优化：集成Evaluator智能体并支持日志文件序号功能

- 在Step 7之后添加Evaluator评分步骤，将step流程从8步扩展为9步
- 新增evaluator模块的__init__.py文件确保正确导入
- 优化WorkflowLogger支持外部传入的case_index序号，生成更规范的日志文件名
- MedicalWorkflow类新增case_index参数，支持批量处理时的文件标识
- 完善Evaluator agent在workflow中的集成，提供医生问诊质量的多维度评价

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

agent_system/evaluetor/__init__.py

@@ -0,0 +1,5 @@
+from .agent import Evaluator
+from .prompt import EvaluatorPrompt
+from .response_model import EvaluatorResult, EvaluationDimension
+
+__all__ = ["Evaluator", "EvaluatorPrompt", "EvaluatorResult", "EvaluationDimension"]

workflow/medical_workflow.py

@@ -11,7 +11,8 @@ class MedicalWorkflow:
     """
     
     def __init__(self, case_data: Dict[str, Any], model_type: str = "gpt-oss:latest", 
-                 llm_config: Optional[Dict] = None, max_steps: int = 30, log_dir: str = "logs"):
+                 llm_config: Optional[Dict] = None, max_steps: int = 30, log_dir: str = "logs",
+                 case_index: Optional[int] = None):
         """
         初始化医疗问诊工作流
         
@@ -21,6 +22,7 @@ class MedicalWorkflow:
             llm_config: 语言模型配置，默认为None
             max_steps: 最大执行步数，默认为30
             log_dir: 日志目录，默认为"logs"
+            case_index: 病例序号，用于日志文件命名
         """
         self.case_data = case_data
         self.model_type = model_type
@@ -30,7 +32,7 @@ class MedicalWorkflow:
         # 初始化核心组件
         self.task_manager = TaskManager()
         self.step_executor = StepExecutor(model_type=model_type, llm_config=self.llm_config)
-        self.logger = WorkflowLogger(case_data=case_data, log_dir=log_dir)
+        self.logger = WorkflowLogger(case_data=case_data, log_dir=log_dir, case_index=case_index)
         
         # 初始化工作流状态
         self.current_step = 0

workflow/step_executor.py

@@ -6,6 +6,7 @@ from agent_system.controller import TaskController
 from agent_system.prompter import Prompter
 from agent_system.inquirer import Inquirer
 from agent_system.virtual_patient import VirtualPatientAgent
+from agent_system.evaluetor import Evaluator
 from .task_manager import TaskManager, TaskPhase
 from .workflow_logger import WorkflowLogger
 
@@ -32,6 +33,7 @@ class StepExecutor:
         self.controller = TaskController(model_type=model_type, llm_config=self.llm_config)
         self.prompter = Prompter(model_type=model_type, llm_config=self.llm_config)
         self.virtual_patient = VirtualPatientAgent(model_type=model_type, llm_config=self.llm_config)
+        self.evaluator = Evaluator(model_type=model_type, llm_config=self.llm_config)
     
     def execute_step(self, 
                     step_num: int,
@@ -123,7 +125,13 @@ class StepExecutor:
             )
             step_result["doctor_question"] = doctor_question
             
-            # Step 8: 获取任务完成情况摘要
+            # Step 8: 使用Evaluator进行评分
+            evaluator_result = self._execute_evaluator(
+                step_num, logger, case_data, step_result
+            )
+            step_result["evaluator_result"] = evaluator_result
+            
+            # Step 9: 获取任务完成情况摘要
             step_result["task_completion_summary"] = task_manager.get_completion_summary()
             
