 开始监听
+        2. 执行操作（如点击提交按钮）
+        3. 调用 mobile_get_toast() 获取 Toast 内容
+        
+        Returns:
+            监听状态
+        """
+        if self._is_ios():
+            return {
+                "success": False,
+                "message": "❌ iOS 不支持 Toast 检测，Toast 是 Android 特有功能"
+            }
+        
+        try:
+            # 清除缓存并开始监听
+            self.client.u2.toast.reset()
+            return {
+                "success": True,
+                "message": "✅ Toast 监听已开启，请立即执行操作，然后调用 mobile_get_toast 获取结果"
+            }
+        except Exception as e:
+            return {
+                "success": False,
+                "message": f"❌ 开启 Toast 监听失败: {e}"
+            }
+    
+    def get_toast(self, timeout: float = 5.0, reset_first: bool = False) -> Dict:
+        """获取 Toast 消息（仅 Android）
+        
+        Toast 是 Android 系统级的短暂提示消息，常用于显示操作结果。
+        
+        ⚠️ 推荐用法（两步走）：
+        1. 先调用 mobile_start_toast_watch() 开始监听
+        2. 执行操作（如点击提交按钮）
+        3. 调用 mobile_get_toast() 获取 Toast
+        
+        或者设置 reset_first=True，会自动 reset 后等待（适合操作已自动触发的场景）
+        
+        Args:
+            timeout: 等待 Toast 出现的超时时间（秒），默认 5 秒
+            reset_first: 是否先 reset（清除旧缓存），默认 False
+        
+        Returns:
+            包含 Toast 消息的字典
+        """
+        if self._is_ios():
+            return {
+                "success": False,
+                "message": "❌ iOS 不支持 Toast 检测，Toast 是 Android 特有功能"
+            }
+        
+        try:
+            if reset_first:
+                # 清除旧缓存，适合等待即将出现的 Toast
+                self.client.u2.toast.reset()
+            
+            # 等待并获取 Toast 消息
+            toast_message = self.client.u2.toast.get_message(
+                wait_timeout=timeout,
+                default=None
+            )
+            
+            if toast_message:
+                return {
+                    "success": True,
+                    "toast_found": True,
+                    "message": toast_message,
+                    "tip": "Toast 消息获取成功"
+                }
+            else:
+                return {
+                    "success": True,
+                    "toast_found": False,
+                    "message": None,
+                    "tip": f"在 {timeout} 秒内未检测到 Toast。提示：先调用 mobile_start_toast_watch，再执行操作，最后调用此工具"
+                }
+        except Exception as e:
+            return {
+                "success": False,
+                "message": f"❌ 获取 Toast 失败: {e}"
+            }
+    
+    def assert_toast(self, expected_text: str, timeout: float = 5.0, contains: bool = True) -> Dict:
+        """断言 Toast 消息（仅 Android）
+        
+        等待 Toast 出现并验证内容是否符合预期。
+        
+        ⚠️ 推荐用法：先调用 mobile_start_toast_watch，再执行操作，最后调用此工具
+        
+        Args:
+            expected_text: 期望的 Toast 文本
+            timeout: 等待超时时间（秒）
+            contains: True 表示包含匹配，False 表示精确匹配
+        
+        Returns:
+            断言结果
+        """
+        if self._is_ios():
+            return {
+                "success": False,
+                "passed": False,
+                "message": "❌ iOS 不支持 Toast 检测"
+            }
+        
+        try:
+            # 获取 Toast（不 reset，假设之前已经调用过 start_toast_watch）
+            toast_message = self.client.u2.toast.get_message(
+                wait_timeout=timeout,
+                default=None
+            )
+            
+            if toast_message is None:
+                return {
+                    "success": True,
+                    "passed": False,
+                    "expected": expected_text,
+                    "actual": None,
+                    "message": f"❌ 断言失败：未检测到 Toast 消息"
+                }
+            
+            # 匹配检查
+            if contains:
+                passed = expected_text in toast_message
+                match_type = "包含"
+            else:
+                passed = expected_text == toast_message
+                match_type = "精确"
+            
+            if passed:
+                return {
+                    "success": True,
+                    "passed": True,
+                    "expected": expected_text,
+                    "actual": toast_message,
+                    "match_type": match_type,
+                    "message": f"✅ Toast 断言通过：'{toast_message}'"
+                }
+            else:
+                return {
+                    "success": True,
+                    "passed": False,
+                    "expected": expected_text,
+                    "actual": toast_message,
+                    "match_type": match_type,
+                    "message": f"❌ Toast 断言失败：期望 '{expected_text}'，实际 '{toast_message}'"
+                }
+        except Exception as e:
+            return {
+                "success": False,
+                "passed": False,
+                "message": f"❌ Toast 断言异常: {e}"
+            }
     
