diff --git a/metagpt/actions/action_node.py b/metagpt/actions/action_node.py
index 092dd5755..4376e09ed 100644
--- a/metagpt/actions/action_node.py
+++ b/metagpt/actions/action_node.py
@@ -4,6 +4,9 @@
 @Time    : 2023/12/11 18:45
 @Author  : alexanderwu
 @File    : action_node.py
+
+NOTE: You should use typing.List instead of list to do type annotation. Because in the markdown extraction process,
+  we can use typing to extract the type of the node, but we cannot use built-in list to extract.
 """
 import json
 from typing import Any, Dict, Generic, List, Optional, Tuple, Type, TypeVar
@@ -41,10 +44,10 @@ Fill in the above nodes based on the format example.
 """
 
 
-def dict_to_markdown(d, prefix="-", postfix="\n"):
+def dict_to_markdown(d, prefix="##", kv_sep="\n", postfix="\n"):
     markdown_str = ""
     for key, value in d.items():
-        markdown_str += f"{prefix} {key}: {value}{postfix}"
+        markdown_str += f"{prefix}{key}{kv_sep}{value}{postfix}"
     return markdown_str
 
 
diff --git a/metagpt/actions/design_api_an.py b/metagpt/actions/design_api_an.py
index 0a303cdd5..7d6802381 100644
--- a/metagpt/actions/design_api_an.py
+++ b/metagpt/actions/design_api_an.py
@@ -5,6 +5,8 @@
 @Author  : alexanderwu
 @File    : design_api_an.py
 """
+from typing import List
+
 from metagpt.actions.action_node import ActionNode
 from metagpt.logs import logger
 from metagpt.utils.mermaid import MMC1, MMC2
@@ -22,7 +24,7 @@ PROJECT_NAME = ActionNode(
 
 FILE_LIST = ActionNode(
     key="File list",
-    expected_type=list[str],
+    expected_type=List[str],
     instruction="Only need relative paths. ALWAYS write a main.py or app.py here",
     example=["main.py", "game.py"],
 )
diff --git a/metagpt/actions/project_management_an.py b/metagpt/actions/project_management_an.py
index 6208c1051..215a67202 100644
--- a/metagpt/actions/project_management_an.py
+++ b/metagpt/actions/project_management_an.py
@@ -5,26 +5,28 @@
 @Author  : alexanderwu
 @File    : project_management_an.py
 """
+from typing import List
+
 from metagpt.actions.action_node import ActionNode
 from metagpt.logs import logger
 
 REQUIRED_PYTHON_PACKAGES = ActionNode(
     key="Required Python packages",
-    expected_type=list[str],
+    expected_type=List[str],
     instruction="Provide required Python packages in requirements.txt format.",
     example=["flask==1.1.2", "bcrypt==3.2.0"],
 )
 
 REQUIRED_OTHER_LANGUAGE_PACKAGES = ActionNode(
     key="Required Other language third-party packages",
-    expected_type=list[str],
+    expected_type=List[str],
     instruction="List down the required packages for languages other than Python.",
     example=["No third-party dependencies required"],
 )
 
 LOGIC_ANALYSIS = ActionNode(
     key="Logic Analysis",
-    expected_type=list[list[str]],
+    expected_type=List[List[str]],
     instruction="Provide a list of files with the classes/methods/functions to be implemented, "
     "including dependency analysis and imports.",
     example=[
@@ -35,7 +37,7 @@ LOGIC_ANALYSIS = ActionNode(
 
 TASK_LIST = ActionNode(
     key="Task list",
-    expected_type=list[str],
+    expected_type=List[str],
     instruction="Break down the tasks into a list of filenames, prioritized by dependency order.",
     example=["game.py", "main.py"],
 )
diff --git a/metagpt/actions/write_prd_an.py b/metagpt/actions/write_prd_an.py
index 8698c739f..d58d72f64 100644
--- a/metagpt/actions/write_prd_an.py
+++ b/metagpt/actions/write_prd_an.py
@@ -5,6 +5,7 @@
 @Author  : alexanderwu
 @File    : write_prd_an.py
 """
+from typing import List
 
 from metagpt.actions.action_node import ActionNode
 from metagpt.logs import logger
@@ -39,14 +40,14 @@ PROJECT_NAME = ActionNode(
 
 PRODUCT_GOALS = ActionNode(
     key="Product Goals",
-    expected_type=list[str],
+    expected_type=List[str],
     instruction="Provide up to three clear, orthogonal product goals.",
     example=["Create an engaging user experience", "Improve accessibility, be responsive", "More beautiful UI"],
 )
 
 USER_STORIES = ActionNode(
     key="User Stories",
-    expected_type=list[str],
+    expected_type=List[str],
     instruction="Provide up to 3 to 5 scenario-based user stories.",
     example=[
         "As a player, I want to be able to choose difficulty levels",
@@ -59,7 +60,7 @@ USER_STORIES = ActionNode(
 
 COMPETITIVE_ANALYSIS = ActionNode(
     key="Competitive Analysis",
-    expected_type=list[str],
+    expected_type=List[str],
     instruction="Provide 5 to 7 competitive products.",
     example=[
         "2048 Game A: Simple interface, lacks responsive features",
@@ -98,7 +99,7 @@ REQUIREMENT_ANALYSIS = ActionNode(
 
