diff --git a/metagpt/actions/ask_review.py b/metagpt/actions/ask_review.py
index 7eb553b7e..0d671648b 100644
--- a/metagpt/actions/ask_review.py
+++ b/metagpt/actions/ask_review.py
@@ -23,14 +23,15 @@ class ReviewConst:
 
 
 class AskReview(Action):
-    async def run(self, context: List[Message], plan: Plan = None, trigger: str = "task"):
-        logger.info("Current overall plan:")
-        logger.info(
-            "\n".join([f"{task.task_id}: {task.instruction}, is_finished: {task.is_finished}" for task in plan.tasks])
-        )
+    async def run(self, context: List[Message] = [], plan: Plan = None, trigger: str = "task"):
+        if plan:
+            logger.info("Current overall plan:")
+            logger.info(
+                "\n".join([f"{task.task_id}: {task.instruction}, is_finished: {task.is_finished}" for task in plan.tasks])
+            )
 
         logger.info("most recent context:")
-        latest_action = context[-1].cause_by if context[-1].cause_by else ""
+        latest_action = context[-1].cause_by if context and context[-1].cause_by else ""
         review_instruction = (
             ReviewConst.TASK_REVIEW_INSTRUCTION
             if trigger == ReviewConst.TASK_REVIEW_TRIGGER
diff --git a/metagpt/actions/write_analysis_code.py b/metagpt/actions/write_analysis_code.py
index aef86122b..c5f9c9166 100644
--- a/metagpt/actions/write_analysis_code.py
+++ b/metagpt/actions/write_analysis_code.py
@@ -12,7 +12,7 @@ import yaml
 from tenacity import retry, stop_after_attempt, wait_fixed
 
 from metagpt.actions import Action
-from metagpt.const import METAGPT_ROOT
+from metagpt.const import METAGPT_ROOT, TOOL_SCHEMA_PATH
 from metagpt.llm import LLM
 from metagpt.logs import logger
 from metagpt.prompts.ml_engineer import (
@@ -93,7 +93,7 @@ class WriteCodeByGenerate(BaseWriteAnalysisCode):
 class WriteCodeWithTools(BaseWriteAnalysisCode):
     """Write code with help of local available tools. Choose tools first, then generate code to use the tools"""
 
-    schema_path: Union[Path, str] = METAGPT_ROOT / "metagpt/tools/functions/schemas"
+    schema_path: Union[Path, str] = TOOL_SCHEMA_PATH
     available_tools: dict = {}
 
     def __init__(self, **kwargs):
diff --git a/metagpt/const.py b/metagpt/const.py
index 811ff9516..b1666e092 100644
--- a/metagpt/const.py
+++ b/metagpt/const.py
@@ -70,6 +70,7 @@ TMP = METAGPT_ROOT / "tmp"
 SOURCE_ROOT = METAGPT_ROOT / "metagpt"
 PROMPT_PATH = SOURCE_ROOT / "prompts"
 SKILL_DIRECTORY = SOURCE_ROOT / "skills"
+TOOL_SCHEMA_PATH = METAGPT_ROOT / "metagpt/tools/functions/schemas"
 
 
 # REAL CONSTS
diff --git a/metagpt/roles/code_interpreter.py b/metagpt/roles/code_interpreter.py
index 6bbd923e6..9b13d8dcb 100644
--- a/metagpt/roles/code_interpreter.py
+++ b/metagpt/roles/code_interpreter.py
@@ -13,6 +13,7 @@ from metagpt.utils.save_code import save_code_file
 
 
 class CodeInterpreter(Role):
+    auto_run: bool = True
     use_tools: bool = False
     make_udfs: bool = False  # whether to save user-defined functions
     execute_code: ExecutePyCode = Field(default_factory=ExecutePyCode, exclude=True)
@@ -22,12 +23,12 @@ class CodeInterpreter(Role):
         name="Charlie",
         profile="CodeInterpreter",
         goal="",
-        auto_run=False,
+        auto_run=True,
         use_tools=False,
         make_udfs=False,
         **kwargs,
     ):
-        super().__init__(name=name, profile=profile, goal=goal, use_tools=use_tools, make_udfs=make_udfs, **kwargs)
+        super().__init__(name=name, profile=profile, goal=goal, auto_run=auto_run, use_tools=use_tools, make_udfs=make_udfs, **kwargs)
         self._set_react_mode(react_mode="plan_and_act", auto_run=auto_run, use_tools=use_tools)
 
     @property
@@ -96,4 +97,4 @@ class CodeInterpreter(Role):
         logger.info("Plan completed. Now start to make tools ...")
         tool_maker = ToolMaker()
         for task in self.planner.plan.get_finished_tasks():
-            await tool_maker.make_tool(task.code, task.instruction, task.task_id)
+            await tool_maker.make_tool(code=task.code, instruction=task.instruction, task_id=task.task_id, auto_run=self.auto_run)
diff --git a/metagpt/roles/ml_engineer.py b/metagpt/roles/ml_engineer.py
index 639a517d6..cf903347d 100644
--- a/metagpt/roles/ml_engineer.py
+++ b/metagpt/roles/ml_engineer.py
@@ -12,7 +12,6 @@ from metagpt.utils.common import any_to_str
 
 
 class MLEngineer(CodeInterpreter):
-    auto_run: bool = False
     use_code_steps: bool = False
     use_udfs: bool = False
     data_desc: dict = {}
diff --git a/metagpt/roles/tool_maker.py b/metagpt/roles/tool_maker.py
index 7fec7b739..5453fd807 100644
--- a/metagpt/roles/tool_maker.py
+++ b/metagpt/roles/tool_maker.py
@@ -1,5 +1,6 @@
 from pydantic import Field
 
+from metagpt.actions.ask_review import AskReview
 from metagpt.actions.execute_code import ExecutePyCode
 from metagpt.actions.write_analysis_code import MakeTools
 from metagpt.logs import logger
@@ -10,7 +11,7 @@ from metagpt.utils.common import remove_comments
 class ToolMaker(Role):
     execute_code: ExecutePyCode = Field(default_factory=ExecutePyCode, exclude=True)
 
-    async def make_tool(self, code: str, instruction: str, task_id: str = ""):
+    async def make_tool(self, code: str, instruction: str, task_id: str = "", auto_run=True):
         if len(remove_comments(code).split("\n")) < 5:  # no need to consider trivial codes with fewer than 5 lines
             return
 
@@ -41,4 +42,12 @@ class ToolMaker(Role):
                 break
         # save successful tool code in udf
         if execute_success:
-            make_tools.save(tool_code)
+            _, confirmed = await self.ask_review(auto_run=auto_run)
+            if confirmed:
+                make_tools.save(tool_code)
+    
+    async def ask_review(self, auto_run: bool = True):
+        if not auto_run:
+            review, confirmed = await AskReview().run()
+            return review, confirmed
+        return "", True