update to deepagents 0.5.2

enable_thinking 开启时，thinking 中间件将 AIMessage 追加到 messages 末尾，
导致不支持 assistant prefill 的模型返回 400 错误。
修复方式：在 AIMessage 后追加多语言 HumanMessage，确保消息以 user 结尾。

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

agent/custom_filesystem_middleware.py

@@ -4,7 +4,9 @@
 """
 
 from pathlib import Path
-from typing import Annotated, Literal, cast
+from typing import Annotated, cast
+import mimetypes
+import warnings
 
 from langchain.tools import ToolRuntime
 from langchain_core.messages import ToolMessage
@@ -15,19 +17,24 @@ from deepagents.backends import StateBackend
 from deepagents.backends.composite import CompositeBackend
 from deepagents.backends.protocol import (
     BACKEND_TYPES,
+    ReadResult,
 )
+from deepagents.backends.utils import _get_file_type, validate_path
+from langchain_core.messages.content import ContentBlock
 from deepagents.middleware.filesystem import (
     DEFAULT_READ_OFFSET,
     DEFAULT_READ_LIMIT,
-    IMAGE_EXTENSIONS,
-    IMAGE_MEDIA_TYPES,
     FilesystemMiddleware,
     FilesystemState,
+    READ_FILE_TOOL_DESCRIPTION,
+    READ_FILE_TRUNCATION_MSG,
+    NUM_CHARS_PER_TOKEN,
+    ReadFileSchema,
+    check_empty_content,
+    format_content_with_line_numbers,
 )
 
 from langgraph.types import Command
-import base64
-from langchain_core.messages.content import create_image_block
 
 
 # SKILL.md 文件的行数限制（设置为较大的值以完整读取）
@@ -44,10 +51,62 @@ class CustomFilesystemMiddleware(FilesystemMiddleware):
     @override
     def _create_read_file_tool(self) -> BaseTool:
         """创建自定义的 read_file 工具，支持 SKILL.md 完整读取。"""
-        # 从父类获取工具描述
+        tool_description = self._custom_tool_descriptions.get("read_file") or READ_FILE_TOOL_DESCRIPTION
-        tool_description = self._custom_tool_descriptions.get("read_file") or self._get_read_file_description()
         token_limit = self._tool_token_limit_before_evict
 
+        def _truncate(content: str, file_path: str, limit: int) -> str:
+            lines = content.splitlines(keepends=True)
+            if len(lines) > limit:
+                lines = lines[:limit]
+                content = "".join(lines)
+
+            if token_limit and len(content) >= NUM_CHARS_PER_TOKEN * token_limit:
+                truncation_msg = READ_FILE_TRUNCATION_MSG.format(file_path=file_path)
+                max_content_length = NUM_CHARS_PER_TOKEN * token_limit - len(truncation_msg)
+                content = content[:max_content_length] + truncation_msg
+
+            return content
+
+        def _handle_read_result(
+            read_result: ReadResult | str,
+            validated_path: str,
+            tool_call_id: str | None,
+            offset: int,
+            limit: int,
+        ) -> ToolMessage | str:
+            if isinstance(read_result, str):
+                warnings.warn(
+                    "Returning a plain `str` from `backend.read()` is deprecated. ",
+                    DeprecationWarning,
+                    stacklevel=2,
+                )
+                return _truncate(read_result, validated_path, limit)
+
+            if read_result.error:
+                return f"Error: {read_result.error}"
+
+            if read_result.file_data is None:
+                return f"Error: no data returned for '{validated_path}'"
+
+            file_type = _get_file_type(validated_path)
+            content = read_result.file_data["content"]
+
+            if file_type != "text":
+                mime_type = mimetypes.guess_type("file" + Path(validated_path).suffix)[0] or "application/octet-stream"
+                return ToolMessage(
+                    content_blocks=cast("list[ContentBlock]", [{"type": file_type, "base64": content, "mime_type": mime_type}]),
+                    name="read_file",
+                    tool_call_id=tool_call_id,
+                    additional_kwargs={"read_file_path": validated_path, "read_file_media_type": mime_type},
+                )
+
+            empty_msg = check_empty_content(content)
+            if empty_msg:
+                return empty_msg
+
+            content = format_content_with_line_numbers(content, start_line=offset + 1)
+            return _truncate(content, validated_path, limit)
+
         def sync_read_file(
             file_path: Annotated[str, "Absolute path to the file to read. Must be absolute, not relative."],
             runtime: ToolRuntime[None, FilesystemState],
@@ -55,57 +114,18 @@ class CustomFilesystemMiddleware(FilesystemMiddleware):
             limit: Annotated[int, "Maximum number of lines to read. Use for pagination of large files."] = DEFAULT_READ_LIMIT,
         ) -> ToolMessage | str:
             """Synchronous wrapper for read_file tool with SKILL.md special handling."""
-            from deepagents.backends.utils import validate_path
-            from deepagents.middleware.filesystem import READ_FILE_TRUNCATION_MSG, NUM_CHARS_PER_TOKEN
-
             resolved_backend = self._get_backend(runtime)
             try:
                 validated_path = validate_path(file_path)
             except ValueError as e:
                 return f"Error: {e}"
 