     def assert_text(self, text: str) -> Dict:
         """检查页面是否包含文本（支持精确匹配和包含匹配）"""
@@ -3102,8 +3283,13 @@ class BasicMobileToolsLite:
             "1. 文本定位 - 最稳定，跨设备兼容",
             "2. ID 定位 - 稳定，跨设备兼容",
             "3. 百分比定位 - 跨分辨率兼容（坐标自动转换）",
+            "",
+            "运行方式：",
+            "  pytest {filename} -v        # 使用 pytest 运行",
+            "  python {filename}           # 直接运行",
             f'"""',
             "import time",
+            "import pytest",
             "import uiautomator2 as u2",
             "",
             f'PACKAGE_NAME = "{package_name}"',
@@ -3179,22 +3365,52 @@ class BasicMobileToolsLite:
             "    return True",
             "",
             "",
-            "def test_main():",
-            "    # 连接设备",
-            "    d = u2.connect()",
-            "    d.implicitly_wait(10)  # 设置全局等待",
-            "    ",
-            "    # 启动应用",
-            f"    d.app_start(PACKAGE_NAME)",
-            "    time.sleep(LAUNCH_WAIT)  # 等待启动（可调整）",
+            "def swipe_direction(d, direction):",
+            '    """',
+            '    通用滑动方法（兼容所有 uiautomator2 版本）',
+            '    ',
+            '    Args:',
+            '        d: uiautomator2 设备对象',
+            '        direction: 滑动方向 (up/down/left/right)',
+            '    """',
+            "    info = d.info",
+            "    width = info.get('displayWidth', 0)",
+            "    height = info.get('displayHeight', 0)",
+            "    cx, cy = width // 2, height // 2",
             "    ",
-            "    # 尝试关闭启动广告（可选，根据 App 情况调整）",
+            "    if direction == 'up':",
+            "        d.swipe(cx, int(height * 0.8), cx, int(height * 0.3))",
+            "    elif direction == 'down':",
+            "        d.swipe(cx, int(height * 0.3), cx, int(height * 0.8))",
+            "    elif direction == 'left':",
+            "        d.swipe(int(width * 0.8), cy, int(width * 0.2), cy)",
+            "    elif direction == 'right':",
+            "        d.swipe(int(width * 0.2), cy, int(width * 0.8), cy)",
+            "    return True",
+            "",
+            "",
+            "# ========== pytest fixture ==========",
+            "@pytest.fixture(scope='function')",
+            "def device():",
+            '    """pytest fixture: 连接设备并启动应用"""',
+            "    d = u2.connect()",
+            "    d.implicitly_wait(10)",
+            "    d.app_start(PACKAGE_NAME)",
+            "    time.sleep(LAUNCH_WAIT)",
             "    if CLOSE_AD_ON_LAUNCH:",
             "        close_ad_if_exists(d)",
+            "    yield d",
+            "    # 测试结束后可选择关闭应用",
+            "    # d.app_stop(PACKAGE_NAME)",
+            "",
+            "",
+            f"def test_{safe_name}(device):",
+            '    """测试用例主函数"""',
+            "    d = device",
             "    ",
         ]
         