             step_result["success"] = True
@@ -372,3 +380,73 @@ class StepExecutor:
             logger.log_error(step_num, "inquirer_error", error_msg)
             # 返回默认问题
             return "请您详细描述一下您的症状，包括什么时候开始的，有什么特点？"
+    
+    def _execute_evaluator(self, step_num: int, logger: WorkflowLogger, 
+                          case_data: Dict[str, Any], step_result: Dict[str, Any]):
+        """执行Evaluator agent"""
+        start_time = time.time()
+        
+        try:
+            # 准备评价器需要的数据格式
+            round_data = {
+                "patient_response": step_result.get("patient_response", ""),
+                "doctor_inquiry": step_result.get("doctor_question", ""),
+                "HPI": step_result.get("updated_hpi", ""),
+                "PH": step_result.get("updated_ph", ""),
+                "chief_complaint": step_result.get("updated_chief_complaint", "")
+            }
+            
+            # 调用评价器进行单轮评价
+            input_data = {
+                "patient_case": case_data,
+                "current_round": step_num,
+                "round_data": round_data
+            }
+            
+            result = self.evaluator.evaluate_single_round(
+                patient_case=case_data,
+                round_data=round_data
+            )
+            
+            execution_time = time.time() - start_time
+            
+            output_data = {
+                "clinical_inquiry": {
+                    "score": result.clinical_inquiry.score,
+                    "comment": result.clinical_inquiry.comment
+                },
+                "communication_quality": {
+                    "score": result.communication_quality.score,
+                    "comment": result.communication_quality.comment
+                },
+                "overall_professionalism": {
+                    "score": result.overall_professionalism.score,
+                    "comment": result.overall_professionalism.comment
+                },
+                "summary": result.summary,
+                "key_suggestions": result.key_suggestions
+            }
+            
+            logger.log_agent_execution(step_num, "evaluator", input_data, output_data, execution_time)
+            
+            return result
+            
+        except Exception as e:
+            error_msg = f"Evaluator执行失败: {str(e)}"
+            logger.log_error(step_num, "evaluator_error", error_msg)
+            # 返回默认评价结果
+            from agent_system.evaluetor.response_model import EvaluatorResult, EvaluationDimension
+            
+            default_dimension = EvaluationDimension(score=0.0, comment="评价失败")
+            return EvaluatorResult(
+                clinical_inquiry=default_dimension,
+                diagnostic_reasoning=default_dimension,
+                communication_quality=default_dimension,
+                multi_round_consistency=default_dimension,
+                overall_professionalism=default_dimension,
+                present_illness_similarity=default_dimension,
+                past_history_similarity=default_dimension,
+                chief_complaint_similarity=default_dimension,
+                summary="评价失败",
+                key_suggestions=["系统需要调试"]
+            )

workflow/workflow_logger.py

@@ -10,16 +10,18 @@ class WorkflowLogger:
     负责将每个step的详细信息记录到jsonl格式文件中
     """
     
-    def __init__(self, case_data: Dict[str, Any], log_dir: str = "logs"):
+    def __init__(self, case_data: Dict[str, Any], log_dir: str = "logs", case_index: Optional[int] = None):
         """
         初始化日志记录器
         
         Args:
             case_data: 病例数据
             log_dir: 日志目录，默认为"logs"  
+            case_index: 病例序号，用于文件名标识
         """
         self.case_data = case_data
         self.log_dir = log_dir
+        self.case_index = case_index
         self.log_file_path = self._generate_log_file_path()
         self.step_count = 0
         
@@ -36,15 +38,18 @@ class WorkflowLogger:
         Returns:
             str: 日志文件路径
         """
-        # 生成基于病例内容的唯一标识
-        case_str = json.dumps(self.case_data, ensure_ascii=False, sort_keys=True)
-        case_hash = hashlib.md5(case_str.encode('utf-8')).hexdigest()[:8]
-        
         # 生成时间戳
         timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
         
-        # 构建文件名
+        # 构建文件名，如果有序号则包含序号
+        if self.case_index is not None:
+            filename = f"workflow_{timestamp}_case_{self.case_index:04d}.jsonl"
+        else:
+            # 生成基于病例内容的唯一标识作为后备
+            case_str = json.dumps(self.case_data, ensure_ascii=False, sort_keys=True)
+            case_hash = hashlib.md5(case_str.encode('utf-8')).hexdigest()[:8]
             filename = f"workflow_{timestamp}_{case_hash}.jsonl"
+        
         return os.path.join(self.log_dir, filename)
     
     def _log_workflow_start(self):