-            ext = Path(validated_path).suffix.lower()
-            # 处理图片文件
-            if ext in IMAGE_EXTENSIONS:
-                responses = resolved_backend.download_files([validated_path])
-                if responses and responses[0].content is not None:
-                    media_type = IMAGE_MEDIA_TYPES.get(ext, "image/png")
-                    image_b64 = base64.standard_b64encode(responses[0].content).decode("utf-8")
-                    return ToolMessage(
-                        content_blocks=[create_image_block(base64=image_b64, mime_type=media_type)],
-                        name="read_file",
-                        tool_call_id=runtime.tool_call_id,
-                        additional_kwargs={
-                            "read_file_path": validated_path,
-                            "read_file_media_type": media_type,
-                        },
-                    )
-                if responses and responses[0].error:
-                    return f"Error reading image: {responses[0].error}"
-                return "Error reading image: unknown error"
-
             # 如果是 SKILL.md 文件，使用大限制读取完整内容
             if validated_path.endswith("SKILL.md") or validated_path.endswith("/SKILL.md"):
-                actual_limit = SKILL_MD_READ_LIMIT
+                limit = SKILL_MD_READ_LIMIT
-            else:
-                actual_limit = limit
 
-            result = resolved_backend.read(validated_path, offset=offset, limit=actual_limit)
+            read_result = resolved_backend.read(validated_path, offset=offset, limit=limit)
-
+            return _handle_read_result(read_result, validated_path, runtime.tool_call_id, offset, limit)
-            lines = result.splitlines(keepends=True)
-            if len(lines) > actual_limit:
-                lines = lines[:actual_limit]
-                result = "".join(lines)
-
-            # Check if result exceeds token threshold and truncate if necessary
-            if token_limit and len(result) >= NUM_CHARS_PER_TOKEN * token_limit:
-                # Calculate truncation message length to ensure final result stays under threshold
-                truncation_msg = READ_FILE_TRUNCATION_MSG.format(file_path=validated_path)
-                max_content_length = NUM_CHARS_PER_TOKEN * token_limit - len(truncation_msg)
-                result = result[:max_content_length]
-                result += truncation_msg
-
-            return result
 
         async def async_read_file(
             file_path: Annotated[str, "Absolute path to the file to read. Must be absolute, not relative."],
@@ -114,66 +134,28 @@ class CustomFilesystemMiddleware(FilesystemMiddleware):
             limit: Annotated[int, "Maximum number of lines to read. Use for pagination of large files."] = DEFAULT_READ_LIMIT,
         ) -> ToolMessage | str:
             """Asynchronous wrapper for read_file tool with SKILL.md special handling."""
-            from deepagents.backends.utils import validate_path
-            from deepagents.middleware.filesystem import READ_FILE_TRUNCATION_MSG, NUM_CHARS_PER_TOKEN
-
             resolved_backend = self._get_backend(runtime)
             try:
                 validated_path = validate_path(file_path)
             except ValueError as e:
                 return f"Error: {e}"
 
