diff --git a/src/bw_learner/expression_reflect_tracker.py b/src/bw_learner/expression_reflect_tracker.py
index 1ab56d3f..e4173c6a 100644
--- a/src/bw_learner/expression_reflect_tracker.py
+++ b/src/bw_learner/expression_reflect_tracker.py
@@ -1,16 +1,23 @@
+import json
+import re
+import time
 from typing import TYPE_CHECKING, Optional
 
-import time
+from json_repair import repair_json
 
-from src.common.logger import get_logger
+from src.chat.utils.chat_message_builder import (
+    build_readable_messages,
+    get_raw_msg_by_timestamp_with_chat,
+)
 from src.common.database.database import get_db_session
-from src.llm_models.utils_model import LLMRequest
+from src.common.logger import get_logger
 from src.config.config import model_config
+from src.llm_models.utils_model import LLMRequest
+from src.prompt.prompt_manager import prompt_manager
 
 if TYPE_CHECKING:
     from src.common.data_models.expression_data_model import MaiExpression
 
-# TODO: 这个LLMRequest实例被更优雅的方式替换掉
 judge_model = LLMRequest(model_set=model_config.model_task_config.tool_use, request_type="reflect.tracker")
 
 logger = get_logger("reflect_tracker")
@@ -57,4 +64,119 @@ class ReflectTracker:
             self._reset_tracker()
             return True
         
-        # TODO: 完成追踪检查逻辑
\ No newline at end of file
+        # 获取消息列表
+        msg_list = get_raw_msg_by_timestamp_with_chat(
+            chat_id=self.session_id,
+            timestamp_start=self.tracking_start_time,
+            timestamp_end=time.time(),
+        )
+
+        current_msg_count = len(msg_list)
+
+        # 检查消息数量是否超限
+        if current_msg_count > self.max_msg_count:
+            logger.info(f"ReflectTracker for expr {self.expression.item_id} timed out (message count).")
+            self._reset_tracker()
+            return True
+
+        # 如果没有新消息，跳过本次检查
+        if current_msg_count <= self.last_check_msg_count:
+            return False
+
+        self.last_check_msg_count = current_msg_count
+
+        # 构建上下文
+        context_block = build_readable_messages(
+            msg_list,
+            replace_bot_name=True,
+            timestamp_mode="relative",
+            read_mark=0.0,
+            show_actions=False,
+        )
+
+        # LLM 判断
+        try:
+            prompt_template = prompt_manager.get_prompt("reflect_judge")
+            prompt_template.add_context("situation", str(self.expression.situation))
+            prompt_template.add_context("style", str(self.expression.style))
+            prompt_template.add_context("context_block", context_block)
+            prompt = await prompt_manager.render_prompt(prompt_template)
+
+            logger.info(f"ReflectTracker LLM Prompt: {prompt}")
+
+            response, _ = await judge_model.generate_response_async(prompt, temperature=0.1)
+
+            logger.info(f"ReflectTracker LLM Response: {response}")
+
+            # 解析 JSON 响应
+            json_pattern = r"```json\s*(.*?)\s*```"
+            matches = re.findall(json_pattern, response, re.DOTALL)
+            if not matches:
+                matches = [response]
+
+            json_obj = json.loads(repair_json(matches[0]))
+            judgment = json_obj.get("judgment")
+
+            if judgment == "Approve":
+                self._update_expression(checked=True, rejected=False, modified_by="ai")
+                logger.info(f"Expression {self.expression.item_id} approved by operator.")
+                self._reset_tracker()
+                return True
+
+            elif judgment == "Reject":
+                corrected_situation = json_obj.get("corrected_situation")
+                corrected_style = json_obj.get("corrected_style")
+                has_update = bool(corrected_situation or corrected_style)
+
+                update_kwargs = {"checked": True, "modified_by": "ai"}
+                if corrected_situation:
+                    update_kwargs["situation"] = corrected_situation
+                if corrected_style:
+                    update_kwargs["style"] = corrected_style
+                if not has_update:
+                    update_kwargs["rejected"] = True
+                else:
+                    update_kwargs["rejected"] = False
+
+                self._update_expression(**update_kwargs)
+
+                if has_update:
+                    logger.info(
+                        f"Expression {self.expression.item_id} rejected and updated. "
+                        f"New situation: {corrected_situation}, New style: {corrected_style}"
+                    )
+                else:
+                    logger.info(
+                        f"Expression {self.expression.item_id} rejected but no correction provided, marked as rejected."
+                    )
+                self._reset_tracker()
+                return True
+
+            elif judgment == "Ignore":
+                logger.info(f"ReflectTracker for expr {self.expression.item_id} judged as Ignore.")
+                return False
+
+        except Exception as e:
+            logger.error(f"Error in ReflectTracker check: {e}")
+            return False
+
+        return False
+
+    def _update_expression(self, **kwargs):
+        """更新表达并持久化到数据库"""
+        if not self.expression:
+            return
+
+        # 更新内存中的表达对象
+        for key, value in kwargs.items():
+            if hasattr(self.expression, key):
+                setattr(self.expression, key, value)
+
+        # 持久化到数据库
+        try:
+            with get_db_session() as session:
+                db_expr = self.expression.to_db_instance()
+                session.merge(db_expr)
+                session.commit()
+        except Exception as e:
+            logger.error(f"Failed to persist expression update: {e}")
\ No newline at end of file