feat(agent): 使用用户配置的LLM参数替代硬编码值

重构所有Agent和LLM服务，移除硬编码的temperature和max_tokens参数
添加get_analysis_config函数统一处理分析配置
在LLM测试接口中显示用户保存的配置参数
前端调试面板默认显示LLM测试详细信息

backend/app/api/v1/endpoints/config.py

@@ -299,6 +299,7 @@ class LLMTestResponse(BaseModel):
 @router.post("/test-llm", response_model=LLMTestResponse)
 async def test_llm_connection(
     request: LLMTestRequest,
+    db: AsyncSession = Depends(get_db),
     current_user: User = Depends(deps.get_current_user),
 ) -> Any:
     """测试LLM连接是否正常"""
@@ -309,12 +310,53 @@ async def test_llm_connection(
     import time
 
     start_time = time.time()
+
+    # 获取用户保存的配置
+    result = await db.execute(
+        select(UserConfig).where(UserConfig.user_id == current_user.id)
+    )
+    user_config_record = result.scalar_one_or_none()
+
+    # 解析用户配置
+    saved_llm_config = {}
+    saved_other_config = {}
+    if user_config_record:
+        if user_config_record.llm_config:
+            saved_llm_config = decrypt_config(
+                json.loads(user_config_record.llm_config),
+                SENSITIVE_LLM_FIELDS
+            )
+        if user_config_record.other_config:
+            saved_other_config = decrypt_config(
+                json.loads(user_config_record.other_config),
+                SENSITIVE_OTHER_FIELDS
+            )
+
+    # 从保存的配置中获取参数（用于调试显示）
+    saved_timeout_ms = saved_llm_config.get('llmTimeout', settings.LLM_TIMEOUT * 1000)
+    saved_temperature = saved_llm_config.get('llmTemperature', settings.LLM_TEMPERATURE)
+    saved_max_tokens = saved_llm_config.get('llmMaxTokens', settings.LLM_MAX_TOKENS)
+    saved_concurrency = saved_other_config.get('llmConcurrency', settings.LLM_CONCURRENCY)
+    saved_gap_ms = saved_other_config.get('llmGapMs', settings.LLM_GAP_MS)
+    saved_max_files = saved_other_config.get('maxAnalyzeFiles', settings.MAX_ANALYZE_FILES)
+    saved_output_lang = saved_other_config.get('outputLanguage', settings.OUTPUT_LANGUAGE)
+
     debug_info = {
         "provider": request.provider,
         "model_requested": request.model,
         "base_url_requested": request.baseUrl,
         "api_key_length": len(request.apiKey) if request.apiKey else 0,
         "api_key_prefix": request.apiKey[:8] + "..." if request.apiKey and len(request.apiKey) > 8 else "(empty)",
+        # 用户保存的配置参数
+        "saved_config": {
+            "timeout_ms": saved_timeout_ms,
+            "temperature": saved_temperature,
+            "max_tokens": saved_max_tokens,
+            "concurrency": saved_concurrency,
+            "gap_ms": saved_gap_ms,
+            "max_analyze_files": saved_max_files,
+            "output_language": saved_output_lang,
+        },
     }
 
     try:
@@ -346,11 +388,21 @@ async def test_llm_connection(
         model = request.model or DEFAULT_MODELS.get(provider)
         base_url = request.baseUrl or DEFAULT_BASE_URLS.get(provider, "")
 
+        # 测试时使用用户保存的所有配置参数
+        test_timeout = int(saved_timeout_ms / 1000) if saved_timeout_ms else settings.LLM_TIMEOUT
+        test_temperature = saved_temperature if saved_temperature is not None else settings.LLM_TEMPERATURE
+        test_max_tokens = saved_max_tokens if saved_max_tokens else settings.LLM_MAX_TOKENS
+
         debug_info["model_used"] = model
         debug_info["base_url_used"] = base_url
         debug_info["is_native_adapter"] = provider in NATIVE_ONLY_PROVIDERS
+        debug_info["test_params"] = {
+            "timeout": test_timeout,
+            "temperature": test_temperature,
+            "max_tokens": test_max_tokens,
+        }
 
-        print(f"[LLM Test] 开始测试: provider={provider.value}, model={model}, base_url={base_url}")
+        print(f"[LLM Test] 开始测试: provider={provider.value}, model={model}, base_url={base_url}, temperature={test_temperature}, timeout={test_timeout}s, max_tokens={test_max_tokens}")
 
