diff --git a/utils/agent.py b/utils/agent.py
index 9270f4c..4ede8cf 100644
--- a/utils/agent.py
+++ b/utils/agent.py
@@ -169,11 +169,26 @@ class Agent:
         )
 
         async with self.agent.run_stream(
-            user_prompt=user_prompt, message_history=message_history
+            user_prompt=user_prompt,
+            message_history=message_history,
         ) as result:
-            async for chunk in result.stream_text(delta=True):  # 只返回新增内容
-                if chunk:
-                    yield chunk
+            async for part in result.stream_output():  # 全量结构化输出分片
+                match part.part_kind:
+                    case "text":  # 就文本分片拆分 content 和 thinking
+                        if not (part_content := part.content.strip()):
+                            continue
+                        if not part.is_reasoning:
+                            yield f"0:{part_content}"  # content
+                        else:
+                            yield f"1:{part_content}"  # thinking
+                    case "tool-call":  # 技能调用分片
+                        yield f"2:技能名称：{part.tool_name}，调用参数：{part.args}"
+                    case "tool-return":  # 技能返回分片
+                        yield f"3:技能结果：{part.content}"
+                    case "error":  # 错误分片
+                        yield f"4:{str(part.error)}"
+                    case _:
+                        continue
 
         self.agent_memory.create_new_messages(
             session_id=self.session_id,
diff --git a/产品需求文档AI生成/application/components/session.py b/产品需求文档AI生成/application/components/session.py
index c5a593f..dfd4606 100644
--- a/产品需求文档AI生成/application/components/session.py
+++ b/产品需求文档AI生成/application/components/session.py
@@ -1,6 +1,6 @@
 # -*- coding: utf-8 -*-
 """
-会话组件
+会话相关组件
 """
 import reflex
 from reflex.constants.colors import ColorType
@@ -33,12 +33,12 @@ def turn(turn: Turn) -> reflex.Component:
     """
     return reflex.box(
         reflex.box(
-            message_bubble(message=turn.input_message, color="mauve"),
+            message_bubble(message=turn.input, color="mauve"),
             text_align="right",
             margin_bottom="8px",
         ),
         reflex.box(
-            message_bubble(message=turn.output_message, color="accent"),
+            message_bubble(message=turn.output, color="accent"),
             text_align="left",
             margin_bottom="8px",
         ),
@@ -69,7 +69,7 @@ def input_bar() -> reflex.Component:
             reflex.form(
                 reflex.hstack(
                     reflex.input(
-                        name="input_message",
+                        name="input",
                         placeholder="请输入...",
                         flex="auto",
                     ),
@@ -83,7 +83,7 @@ def input_bar() -> reflex.Component:
                     margin="0 auto",
                     align_items="center",
                 ),
-                on_submit=State.adapt_input_message,
+                on_submit=State.adapt_input,
                 reset_on_submit=True,  # 提交后清空输入框
             ),
             reflex.text(
@@ -106,4 +106,4 @@ def input_bar() -> reflex.Component:
         background_color=reflex.color("mauve", 2),
         align="stretch",
         width="100%",
-    )
+    )  # reflex.center 等价 reflex.box(display="flex", align_items="center", justify_content="center")
diff --git a/产品需求文档AI生成/application/state.py b/产品需求文档AI生成/application/state.py
index 82570fa..9ae30fd 100644
--- a/产品需求文档AI生成/application/state.py
+++ b/产品需求文档AI生成/application/state.py
@@ -3,7 +3,7 @@
 应用状态管理模块
 """
 
-from typing import Any, AsyncGenerator, Dict, List
+from typing import Any, AsyncGenerator, Dict, List, Literal
 from uuid import uuid4
 
 from pydantic import BaseModel, Field
@@ -32,11 +32,22 @@ def retrieve_agent(state) -> Agent:
     return agents[current_session_name]
 
 
+class MessageBlock(BaseModel):
+    """消息块数据模型，包含类型和内容"""
+
+    type: Literal[
+        "thinking", "content", "skill_call", "skill_result", "skill_error"
+    ] = Field(..., description="类型")
+    content: str = Field(default="", description="内容")
+
+
 class Turn(BaseModel):
     """对话数据模型，包含用户输入消息和智能体输出消息"""
 
-    input_message: str = Field(..., description="用户输入的消息")
-    output_message: str = Field(default="", description="智能体输出的消息")
+    input: str = Field(..., description="用户输入的消息")
+    output: List[MessageBlock] = Field(
+        default_factory=list, description="智能体输出的消息"
+    )
 
 
 class Session(BaseModel):
@@ -153,43 +164,45 @@ class State(reflex.State):
         self.create_session_modal_is_open = is_open
 
     @reflex.event
-    async def adapt_input_message(self, form_data: dict[str, Any]) -> AsyncGenerator:
+    async def adapt_input(self, form_data: dict[str, Any]) -> AsyncGenerator:
         """
-        适配用户输入消息
+        适配用户输入
         :param form_data: 对话表单数据
         :return: AsyncGenerator
         """
-        input_message = form_data["input_message"].strip()
-        if not input_message:
+        input = form_data["input_message"].strip()
+        if not input:
             return
 
-        async for value in self.process_input_message(input_message=input_message):
+        async for value in self.process_input(input=input):
             yield value
 
-    async def process_input_message(self, input_message: str) -> AsyncGenerator:
+    async def process_input(self, input: str) -> AsyncGenerator:
         """
-        处理用户输入消息
-        :param input_message: 用户输入的消息
+        处理用户输入
+        :param input: 用户输入
         :return: AsyncGenerator
         """
         # 当前会话
         current_session = self.sessions[self.current_session_name]
-        # 将用户输入消息添加到当前会话对话列表
-        current_session.turns.append(
-            Turn(
-                input_message=input_message,
-            )
-        )
-
         # 当前会话正在处理
         current_session.is_processing = True
+        # 将用户输入添加到当前会话对话列表
+        current_session.turns.append(
+            Turn(
+                input=input,
+            )
+        )
+        # 当前对话
+        current_turn = current_session.turns[-1]
         yield
 
         agent = retrieve_agent(self)
-        async for chunk in agent.output_message_streamed(user_prompt=input_message):
+        async for chunk in agent.output_message_streamed(user_prompt=input):
             if not chunk:
                 continue
             current_session.turns[-1].output_message += chunk
+
             yield
 
         # 当前会话处理完成