 REQUIREMENT_POOL = ActionNode(
     key="Requirement Pool",
-    expected_type=list[list[str]],
+    expected_type=List[List[str]],
     instruction="List down the top-5 requirements with their priority (P0, P1, P2).",
     example=[["P0", "The main code ..."], ["P0", "The game algorithm ..."]],
 )
diff --git a/metagpt/actions/write_review.py b/metagpt/actions/write_review.py
new file mode 100644
index 000000000..13690a1a5
--- /dev/null
+++ b/metagpt/actions/write_review.py
@@ -0,0 +1,38 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+"""
+@Author  : alexanderwu
+@File    : write_review.py
+"""
+from typing import List
+
+from metagpt.actions import Action
+from metagpt.actions.action_node import ActionNode
+
+REVIEW = ActionNode(
+    key="Review",
+    expected_type=List[str],
+    instruction="Act as an experienced Reviewer and review the given output. Ask a series of critical questions, "
+    "concisely and clearly, to help the writer improve their work.",
+    example=[
+        "This is a good PRD, but I think it can be improved by adding more details.",
+    ],
+)
+
+LGTM = ActionNode(
+    key="LGTM",
+    expected_type=str,
+    instruction="LGTM/LBTM. If the output is good enough, give a LGTM (Looks Good To Me) to the writer, "
+    "else LBTM (Looks Bad To Me).",
+    example="LGTM",
+)
+
+WRITE_REVIEW_NODE = ActionNode.from_children("WRITE_REVIEW_NODE", [REVIEW, LGTM])
+
+
+class WriteReview(Action):
+    """This class allows LLM to further mine noteworthy details based on specific "##TOPIC"(discussion topic) and
+    "##RECORD" (discussion records), thereby deepening the discussion."""
+
+    async def run(self, context):
+        return await WRITE_REVIEW_NODE.fill(context=context, llm=self.llm, schema="markdown")
diff --git a/metagpt/utils/common.py b/metagpt/utils/common.py
index ab7a3d99e..e123e8fd9 100644
--- a/metagpt/utils/common.py
+++ b/metagpt/utils/common.py
@@ -23,7 +23,7 @@ import traceback
 import typing
 from pathlib import Path
 from typing import Any
-from typing import List, Tuple, Union
+from typing import List, Tuple, Union, get_args, get_origin
 
 import aiofiles
 import loguru
@@ -135,8 +135,31 @@ class OutputParser:
             parsed_data[block] = content
         return parsed_data
 
+    @staticmethod
+    def extract_content(text, tag="CONTENT"):
+        # Use regular expression to extract content between [CONTENT] and [/CONTENT]
+        extracted_content = re.search(rf"\[{tag}\](.*?)\[/{tag}\]", text, re.DOTALL)
+
+        if extracted_content:
+            return extracted_content.group(1).strip()
+        else:
+            return "No content found between [CONTENT] and [/CONTENT] tags."
+
+    @staticmethod
+    def is_supported_list_type(i):
+        origin = get_origin(i)
+        if origin is not List:
+            return False
+
+        args = get_args(i)
+        if args == (str,) or args == (Tuple[str, str],) or args == (List[str],):
+            return True
+
+        return False
+
     @classmethod
     def parse_data_with_mapping(cls, data, mapping):
+        data = cls.extract_content(text=data)
         block_dict = cls.parse_blocks(data)
         parsed_data = {}
         for block, content in block_dict.items():
diff --git a/tests/metagpt/actions/test_write_review.py b/tests/metagpt/actions/test_write_review.py
new file mode 100644
index 000000000..2d188b720
--- /dev/null
+++ b/tests/metagpt/actions/test_write_review.py
@@ -0,0 +1,53 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+"""
+@Time    : 2023/12/20 15:01
+@Author  : alexanderwu
+@File    : test_write_review.py
+"""
+import pytest
+
+from metagpt.actions.write_review import WriteReview
+
+CONTEXT = """
+{
+    "Language": "zh_cn",
+    "Programming Language": "Python",
+    "Original Requirements": "写一个简单的2048",
+    "Project Name": "game_2048",
+    "Product Goals": [
+        "创建一个引人入胜的用户体验",
+        "确保高性能",
+        "提供可定制的功能"
+    ],
+    "User Stories": [
+        "作为用户，我希望能够选择不同的难度级别",
+        "作为玩家，我希望在每局游戏结束后能看到我的得分"
+    ],
+    "Competitive Analysis": [
+        "Python Snake Game: 界面简单，缺乏高级功能"
+    ],
+    "Competitive Quadrant Chart": "quadrantChart\n    title \"Reach and engagement of campaigns\"\n    x-axis \"Low Reach\" --> \"High Reach\"\n    y-axis \"Low Engagement\" --> \"High Engagement\"\n    quadrant-1 \"我们应该扩展\"\n    quadrant-2 \"需要推广\"\n    quadrant-3 \"重新评估\"\n    quadrant-4 \"可能需要改进\"\n    \"Campaign A\": [0.3, 0.6]\n    \"Campaign B\": [0.45, 0.23]\n    \"Campaign C\": [0.57, 0.69]\n    \"Campaign D\": [0.78, 0.34]\n    \"Campaign E\": [0.40, 0.34]\n    \"Campaign F\": [0.35, 0.78]\n    \"Our Target Product\": [0.5, 0.6]",
+    "Requirement Analysis": "产品应该用户友好。",
+    "Requirement Pool": [
+        [
+            "P0",
+            "主要代码..."
+        ],
+        [
+            "P0",
+            "游戏算法..."
+        ]
+    ],
+    "UI Design draft": "基本功能描述，简单的风格和布局。",
+    "Anything UNCLEAR": "..."
+}
+"""
+
+
+@pytest.mark.asyncio
+async def test_write_review():
+    write_review = WriteReview()
+    review = await write_review.run(CONTEXT)
+    assert review.instruct_content
+    assert review.get("LGTM") in ["LGTM", "LBTM"]