         # 创建配置
         config = LLMConfig(
@@ -358,8 +410,9 @@ async def test_llm_connection(
             api_key=request.apiKey,
             model=model,
             base_url=request.baseUrl,
-            timeout=30,  # 测试使用较短的超时时间
-            max_tokens=50,  # 测试使用较少的token
+            timeout=test_timeout,
+            temperature=test_temperature,
+            max_tokens=test_max_tokens,
         )
 
         # 直接创建新的适配器实例（不使用缓存），确保使用最新的配置
@@ -375,13 +428,14 @@ async def test_llm_connection(
             adapter = LiteLLMAdapter(config)
             debug_info["adapter_type"] = "LiteLLMAdapter"
             # 获取 LiteLLM 实际使用的模型名
-            debug_info["litellm_model"] = getattr(adapter, '_get_model_name', lambda: model)() if hasattr(adapter, '_get_model_name') else model
+            debug_info["litellm_model"] = getattr(adapter, '_get_litellm_model', lambda: model)() if hasattr(adapter, '_get_litellm_model') else model
 
         test_request = LLMRequest(
             messages=[
                 LLMMessage(role="user", content="Say 'Hello' in one word.")
             ],
-            max_tokens=50,
+            temperature=test_temperature,
+            max_tokens=test_max_tokens,
         )
 
         print(f"[LLM Test] 发送测试请求...")

backend/app/api/v1/endpoints/scan.py

@@ -20,7 +20,7 @@ from app.models.project import Project
 from app.models.analysis import InstantAnalysis
 from app.models.user_config import UserConfig
 from app.services.llm.service import LLMService
-from app.services.scanner import task_control, is_text_file, should_exclude, get_language_from_path
+from app.services.scanner import task_control, is_text_file, should_exclude, get_language_from_path, get_analysis_config
 from app.services.zip_storage import load_project_zip, save_project_zip, has_project_zip
 from app.core.config import settings
 
@@ -93,6 +93,11 @@ async def process_zip_task(task_id: str, file_path: str, db_session_factory, use
                         except:
                             pass
 
+            # 获取分析配置（优先使用用户配置）
+            analysis_config = get_analysis_config(user_config)
+            max_analyze_files = analysis_config['max_analyze_files']
+            llm_gap_ms = analysis_config['llm_gap_ms']
+
             # 限制文件数量
             # 如果指定了特定文件，则只分析这些文件
             target_files = scan_config.get('file_paths', [])
@@ -101,13 +106,13 @@ async def process_zip_task(task_id: str, file_path: str, db_session_factory, use
                 normalized_targets = {normalize_path(p) for p in target_files}
                 print(f"🎯 ZIP任务: 指定分析 {len(normalized_targets)} 个文件")
                 files_to_scan = [f for f in files_to_scan if f['path'] in normalized_targets]
-            elif settings.MAX_ANALYZE_FILES > 0:
-                files_to_scan = files_to_scan[:settings.MAX_ANALYZE_FILES]
-            
+            elif max_analyze_files > 0:
+                files_to_scan = files_to_scan[:max_analyze_files]
+
             task.total_files = len(files_to_scan)
             await db.commit()
 
-            print(f"📊 ZIP任务 {task_id}: 找到 {len(files_to_scan)} 个文件")
+            print(f"📊 ZIP任务 {task_id}: 找到 {len(files_to_scan)} 个文件 (最大文件数: {max_analyze_files}, 请求间隔: {llm_gap_ms}ms)")
 
             total_issues = 0
             total_lines = 0
@@ -178,12 +183,12 @@ async def process_zip_task(task_id: str, file_path: str, db_session_factory, use
                     print(f"📈 ZIP任务 {task_id}: 进度 {scanned_files}/{len(files_to_scan)}")
                     
                     # 请求间隔
-                    await asyncio.sleep(settings.LLM_GAP_MS / 1000)
-                    
+                    await asyncio.sleep(llm_gap_ms / 1000)
+
                 except Exception as file_error:
                     failed_files += 1
                     print(f"❌ ZIP任务分析文件失败 ({file_info['path']}): {file_error}")
-                    await asyncio.sleep(settings.LLM_GAP_MS / 1000)
+                    await asyncio.sleep(llm_gap_ms / 1000)
 
