fix: attach tool-viewed media into model context

server/chat_flow_tool_loop.py

@@ -418,6 +418,7 @@ async def execute_tool_calls(*, web_terminal, tool_calls, sender, messages, clie
             ):
                 img_path = inj.get("path") if isinstance(inj, dict) else None
                 if img_path:
+                    tool_images = [img_path]
                     if metadata_payload is None:
                         metadata_payload = {}
                     metadata_payload["tool_image_path"] = img_path
@@ -436,6 +437,7 @@ async def execute_tool_calls(*, web_terminal, tool_calls, sender, messages, clie
             ):
                 video_path = inj.get("path") if isinstance(inj, dict) else None
                 if video_path:
+                    tool_videos = [video_path]
                     if metadata_payload is None:
                         metadata_payload = {}
                     metadata_payload["tool_video_path"] = video_path
@@ -443,6 +445,17 @@ async def execute_tool_calls(*, web_terminal, tool_calls, sender, messages, clie
                         'content': f'系统已记录视频路径（不再附带二进制数据）: {video_path}'
                     })
 
+        # 将工具结果即时组装为多模态消息，确保同一轮 tool loop 的下一次模型请求能真正看到图片/视频
+        if tool_images or tool_videos:
+            try:
+                tool_message_content = web_terminal.context_manager._build_content_with_images(
+                    str(tool_result_content or ""),
+                    tool_images or [],
+                    tool_videos or []
+                )
+            except Exception:
+                tool_message_content = tool_result_content
+
         # 立即保存工具结果
         web_terminal.context_manager.add_conversation(
             "tool",