-            ext = Path(validated_path).suffix.lower()
-            # 处理图片文件
-            if ext in IMAGE_EXTENSIONS:
-                responses = await resolved_backend.adownload_files([validated_path])
-                if responses and responses[0].content is not None:
-                    media_type = IMAGE_MEDIA_TYPES.get(ext, "image/png")
-                    image_b64 = base64.standard_b64encode(responses[0].content).decode("utf-8")
-                    return ToolMessage(
-                        content_blocks=[create_image_block(base64=image_b64, mime_type=media_type)],
-                        name="read_file",
-                        tool_call_id=runtime.tool_call_id,
-                        additional_kwargs={
-                            "read_file_path": validated_path,
-                            "read_file_media_type": media_type,
-                        },
-                    )
-                if responses and responses[0].error:
-                    return f"Error reading image: {responses[0].error}"
-                return "Error reading image: unknown error"
-
             # 如果是 SKILL.md 文件，使用大限制读取完整内容
             if validated_path.endswith("SKILL.md") or validated_path.endswith("/SKILL.md"):
-                actual_limit = SKILL_MD_READ_LIMIT
+                limit = SKILL_MD_READ_LIMIT
-            else:
-                actual_limit = limit
 
-            result = await resolved_backend.aread(validated_path, offset=offset, limit=actual_limit)
+            read_result = await resolved_backend.aread(validated_path, offset=offset, limit=limit)
-
+            return _handle_read_result(read_result, validated_path, runtime.tool_call_id, offset, limit)
-            lines = result.splitlines(keepends=True)
-            if len(lines) > actual_limit:
-                lines = lines[:actual_limit]
-                result = "".join(lines)
-
-            # Check if result exceeds token threshold and truncate if necessary
-            if token_limit and len(result) >= NUM_CHARS_PER_TOKEN * token_limit:
-                # Calculate truncation message length to ensure final result stays under threshold
-                truncation_msg = READ_FILE_TRUNCATION_MSG.format(file_path=validated_path)
-                max_content_length = NUM_CHARS_PER_TOKEN * token_limit - len(truncation_msg)
-                result = result[:max_content_length]
-                result += truncation_msg
-
-            return result
 
         return StructuredTool.from_function(
             name="read_file",
             description=tool_description,
             func=sync_read_file,
             coroutine=async_read_file,
+            infer_schema=False,
+            args_schema=ReadFileSchema,
         )
 
     def _get_read_file_description(self) -> str:
         """获取 read_file 工具的描述，添加 SKILL.md 完整读取的说明。"""
-        from deepagents.middleware.filesystem import READ_FILE_TOOL_DESCRIPTION
         return READ_FILE_TOOL_DESCRIPTION

agent/deep_assistant.py

@@ -407,19 +407,22 @@ def create_custom_cli_agent(
             # Use LocalShellBackend for filesystem + shell execution
             backend = LocalShellBackend(
                 root_dir=workspace_root,
+                virtual_mode=True,
                 inherit_env=True,
                 env=final_shell_env,
             )
         else:
             # No shell access - use plain FilesystemBackend
-            backend = FilesystemBackend(root_dir=workspace_root)
+            backend = FilesystemBackend(root_dir=workspace_root, virtual_mode=True)
 
         # Set up composite backend with routing (参考新版本实现)
         large_results_backend = FilesystemBackend(
             root_dir=tempfile.mkdtemp(prefix="deepagents_large_results_"),
+            virtual_mode=True,
         )
         conversation_history_backend = FilesystemBackend(
             root_dir=tempfile.mkdtemp(prefix="deepagents_conversation_history_"),
+            virtual_mode=True,
         )
         composite_backend = CompositeBackend(
             default=backend,
@@ -435,7 +438,7 @@ def create_custom_cli_agent(
 
             agent_middleware.append(
                 CustomSkillsMiddleware(
-                    backend=FilesystemBackend(root_dir=workspace_root),
+                    backend=FilesystemBackend(root_dir=workspace_root, virtual_mode=True),
                     sources=skills_sources,
                 )
             )

agent/guideline_middleware.py

@@ -1,4 +1,3 @@
-from ast import Str
 from langchain.agents.middleware import AgentState, AgentMiddleware, ModelRequest, ModelResponse
 from langchain_core.messages import convert_to_openai_messages
 from agent.prompt_loader import load_guideline_prompt
@@ -6,7 +5,7 @@ from utils.fastapi_utils import (extract_block_from_system_prompt, format_messag
 from langchain.chat_models import BaseChatModel
 from langgraph.runtime import Runtime
 