             # 完成任务
             avg_quality_score = sum(quality_scores) / len(quality_scores) if quality_scores else 100.0

backend/app/services/agent/agents/analysis.py

@@ -452,12 +452,11 @@ class AnalysisAgent(BaseAgent):
                     break
                 
                 # 调用 LLM 进行思考和决策（流式输出）
-                # 🔥 增加 max_tokens 到 4096，避免长输出被截断
+                # 🔥 使用用户配置的 temperature 和 max_tokens
                 try:
                     llm_output, tokens_this_round = await self.stream_llm_call(
                         self._conversation_history,
-                        temperature=0.1,
-                        max_tokens=8192,
+                        # 🔥 不传递 temperature 和 max_tokens，使用用户配置
                     )
                 except asyncio.CancelledError:
                     logger.info(f"[{self.name}] LLM call cancelled")
@@ -653,8 +652,7 @@ Final Answer:""",
                 try:
                     summary_output, _ = await self.stream_llm_call(
                         self._conversation_history,
-                        temperature=0.1,
-                        max_tokens=4096,
+                        # 🔥 不传递 temperature 和 max_tokens，使用用户配置
                     )
                     
                     if summary_output and summary_output.strip():

backend/app/services/agent/agents/base.py

@@ -838,25 +838,24 @@ class BaseAgent(ABC):
         Args:
             messages: 消息列表
             tools: 可用工具描述
-            
+
         Returns:
             LLM 响应
         """
         self._iteration += 1
-        
+
         try:
+            # 🔥 不传递 temperature 和 max_tokens，让 LLMService 使用用户配置
             response = await self.llm_service.chat_completion(
                 messages=messages,
-                temperature=self.config.temperature,
-                max_tokens=self.config.max_tokens,
                 tools=tools,
             )
-            
+
             if response.get("usage"):
                 self._total_tokens += response["usage"].get("total_tokens", 0)
-            
+
             return response
-            
+
         except Exception as e:
             logger.error(f"LLM call failed: {e}")
             raise
@@ -925,46 +924,46 @@ class BaseAgent(ABC):
         return messages
     
     # ============ 统一的流式 LLM 调用 ============
-    
+
     async def stream_llm_call(
         self,
         messages: List[Dict[str, str]],
-        temperature: float = 0.1,
-        max_tokens: int = 2048,
+        temperature: Optional[float] = None,
+        max_tokens: Optional[int] = None,
         auto_compress: bool = True,
     ) -> Tuple[str, int]:
         """
         统一的流式 LLM 调用方法
-        
+
         所有 Agent 共用此方法，避免重复代码
-        
+
         Args:
             messages: 消息列表
-            temperature: 温度
-            max_tokens: 最大 token 数
+            temperature: 温度（None 时使用用户配置）
+            max_tokens: 最大 token 数（None 时使用用户配置）
             auto_compress: 是否自动压缩过长的消息历史
-            
+
         Returns:
             (完整响应内容, token数量)
         """
         # 🔥 自动压缩过长的消息历史
         if auto_compress:
             messages = self.compress_messages_if_needed(messages)
-        
+
         accumulated = ""
         total_tokens = 0
-        
+
         # 🔥 在开始 LLM 调用前检查取消
         if self.is_cancelled:
             logger.info(f"[{self.name}] Cancelled before LLM call")
             return "", 0
-        
+
         logger.info(f"[{self.name}] 🚀 Starting stream_llm_call, emitting thinking_start...")
         await self.emit_thinking_start()
         logger.info(f"[{self.name}] ✅ thinking_start emitted, starting LLM stream...")
-        
+
         try:
-            # 获取流式迭代器
+            # 获取流式迭代器（传入 None 时使用用户配置）
             stream = self.llm_service.chat_completion_stream(
                 messages=messages,
                 temperature=temperature,

backend/app/services/agent/agents/orchestrator.py

@@ -241,8 +241,7 @@ class OrchestratorAgent(BaseAgent):
                 try:
                     llm_output, tokens_this_round = await self.stream_llm_call(
                         self._conversation_history,
-                        temperature=0.1,
-                        max_tokens=8192,  # 🔥 增加到 8192，避免截断
+                        # 🔥 不传递 temperature 和 max_tokens，使用用户配置
                     )
                 except asyncio.CancelledError:
                     logger.info(f"[{self.name}] LLM call cancelled")