-        # 生成操作代码（跳过启动应用相关操作，因为脚本头部已处理）
+        # 生成操作代码（使用标准记录格式，逻辑更简洁）
         step_num = 0
         for op in self.operation_history:
             action = op.get('action')
@@ -3206,131 +3422,122 @@ class BasicMobileToolsLite:
             step_num += 1
             
             if action == 'click':
-                ref = op.get('ref', '')
-                element = op.get('element', '')
-                has_coords = 'x' in op and 'y' in op
-                has_percent = 'x_percent' in op and 'y_percent' in op
-                
-                # 判断 ref 是否为坐标格式（coords_ 或 coords:）
-                is_coords_ref = ref.startswith('coords_') or ref.startswith('coords:')
-                is_percent_ref = ref.startswith('percent_')
-                
-                # 优先级：文本 > ID > 百分比 > 坐标（兜底）
-                if ref and not is_coords_ref and not is_percent_ref and ':' not in ref:
-                    # 1️⃣ 使用文本（最稳定，优先）- 排除 "text:xxx" 等带冒号的格式
-                    script_lines.append(f"    # 步骤{step_num}: 点击文本 '{ref}' (文本定位，最稳定)")
-                    script_lines.append(f"    safe_click(d, d(text='{ref}'))")
-                elif ref and ':' in ref and not is_coords_ref and not is_percent_ref:
-                    # 1️⃣-b 使用文本（Android 的 text:xxx 或 description:xxx 格式）
-                    # 提取冒号后面的实际文本值
-                    actual_text = ref.split(':', 1)[1] if ':' in ref else ref
-                    script_lines.append(f"    # 步骤{step_num}: 点击文本 '{actual_text}' (文本定位，最稳定)")
-                    script_lines.append(f"    safe_click(d, d(text='{actual_text}'))")
-                elif ref and (':id/' in ref or ref.startswith('com.')):
-                    # 2️⃣ 使用 resource-id（稳定）
-                    script_lines.append(f"    # 步骤{step_num}: 点击元素 (ID定位)")
-                    script_lines.append(f"    safe_click(d, d(resourceId='{ref}'))")
-                elif has_percent:
-                    # 3️⃣ 使用百分比（跨分辨率兼容）
-                    x_pct = op['x_percent']
-                    y_pct = op['y_percent']
-                    desc = f" ({element})" if element else ""
-                    script_lines.append(f"    # 步骤{step_num}: 点击位置{desc} (百分比定位，跨分辨率兼容)")
-                    script_lines.append(f"    click_by_percent(d, {x_pct}, {y_pct})  # 原坐标: ({op.get('x', '?')}, {op.get('y', '?')})")
-                elif has_coords:
-                    # 4️⃣ 坐标兜底（不推荐，仅用于无法获取百分比的情况）
-                    desc = f" ({element})" if element else ""
-                    script_lines.append(f"    # 步骤{step_num}: 点击坐标{desc} (⚠️ 坐标定位，可能不兼容其他分辨率)")
-                    script_lines.append(f"    d.click({op['x']}, {op['y']})")
+                # 新格式：使用 locator_type 和 locator_value
+                locator_type = op.get('locator_type', '')
+                locator_value = op.get('locator_value', '')
+                locator_attr = op.get('locator_attr', 'text')
+                element_desc = op.get('element_desc', '')
+                x_pct = op.get('x_percent', 0)
+                y_pct = op.get('y_percent', 0)
+                
+                # 转义单引号
+                value_escaped = locator_value.replace("'", "\\'") if locator_value else ''
+                
+                if locator_type == 'text':
+                    # 文本定位（最稳定）
+                    script_lines.append(f"    # 步骤{step_num}: 点击 '{element_desc}' (文本定位)")
+                    if locator_attr == 'description':
+                        script_lines.append(f"    safe_click(d, d(description='{value_escaped}'))")
+                    elif locator_attr == 'descriptionContains':
+                        script_lines.append(f"    safe_click(d, d(descriptionContains='{value_escaped}'))")
+                    elif locator_attr == 'textContains':
+                        script_lines.append(f"    safe_click(d, d(textContains='{value_escaped}'))")
+                    else:
+                        script_lines.append(f"    safe_click(d, d(text='{value_escaped}'))")
+                elif locator_type == 'id':
+                    # ID 定位（稳定）
+                    script_lines.append(f"    # 步骤{step_num}: 点击 '{element_desc}' (ID定位)")
+                    script_lines.append(f"    safe_click(d, d(resourceId='{value_escaped}'))")
+                elif locator_type == 'percent':
+                    # 百分比定位（跨分辨率兼容）
+                    script_lines.append(f"    # 步骤{step_num}: 点击 '{element_desc}' (百分比定位)")
+                    script_lines.append(f"    click_by_percent(d, {x_pct}, {y_pct})")
                 else:
-                    continue  # 无效操作，跳过
-                    
-                script_lines.append("    time.sleep(0.5)  # 等待响应")
+                    # 兼容旧格式
+                    ref = op.get('ref', '')
+                    if ref:
+                        ref_escaped = ref.replace("'", "\\'")
+                        script_lines.append(f"    # 步骤{step_num}: 点击 '{ref}'")
+                        script_lines.append(f"    safe_click(d, d(text='{ref_escaped}'))")
+                    else:
+                        continue
+                
+                script_lines.append("    time.sleep(0.5)")
                 script_lines.append("    ")
             
