diff --git a/metagpt/prompts/di/role_zero.py b/metagpt/prompts/di/role_zero.py
index f85128b38..2294fa731 100644
--- a/metagpt/prompts/di/role_zero.py
+++ b/metagpt/prompts/di/role_zero.py
@@ -220,20 +220,19 @@ However, you MUST respond to the user message by yourself directly, DON'T ask yo
 """
 
 REPORT_TO_HUMAN_PROMPT = """
-# Current Plan
-{plan_status}
+# Restrictions
+{requirements_constraints}
 
-Your have just finish a task, Use "RoleZero.reply_to_human" to report what you have done.
-The output format is :
-```json
-[
-    {{
-        "command_name": "RoleZero.reply_to_human",
-        "args": {{
-            "content": ""
-        }}
-    }}
-]
-```
+Your have just finish all tasks, Use "RoleZero.reply_to_human" to answer the user requirements.
+Report to human what you have done. Do Not ouput any other format.
+Your reply is:
 
 """
+SUMMARY_PROMPY = """
+# Restrictions
+{requirements_constraints}
+
+You have just completed some tasks. 
+Summarize the tasks you have accomplished without including detailed information. 
+If there are any deliverables, list their descriptions and provide their file paths.
+"""
diff --git a/metagpt/roles/di/role_zero.py b/metagpt/roles/di/role_zero.py
index 727360ae9..48967d7ef 100644
--- a/metagpt/roles/di/role_zero.py
+++ b/metagpt/roles/di/role_zero.py
@@ -28,6 +28,7 @@ from metagpt.prompts.di.role_zero import (
     REGENERATE_PROMPT,
     REPORT_TO_HUMAN_PROMPT,
     ROLE_INSTRUCTION,
+    SUMMARY_PROMPY,
     SYSTEM_PROMPT,
     THOUGHT_GUIDANCE,
 )
@@ -68,6 +69,8 @@ class RoleZero(Role):
     react_mode: Literal["react"] = "react"
     max_react_loop: int = 20  # used for react mode
 
+    # Summary Mode
+    use_summary: bool = True
     # Tools
     tools: list[str] = []  # Use special symbol ["<all>"] to indicate use of all registered tools
     tool_recommender: Optional[ToolRecommender] = None
@@ -87,8 +90,6 @@ class RoleZero(Role):
     use_fixed_sop: bool = False
     requirements_constraints: str = ""  # the constraints in user requirements
 
-    command_history: list[str] = []
-
     @model_validator(mode="after")
     def set_plan_and_tool(self) -> "RoleZero":
         # We force using this parameter for DataAnalyst
@@ -247,26 +248,24 @@ class RoleZero(Role):
         logger.info(f"Commands outputs: \n{outputs}")
         self.rc.memory.add(UserMessage(content=outputs))
 
-        # Report what is done when finishing the task.
-        current_command_list = [command["command_name"] for command in commands]
-        self.command_history.extend(current_command_list)
-        if self.check_whether_report_to_human():
-            memory = self.rc.memory.get(self.memory_k)
-            memory = await self.parse_browser_actions(memory)
-            memory = self.parse_images(memory)
-            plan_status, _ = self._get_plan_status()
-            prompt = REPORT_TO_HUMAN_PROMPT.format(plan_status=plan_status)
-            req = self.llm.format_msg(memory + [UserMessage(content=prompt)])
-            respond_command = await self.llm.aask(msg=req)
-            commands, ok = await self._parse_commands(respond_command)
-            if ok and len(commands) == 1 and commands[0]["command_name"] == "RoleZero.reply_to_human":
-                cmd = commands[0]
-                report_result = await self.reply_to_human(cmd["args"])
-                self.rc.memory.add(AIMessage(content=respond_command))
-                self.rc.memory.add(UserMessage(content=report_result))
-                self.command_history.append("RoleZero.reply_to_human")
-                logger.info(f"Commands outputs: \n{report_result}")
-                outputs += report_result
+        if any(["end" == command["command_name"] for command in commands]):
+            # Ensure reply to the human before the "end" command is executed.
+            if all(["reply_to_human" not in memory.content for memory in self.get_memories(k=5)]):
+                memory = self.rc.memory.get(self.memory_k)
+                reply_to_human_prompt = REPORT_TO_HUMAN_PROMPT.format(
+                    requirements_constraints=self.requirements_constraints,
+                )
+                reply_content = await self.llm.aask(self.llm.format_msg(memory + [UserMessage(reply_to_human_prompt)]))
+                await self.reply_to_human(content=reply_content)
+                self.rc.memory.add(AIMessage(content=reply_content, cause_by=RunCommand))
+
+            # Summary of the Completed Task and Deliverables
+            if self.use_summary:
+                memory = self.rc.memory.get(self.memory_k)
+                summary_prompt = SUMMARY_PROMPY.format(
+                    requirements_constraints=self.requirements_constraints,
+                )
+                outputs = await self.llm.aask(self.llm.format_msg(memory + [UserMessage(summary_prompt)]))
 
         return AIMessage(
             content=f"I have finished the task, please mark my task as finished. Outputs: {outputs}",
@@ -274,20 +273,6 @@ class RoleZero(Role):
             cause_by=RunCommand,
         )
 
-    def check_whether_report_to_human(self):
-        """ "Check whether add reply to human command when finish current task"""
-        interaction_with_human = ["RoleZero.ask_human", "RoleZero.reply_to_human"]
-        plan_end_action = ["Plan.finish_current_task", "end"]
-        flag = 0
-        for command_name in self.command_history[::-1]:
-            if command_name in plan_end_action:
-                flag |= 1
-            elif command_name in interaction_with_human:
-                flag |= 2
-            else:
-                break
-        return flag == 1
-
     async def _react(self) -> Message:
         # NOTE: Diff 1: Each time landing here means news is observed, set todo to allow news processing in _think
         self._set_state(0)
diff --git a/metagpt/roles/di/swe_agent.py b/metagpt/roles/di/swe_agent.py
index e90fc3045..d54c9dd44 100644
--- a/metagpt/roles/di/swe_agent.py
+++ b/metagpt/roles/di/swe_agent.py
@@ -69,11 +69,10 @@ class SWEAgent(RoleZero):
 
         This function is specifically added for SWE bench evaluation.
         """
-        # only import when evaluation is needed
+        # If todo switches to None, it indicates that this is the final round of reactions, and the Swe-Agent will stop. Use git diff to store any changes made.
         if not self.rc.todo:
             from metagpt.tools.swe_agent_commands.swe_agent_utils import extract_patch
 
-            # swe agent have been stop. it means  'end' or other command which can stop the swe agent have been executed
             try:
                 diff_output = await self.terminal.run("git diff --cached")
                 clear_diff = extract_patch(diff_output)
diff --git a/metagpt/roles/di/team_leader.py b/metagpt/roles/di/team_leader.py
index 97100f295..112ca5a84 100644
--- a/metagpt/roles/di/team_leader.py
+++ b/metagpt/roles/di/team_leader.py
@@ -30,6 +30,8 @@ class TeamLeader(RoleZero):
 
     experience_retriever: Annotated[ExpRetriever, Field(exclude=True)] = SimpleExpRetriever()
 
+    use_summary: bool = False
+
     def _update_tool_execution(self):
         self.tool_execution_map.update(
             {