backend/app/services/agent/agents/recon.py

@@ -358,8 +358,7 @@ class ReconAgent(BaseAgent):
                 try:
                     llm_output, tokens_this_round = await self.stream_llm_call(
                         self._conversation_history,
-                        temperature=0.1,
-                        max_tokens=8192,  # 🔥 增加到 8192，避免截断
+                        # 🔥 不传递 temperature 和 max_tokens，使用用户配置
                     )
                 except asyncio.CancelledError:
                     logger.info(f"[{self.name}] LLM call cancelled")
@@ -525,8 +524,7 @@ Final Answer:""",
                 try:
                     summary_output, _ = await self.stream_llm_call(
                         self._conversation_history,
-                        temperature=0.1,
-                        max_tokens=2048,
+                        # 🔥 不传递 temperature 和 max_tokens，使用用户配置
                     )
                     
                     if summary_output and summary_output.strip():

backend/app/services/agent/agents/verification.py

@@ -587,8 +587,7 @@ class VerificationAgent(BaseAgent):
                 try:
                     llm_output, tokens_this_round = await self.stream_llm_call(
                         self._conversation_history,
-                        temperature=0.1,
-                        max_tokens=8192,  # 🔥 增加到 8192，避免截断
+                        # 🔥 不传递 temperature 和 max_tokens，使用用户配置
                     )
                 except asyncio.CancelledError:
                     logger.info(f"[{self.name}] LLM call cancelled")

backend/app/services/agent/graph/audit_graph.py

@@ -125,8 +125,7 @@ class LLMRouter:
                     {"role": "system", "content": "你是安全审计流程的决策者，负责决定下一步行动。"},
                     {"role": "user", "content": prompt},
                 ],
-                temperature=0.1,
-                max_tokens=200,
+                # 🔥 不传递 temperature 和 max_tokens，使用用户配置
             )
             
             content = response.get("content", "")
@@ -180,8 +179,7 @@ class LLMRouter:
                     {"role": "system", "content": "你是安全审计流程的决策者，负责决定下一步行动。"},
                     {"role": "user", "content": prompt},
                 ],
-                temperature=0.1,
-                max_tokens=200,
+                # 🔥 不传递 temperature 和 max_tokens，使用用户配置
             )
             
             content = response.get("content", "")
@@ -227,8 +225,7 @@ class LLMRouter:
                     {"role": "system", "content": "你是安全审计流程的决策者，负责决定下一步行动。"},
                     {"role": "user", "content": prompt},
                 ],
-                temperature=0.1,
-                max_tokens=200,
+                # 🔥 不传递 temperature 和 max_tokens，使用用户配置
             )
             
             content = response.get("content", "")

backend/app/services/llm/service.py

@@ -359,12 +359,14 @@ Please analyze the following code:
         try:
             adapter = LLMFactory.create_adapter(self.config)
             
+            # 使用用户配置的 temperature（如果未设置则使用 config 中的默认值）
             request = LLMRequest(
                 messages=[
                     LLMMessage(role="system", content=system_prompt),
                     LLMMessage(role="user", content=user_prompt)
                 ],
-                temperature=0.1,
+                temperature=self.config.temperature,
+                max_tokens=self.config.max_tokens,
             )
             