             elif action == 'input':
                 text = op.get('text', '')
-                ref = op.get('ref', '')
-                has_coords = 'x' in op and 'y' in op
-                has_percent = 'x_percent' in op and 'y_percent' in op
-                
-                # 判断 ref 是否为坐标格式
-                is_coords_ref = ref.startswith('coords_') or ref.startswith('coords:')
-                
-                # 优先使用 ID，其次百分比，最后坐标
-                if ref and not is_coords_ref and (':id/' in ref or ref.startswith('com.')):
-                    # 完整格式的 resource-id
-                    script_lines.append(f"    # 步骤{step_num}: 输入文本 '{text}' (ID定位)")
-                    script_lines.append(f"    d(resourceId='{ref}').set_text('{text}')")
-                elif ref and not is_coords_ref and not has_coords:
-                    # 简短格式的 resource-id（不包含 com. 或 :id/）
-                    script_lines.append(f"    # 步骤{step_num}: 输入文本 '{text}' (ID定位)")
-                    script_lines.append(f"    d(resourceId='{ref}').set_text('{text}')")
-                elif has_percent:
-                    x_pct = op['x_percent']
-                    y_pct = op['y_percent']
-                    script_lines.append(f"    # 步骤{step_num}: 点击后输入 (百分比定位)")
+                locator_type = op.get('locator_type', '')
+                locator_value = op.get('locator_value', '')
+                x_pct = op.get('x_percent', 0)
+                y_pct = op.get('y_percent', 0)
+                
+                text_escaped = text.replace("'", "\\'")
+                value_escaped = locator_value.replace("'", "\\'") if locator_value else ''
+                
+                if locator_type == 'id':
+                    script_lines.append(f"    # 步骤{step_num}: 输入 '{text}' (ID定位)")
+                    script_lines.append(f"    d(resourceId='{value_escaped}').set_text('{text_escaped}')")
+                elif locator_type == 'class':
+                    script_lines.append(f"    # 步骤{step_num}: 输入 '{text}' (类名定位)")
+                    script_lines.append(f"    d(className='android.widget.EditText').set_text('{text_escaped}')")
+                elif x_pct > 0 and y_pct > 0:
+                    script_lines.append(f"    # 步骤{step_num}: 点击后输入 '{text}'")
                     script_lines.append(f"    click_by_percent(d, {x_pct}, {y_pct})")
-                    script_lines.append(f"    time.sleep(0.3)")
-                    script_lines.append(f"    d.send_keys('{text}')")
-                elif has_coords:
-                    script_lines.append(f"    # 步骤{step_num}: 点击坐标后输入 (⚠️ 可能不兼容其他分辨率)")
-                    script_lines.append(f"    d.click({op['x']}, {op['y']})")
-                    script_lines.append(f"    time.sleep(0.3)")
-                    script_lines.append(f"    d.send_keys('{text}')")
+                    script_lines.append("    time.sleep(0.3)")
+                    script_lines.append(f"    d.send_keys('{text_escaped}')")
                 else:
-                    # 兜底：无法识别的格式，跳过
-                    continue
+                    # 兼容旧格式
+                    ref = op.get('ref', '')
+                    if ref:
+                        script_lines.append(f"    # 步骤{step_num}: 输入 '{text}'")
+                        script_lines.append(f"    d(resourceId='{ref}').set_text('{text_escaped}')")
+                    else:
+                        continue
+                
                 script_lines.append("    time.sleep(0.5)")
                 script_lines.append("    ")
             