             response = await adapter.complete(request)
@@ -401,39 +403,97 @@ Please analyze the following code:
             logger.error(f"Provider: {self.config.provider.value}, Model: {self.config.model}")
             # 重新抛出异常，让调用者处理
             raise
-    
-    async def chat_completion_raw(
+
+    async def chat_completion(
         self,
         messages: List[Dict[str, str]],
-        temperature: float = 0.1,
-        max_tokens: int = 4096,
+        temperature: Optional[float] = None,
+        max_tokens: Optional[int] = None,
+        tools: Optional[List[Dict[str, Any]]] = None,
     ) -> Dict[str, Any]:
         """
-        🔥 Agent 使用的原始聊天完成接口（兼容旧接口）
-        
+        🔥 Agent 使用的聊天完成接口（支持工具调用）
+
         Args:
             messages: 消息列表，格式为 [{"role": "user", "content": "..."}]
-            temperature: 温度参数
-            max_tokens: 最大token数
-            
+            temperature: 温度参数（None 时使用用户配置）
+            max_tokens: 最大token数（None 时使用用户配置）
+            tools: 工具描述列表（可选）
+
         Returns:
-            包含 content 和 usage 的字典
+            包含 content、usage 和 tool_calls 的字典
         """
+        # 使用用户配置作为默认值
+        actual_temperature = temperature if temperature is not None else self.config.temperature
+        actual_max_tokens = max_tokens if max_tokens is not None else self.config.max_tokens
+
         # 转换消息格式
         llm_messages = [
             LLMMessage(role=msg["role"], content=msg["content"])
             for msg in messages
         ]
-        
+
         request = LLMRequest(
             messages=llm_messages,
-            temperature=temperature,
-            max_tokens=max_tokens,
+            temperature=actual_temperature,
+            max_tokens=actual_max_tokens,
+            tools=tools,
         )
-        
+
         adapter = LLMFactory.create_adapter(self.config)
         response = await adapter.complete(request)
-        
+
+        result = {
+            "content": response.content,
+            "usage": {
+                "prompt_tokens": response.usage.prompt_tokens if response.usage else 0,
+                "completion_tokens": response.usage.completion_tokens if response.usage else 0,
+                "total_tokens": response.usage.total_tokens if response.usage else 0,
+            },
+        }
+
+        # 添加工具调用信息
+        if response.tool_calls:
+            result["tool_calls"] = response.tool_calls
+
+        return result
+
+    async def chat_completion_raw(
+        self,
+        messages: List[Dict[str, str]],
+        temperature: Optional[float] = None,
+        max_tokens: Optional[int] = None,
+    ) -> Dict[str, Any]:
+        """
+        🔥 Agent 使用的原始聊天完成接口（兼容旧接口）
+
+        Args:
+            messages: 消息列表，格式为 [{"role": "user", "content": "..."}]
+            temperature: 温度参数（None 时使用用户配置）
+            max_tokens: 最大token数（None 时使用用户配置）
+
+        Returns:
+            包含 content 和 usage 的字典
+        """
+        # 使用用户配置作为默认值
+        actual_temperature = temperature if temperature is not None else self.config.temperature
+        actual_max_tokens = max_tokens if max_tokens is not None else self.config.max_tokens
+
+        # 转换消息格式
+        llm_messages = [
+            LLMMessage(role=msg["role"], content=msg["content"])
+            for msg in messages
+        ]
+
+        request = LLMRequest(
+            messages=llm_messages,
+            temperature=actual_temperature,
+            max_tokens=actual_max_tokens,
+        )
+
+        adapter = LLMFactory.create_adapter(self.config)
+        response = await adapter.complete(request)
+
         return {
             "content": response.content,
             "usage": {
@@ -446,29 +506,33 @@ Please analyze the following code:
     async def chat_completion_stream(
         self,
         messages: List[Dict[str, str]],
-        temperature: float = 0.1,
-        max_tokens: int = 4096,
+        temperature: Optional[float] = None,
+        max_tokens: Optional[int] = None,
     ):
         """
         流式聊天完成接口，逐 token 返回
-        
+
         Args:
             messages: 消息列表
-            temperature: 温度参数
-            max_tokens: 最大token数
-            
+            temperature: 温度参数（None 时使用用户配置）
+            max_tokens: 最大token数（None 时使用用户配置）
+
         Yields:
             dict: {"type": "token", "content": str} 或 {"type": "done", ...}
         """
+        # 使用用户配置作为默认值
+        actual_temperature = temperature if temperature is not None else self.config.temperature
+        actual_max_tokens = max_tokens if max_tokens is not None else self.config.max_tokens
+
         llm_messages = [
             LLMMessage(role=msg["role"], content=msg["content"])
             for msg in messages
         ]
-        
+
         request = LLMRequest(
             messages=llm_messages,
-            temperature=temperature,
-            max_tokens=max_tokens,
+            temperature=actual_temperature,
+            max_tokens=actual_max_tokens,
         )
         
         if self.config.provider in NATIVE_ONLY_PROVIDERS:
@@ -869,15 +933,17 @@ Please analyze the following code:
         
         try:
             adapter = LLMFactory.create_adapter(self.config)
-            
+
+            # 使用用户配置的 temperature 和 max_tokens
             request = LLMRequest(
                 messages=[
                     LLMMessage(role="system", content=full_system_prompt),
                     LLMMessage(role="user", content=user_prompt)
                 ],
-                temperature=0.1,
+                temperature=self.config.temperature,
+                max_tokens=self.config.max_tokens,
             )
-            
+
             response = await adapter.complete(request)
             content = response.content
             

backend/app/services/scanner.py

@@ -15,6 +15,25 @@ from app.services.llm.service import LLMService
 from app.core.config import settings
 
 
+def get_analysis_config(user_config: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:
+    """
+    获取分析配置参数（优先使用用户配置，然后使用系统配置）
+
+    Returns:
+        包含以下字段的字典:
+        - max_analyze_files: 最大分析文件数
+        - llm_concurrency: LLM 并发数
+        - llm_gap_ms: LLM 请求间隔（毫秒）
+    """
+    other_config = (user_config or {}).get('otherConfig', {})
+
+    return {
+        'max_analyze_files': other_config.get('maxAnalyzeFiles') or settings.MAX_ANALYZE_FILES,
+        'llm_concurrency': other_config.get('llmConcurrency') or settings.LLM_CONCURRENCY,
+        'llm_gap_ms': other_config.get('llmGapMs') or settings.LLM_GAP_MS,
+    }
+
+
 # 支持的文本文件扩展名
 TEXT_EXTENSIONS = [
     ".js", ".ts", ".tsx", ".jsx", ".py", ".java", ".go", ".rs", 
@@ -344,19 +363,24 @@ async def scan_repo_task(task_id: str, db_session_factory, user_config: dict = N
 
             print(f"✅ 成功获取分支 {actual_branch} 的文件列表")
 
+            # 获取分析配置（优先使用用户配置）
+            analysis_config = get_analysis_config(user_config)
+            max_analyze_files = analysis_config['max_analyze_files']
+            llm_gap_ms = analysis_config['llm_gap_ms']
+
             # 限制文件数量
             # 如果指定了特定文件，则只分析这些文件
             target_files = (user_config or {}).get('scan_config', {}).get('file_paths', [])
             if target_files:
                 print(f"🎯 指定分析 {len(target_files)} 个文件")
                 files = [f for f in files if f['path'] in target_files]
-            elif settings.MAX_ANALYZE_FILES > 0:
-                files = files[:settings.MAX_ANALYZE_FILES]
-            
+            elif max_analyze_files > 0:
+                files = files[:max_analyze_files]
+
             task.total_files = len(files)
             await db.commit()
 
-            print(f"📊 获取到 {len(files)} 个文件，开始分析")
+            print(f"📊 获取到 {len(files)} 个文件，开始分析 (最大文件数: {max_analyze_files}, 请求间隔: {llm_gap_ms}ms)")
 
             # 4. 分析文件
             total_issues = 0
@@ -484,7 +508,7 @@ async def scan_repo_task(task_id: str, db_session_factory, user_config: dict = N
                     print(f"📈 任务 {task_id}: 进度 {scanned_files}/{len(files)} ({int(scanned_files/len(files)*100)}%)")
                     
                     # 请求间隔
-                    await asyncio.sleep(settings.LLM_GAP_MS / 1000)
+                    await asyncio.sleep(llm_gap_ms / 1000)
                     
                 except Exception as file_error:
                     failed_files += 1
@@ -494,7 +518,7 @@ async def scan_repo_task(task_id: str, db_session_factory, user_config: dict = N
                     print(f"❌ 分析文件失败 ({file_info['path']}): {file_error}")
                     print(f"   错误类型: {type(file_error).__name__}")
                     print(f"   详细信息: {traceback.format_exc()}")
-                    await asyncio.sleep(settings.LLM_GAP_MS / 1000)
+                    await asyncio.sleep(llm_gap_ms / 1000)
 
             # 5. 完成任务
             avg_quality_score = sum(quality_scores) / len(quality_scores) if quality_scores else 100.0

frontend/src/components/system/SystemConfig.tsx

@@ -52,7 +52,7 @@ export function SystemConfig() {
   const [hasChanges, setHasChanges] = useState(false);
   const [testingLLM, setTestingLLM] = useState(false);
   const [llmTestResult, setLlmTestResult] = useState<{ success: boolean; message: string; debug?: Record<string, unknown> } | null>(null);
-  const [showDebugInfo, setShowDebugInfo] = useState(false);
+  const [showDebugInfo, setShowDebugInfo] = useState(true);
 
   useEffect(() => { loadConfig(); }, []);
 