             elif action == 'long_press':
-                ref = op.get('ref', '')
-                element = op.get('element', '')
+                locator_type = op.get('locator_type', '')
+                locator_value = op.get('locator_value', '')
+                locator_attr = op.get('locator_attr', 'text')
+                element_desc = op.get('element_desc', '')
                 duration = op.get('duration', 1.0)
-                has_coords = 'x' in op and 'y' in op
-                has_percent = 'x_percent' in op and 'y_percent' in op
-                
-                # 判断 ref 是否为坐标格式
-                is_coords_ref = ref.startswith('coords_') or ref.startswith('coords:')
-                is_percent_ref = ref.startswith('percent_')
-                
-                # 优先级：文本 > ID > 百分比 > 坐标
-                if ref and not is_coords_ref and not is_percent_ref and ':' not in ref:
-                    # 1️⃣ 使用文本（最稳定，优先）
-                    script_lines.append(f"    # 步骤{step_num}: 长按文本 '{ref}' (文本定位，最稳定)")
-                    script_lines.append(f"    d(text='{ref}').long_click(duration={duration})")
-                elif ref and ':' in ref and not is_coords_ref and not is_percent_ref:
-                    # 1️⃣-b 使用文本（Android 的 text:xxx 或 description:xxx 格式）
-                    actual_text = ref.split(':', 1)[1] if ':' in ref else ref
-                    script_lines.append(f"    # 步骤{step_num}: 长按文本 '{actual_text}' (文本定位，最稳定)")
-                    script_lines.append(f"    d(text='{actual_text}').long_click(duration={duration})")
-                elif ref and (':id/' in ref or ref.startswith('com.')):
-                    # 2️⃣ 使用 resource-id（稳定）
-                    script_lines.append(f"    # 步骤{step_num}: 长按元素 (ID定位)")
-                    script_lines.append(f"    d(resourceId='{ref}').long_click(duration={duration})")
-                elif has_percent:
-                    # 使用百分比
-                    x_pct = op['x_percent']
-                    y_pct = op['y_percent']
-                    desc = f" ({element})" if element else ""
-                    script_lines.append(f"    # 步骤{step_num}: 长按位置{desc} (百分比定位，跨分辨率兼容)")
-                    script_lines.append(f"    long_press_by_percent(d, {x_pct}, {y_pct}, duration={duration})  # 原坐标: ({op.get('x', '?')}, {op.get('y', '?')})")
-                elif has_coords:
-                    # 坐标兜底
-                    desc = f" ({element})" if element else ""
-                    script_lines.append(f"    # 步骤{step_num}: 长按坐标{desc} (⚠️ 坐标定位，可能不兼容其他分辨率)")
-                    script_lines.append(f"    d.long_click({op['x']}, {op['y']}, duration={duration})")
+                x_pct = op.get('x_percent', 0)
+                y_pct = op.get('y_percent', 0)
+                
+                value_escaped = locator_value.replace("'", "\\'") if locator_value else ''
+                
+                if locator_type == 'text':
+                    script_lines.append(f"    # 步骤{step_num}: 长按 '{element_desc}'")
+                    if locator_attr == 'description':
+                        script_lines.append(f"    d(description='{value_escaped}').long_click(duration={duration})")
+                    else:
+                        script_lines.append(f"    d(text='{value_escaped}').long_click(duration={duration})")
+                elif locator_type == 'id':
+                    script_lines.append(f"    # 步骤{step_num}: 长按 '{element_desc}'")
+                    script_lines.append(f"    d(resourceId='{value_escaped}').long_click(duration={duration})")
+                elif locator_type == 'percent':
+                    script_lines.append(f"    # 步骤{step_num}: 长按 '{element_desc}'")
+                    script_lines.append(f"    long_press_by_percent(d, {x_pct}, {y_pct}, duration={duration})")
                 else:
-                    continue
-                    
-                script_lines.append("    time.sleep(0.5)  # 等待响应")
+                    # 兼容旧格式
+                    ref = op.get('ref', '')
+                    if ref:
+                        ref_escaped = ref.replace("'", "\\'")
+                        script_lines.append(f"    # 步骤{step_num}: 长按 '{ref}'")
+                        script_lines.append(f"    d(text='{ref_escaped}').long_click(duration={duration})")
+                    else:
+                        continue
+                
+                script_lines.append("    time.sleep(0.5)")
                 script_lines.append("    ")
             