@@ -396,15 +396,44 @@ export function SystemConfig() {
                 {showDebugInfo && llmTestResult.debug && (
                   <div className="mt-3 pt-3 border-t border-border/50">
                     <div className="text-xs font-mono space-y-1 text-muted-foreground">
-                      <div className="font-bold text-foreground mb-2">调试信息:</div>
+                      <div className="font-bold text-foreground mb-2">连接信息:</div>
                       <div>Provider: <span className="text-foreground">{String(llmTestResult.debug.provider)}</span></div>
                       <div>Model: <span className="text-foreground">{String(llmTestResult.debug.model_used || llmTestResult.debug.model_requested || 'N/A')}</span></div>
                       <div>Base URL: <span className="text-foreground">{String(llmTestResult.debug.base_url_used || llmTestResult.debug.base_url_requested || '(default)')}</span></div>
                       <div>Adapter: <span className="text-foreground">{String(llmTestResult.debug.adapter_type || 'N/A')}</span></div>
                       <div>API Key: <span className="text-foreground">{String(llmTestResult.debug.api_key_prefix)} (长度: {String(llmTestResult.debug.api_key_length)})</span></div>
                       <div>耗时: <span className="text-foreground">{String(llmTestResult.debug.elapsed_time_ms || 'N/A')} ms</span></div>
+
+                      {/* 用户保存的配置参数 */}
+                      {llmTestResult.debug.saved_config && (
+                        <div className="mt-3 pt-2 border-t border-border/30">
+                          <div className="font-bold text-cyan-400 mb-2">已保存的配置参数:</div>
+                          <div className="grid grid-cols-2 gap-x-4 gap-y-1">
+                            <div>温度: <span className="text-foreground">{String((llmTestResult.debug.saved_config as Record<string, unknown>).temperature ?? 'N/A')}</span></div>
+                            <div>最大Tokens: <span className="text-foreground">{String((llmTestResult.debug.saved_config as Record<string, unknown>).max_tokens ?? 'N/A')}</span></div>
+                            <div>超时: <span className="text-foreground">{String((llmTestResult.debug.saved_config as Record<string, unknown>).timeout_ms ?? 'N/A')} ms</span></div>
+                            <div>请求间隔: <span className="text-foreground">{String((llmTestResult.debug.saved_config as Record<string, unknown>).gap_ms ?? 'N/A')} ms</span></div>
+                            <div>并发数: <span className="text-foreground">{String((llmTestResult.debug.saved_config as Record<string, unknown>).concurrency ?? 'N/A')}</span></div>
+                            <div>最大文件数: <span className="text-foreground">{String((llmTestResult.debug.saved_config as Record<string, unknown>).max_analyze_files ?? 'N/A')}</span></div>
+                            <div>输出语言: <span className="text-foreground">{String((llmTestResult.debug.saved_config as Record<string, unknown>).output_language ?? 'N/A')}</span></div>
+                          </div>
+                        </div>
+                      )}
+
+                      {/* 测试时实际使用的参数 */}
+                      {llmTestResult.debug.test_params && (
+                        <div className="mt-2 pt-2 border-t border-border/30">
+                          <div className="font-bold text-emerald-400 mb-2">测试时使用的参数:</div>
+                          <div className="grid grid-cols-3 gap-x-4">
+                            <div>温度: <span className="text-foreground">{String((llmTestResult.debug.test_params as Record<string, unknown>).temperature ?? 'N/A')}</span></div>
+                            <div>超时: <span className="text-foreground">{String((llmTestResult.debug.test_params as Record<string, unknown>).timeout ?? 'N/A')}s</span></div>
+                            <div>MaxTokens: <span className="text-foreground">{String((llmTestResult.debug.test_params as Record<string, unknown>).max_tokens ?? 'N/A')}</span></div>
+                          </div>
+                        </div>
+                      )}
+
                       {llmTestResult.debug.error_category && (
-                        <div>错误类型: <span className="text-rose-400">{String(llmTestResult.debug.error_category)}</span></div>
+                        <div className="mt-2">错误类型: <span className="text-rose-400">{String(llmTestResult.debug.error_category)}</span></div>
                       )}
                       {llmTestResult.debug.error_type && (
                         <div>异常类型: <span className="text-rose-400">{String(llmTestResult.debug.error_type)}</span></div>