             elif action == 'swipe':
                 direction = op.get('direction', 'up')
                 script_lines.append(f"    # 步骤{step_num}: 滑动 {direction}")
-                script_lines.append(f"    d.swipe_ext('{direction}')")
+                script_lines.append(f"    swipe_direction(d, '{direction}')")
                 script_lines.append("    time.sleep(0.5)")
                 script_lines.append("    ")
             
@@ -3345,8 +3552,16 @@ class BasicMobileToolsLite:
             "    print('✅ 测试完成')",
             "",
             "",
+            "# ========== 直接运行入口 ==========",
             "if __name__ == '__main__':",
-            "    test_main()",
+            "    # 直接运行时，手动创建设备连接",
+            "    _d = u2.connect()",
+            "    _d.implicitly_wait(10)",
+            "    _d.app_start(PACKAGE_NAME)",
+            "    time.sleep(LAUNCH_WAIT)",
+            "    if CLOSE_AD_ON_LAUNCH:",
+            "        close_ad_if_exists(_d)",
+            f"    test_{safe_name}(_d)",
         ])
         
         script = '\n'.join(script_lines)
@@ -3355,8 +3570,11 @@ class BasicMobileToolsLite:
         output_dir = Path("tests")
         output_dir.mkdir(exist_ok=True)
         
+        # 确保文件名符合 pytest 规范（以 test_ 开头）
         if not filename.endswith('.py'):
             filename = f"{filename}.py"
+        if not filename.startswith('test_'):
+            filename = f"test_{filename}"
         
         file_path = output_dir / filename
         file_path.write_text(script, encoding='utf-8')
@@ -3364,7 +3582,7 @@ class BasicMobileToolsLite:
         return {
             "success": True,
             "file_path": str(file_path),
-            "message": f"✅ 脚本已生成: {file_path}",
+            "message": f"✅ 脚本已生成: {file_path}\n💡 运行方式: pytest {file_path} -v 或 python {file_path}",
             "operations_count": len(self.operation_history),
             "preview": script[:500] + "..."
         }
@@ -3533,10 +3751,28 @@ class BasicMobileToolsLite:
         try:
             import xml.etree.ElementTree as ET
             
-            # ========== 第1步：控件树查找关闭按钮 ==========
+            # ========== 第0步：先检测是否有弹窗 ==========
             xml_string = self.client.u2.dump_hierarchy(compressed=False)
             root = ET.fromstring(xml_string)
             
+            screen_width = self.client.u2.info.get('displayWidth', 1440)
+            screen_height = self.client.u2.info.get('displayHeight', 3200)
+            
+            popup_bounds, popup_confidence = self._detect_popup_with_confidence(
+                root, screen_width, screen_height
+            )
+            
+            # 如果没有检测到弹窗，直接返回"无弹窗"
+            if popup_bounds is None or popup_confidence < 0.5:
+                result["success"] = True
+                result["method"] = None
+                result["message"] = "ℹ️ 当前页面未检测到弹窗，无需关闭"
+                result["popup_detected"] = False
+                result["popup_confidence"] = popup_confidence
+                return result
+            
+            # ========== 第1步：控件树查找关闭按钮 ==========
+            
             # 关闭按钮的常见特征
             close_keywords = ['关闭', '跳过', '×', 'X', 'x', 'close', 'skip', '取消']
             close_content_desc = ['关闭', '跳过', 'close', 'skip', 'dismiss']
@@ -3615,12 +3851,6 @@ class BasicMobileToolsLite:
                 cx, cy = best['center']
                 bounds = best['bounds']
                 
-                # 点击前截图（用于自动学习）
-                pre_screenshot = None
-                if auto_learn:
-                    pre_result = self.take_screenshot(description="关闭前", compress=False)
-                    pre_screenshot = pre_result.get("screenshot_path")
-                
                 # 点击（click_at_coords 内部已包含应用状态检查和自动返回）
                 click_result = self.click_at_coords(cx, cy)
                 time.sleep(0.5)
@@ -3650,17 +3880,11 @@ class BasicMobileToolsLite:
                 result["message"] = msg
                 result["app_check"] = app_check
                 result["return_to_app"] = return_result
-                
-                # 自动学习：检查这个 X 是否已在模板库，不在就添加
-                if auto_learn and pre_screenshot:
-                    learn_result = self._auto_learn_template(pre_screenshot, bounds)
-                    if learn_result:
-                        result["learned_template"] = learn_result
-                        result["message"] += f"\n📚 自动学习: {learn_result}"
+                result["tip"] = "💡 建议调用 mobile_screenshot_with_som 确认弹窗是否已关闭"
                 
                 return result
             
-            # ========== 第2步：模板匹配 ==========
+            # ========== 第2步：模板匹配（自动执行，不需要 AI 介入）==========
             screenshot_path = None
             try:
                 from .template_matcher import TemplateMatcher
@@ -3679,16 +3903,14 @@ class BasicMobileToolsLite:
                         x_pct = best["percent"]["x"]
                         y_pct = best["percent"]["y"]
                         
-                        # 点击（click_by_percent 内部已包含应用状态检查和自动返回）
+                        # 点击
                         click_result = self.click_by_percent(x_pct, y_pct)
                         time.sleep(0.5)
                         
-                        # 🎯 再次检查应用状态（确保弹窗去除没有导致应用跳转）
                         app_check = self._check_app_switched()
                         return_result = None
                         
                         if app_check['switched']:
-                            # 应用已跳转，说明弹窗去除失败，尝试返回目标应用
                             return_result = self._return_to_target_app()
                         
                         result["success"] = True
@@ -3699,12 +3921,9 @@ class BasicMobileToolsLite:
                               f"   位置: ({x_pct:.1f}%, {y_pct:.1f}%)"
                         
                         if app_check['switched']:
-                            msg += f"\n⚠️ 应用已跳转，说明弹窗去除失败"
+                            msg += f"\n⚠️ 应用已跳转"
                             if return_result:
-                                if return_result['success']:
-                                    msg += f"\n{return_result['message']}"
-                                else:
-                                    msg += f"\n❌ 自动返回失败: {return_result['message']}"
+                                msg += f"\n{return_result['message']}"
                         
                         result["message"] = msg
                         result["app_check"] = app_check
@@ -3716,17 +3935,12 @@ class BasicMobileToolsLite:
             except Exception:
                 pass  # 模板匹配失败，继续下一步
             
-            # ========== 第3步：返回截图供 AI 分析 ==========
-            if not screenshot_path:
-                screenshot_result = self.take_screenshot(description="需要AI分析", compress=True)
-            
+            # ========== 第3步：控件树和模板匹配都失败，提示 AI 使用视觉识别 ==========
             result["success"] = False
+            result["fallback"] = "vision"
             result["method"] = None
-            result["message"] = "❌ 控件树和模板匹配都未找到关闭按钮\n" \
-                               "📸 已截图，请 AI 分析图片中的 X 按钮位置\n" \
-                               "💡 找到后使用 mobile_click_by_percent(x%, y%) 点击"
-            result["screenshot"] = screenshot_result if not screenshot_path else {"screenshot_path": screenshot_path}
-            result["need_ai_analysis"] = True
+            result["popup_detected"] = True
+            result["message"] = "⚠️ 控件树和模板匹配都未找到关闭按钮，请调用 mobile_screenshot_with_som 截图后用 click_by_som 点击"
             
             return result