-from langchain_core.messages import SystemMessage
+from langchain_core.messages import SystemMessage, HumanMessage
 from typing import Any, Callable
 from langchain_core.callbacks import BaseCallbackHandler
 from langchain_core.outputs import LLMResult
@@ -124,8 +123,9 @@ Action: Provide concise, friendly, and personified natural responses.
         response.additional_kwargs["message_tag"] = "THINK"
         response.content = f"<think>{response.content}</think>"
 
-        # 将响应添加到原始消息列表
+        # 将响应添加到原始消息列表，并追加 HumanMessage 确保消息以 user 结尾
-        state['messages'] = state['messages'] + [response]
+        # 某些模型不支持 assistant message prefill，要求最后一条消息必须是 user
+        state['messages'] = state['messages'] + [response, HumanMessage(content=self._get_follow_up_prompt())]
         return state
 
     async def abefore_agent(self, state: AgentState, runtime: Runtime) -> dict[str, Any] | None:
@@ -148,10 +148,23 @@ Action: Provide concise, friendly, and personified natural responses.
         response.additional_kwargs["message_tag"] = "THINK"
         response.content = f"<think>{response.content}</think>"
 
-        # 将响应添加到原始消息列表
+        # 将响应添加到原始消息列表，并追加 HumanMessage 确保消息以 user 结尾
-        state['messages'] = state['messages'] + [response]
+        # 某些模型不支持 assistant message prefill，要求最后一条消息必须是 user
+        state['messages'] = state['messages'] + [response, HumanMessage(content=self._get_follow_up_prompt())]
         return state
 
+    def _get_follow_up_prompt(self) -> str:
+        """根据语言返回引导主 agent 回复的提示"""
+        prompts = {
+            "ja": "以上の分析に基づいて、ユーザーに返信してください。",
+            "jp": "以上の分析に基づいて、ユーザーに返信してください。",
+            "zh": "请根据以上分析，回复用户。",
+            "zh-TW": "請根據以上分析，回覆用戶。",
+            "ko": "위 분석을 바탕으로 사용자에게 답변해 주세요.",
+            "en": "Based on the above analysis, please respond to the user.",
+        }
+        return prompts.get(self.language, prompts["en"])
+
     def wrap_model_call(
         self,
         request: ModelRequest,

agent/plugin_hook_loader.py

@@ -165,6 +165,8 @@ async def _execute_command(skill_path: str, command: str, hook_type: str, config
     """
     try:
         # 设置环境变量，传递给子进程
+        # 注意：subprocess 要求所有 env 值必须是 str 类型，
+        # getattr 可能返回 None（属性存在但值为 None），需要确保转换为 str
         env = os.environ.copy()
         env['ASSISTANT_ID'] = str(getattr(config, 'bot_id', ''))
         env['USER_IDENTIFIER'] = str(getattr(config, 'user_identifier', ''))
@@ -176,16 +178,17 @@ async def _execute_command(skill_path: str, command: str, hook_type: str, config
         # 合并 config 中的自定义 shell 环境变量
         shell_env = getattr(config, 'shell_env', None)
         if shell_env:
-            env.update(shell_env)
+            # 确保所有自定义环境变量值也是字符串
+            env.update({k: str(v) if v is not None else '' for k, v in shell_env.items()})
 
         # 对于 PreSave，传递 content
         if hook_type == 'PreSave':
-            env['CONTENT'] = kwargs.get('content', '')
+            env['CONTENT'] = str(kwargs.get('content', '') or '')
-            env['ROLE'] = kwargs.get('role', '')
+            env['ROLE'] = str(kwargs.get('role', '') or '')
 
         # 对于 PostAgent，传递 response
         if hook_type == 'PostAgent':
-            env['RESPONSE'] = kwargs.get('response', '')
+            env['RESPONSE'] = str(kwargs.get('response', '') or '')
             metadata = kwargs.get('metadata', {})
             env['METADATA'] = json.dumps(metadata) if metadata else ''
 

docker-compose.yml

@@ -12,7 +12,6 @@ services:
       # 应用配置
       - BACKEND_HOST=http://api-dev.gbase.ai
       - MAX_CONTEXT_TOKENS=262144
-      - DEFAULT_THINKING_ENABLE=true
       - R2_UPLOAD_CONFIG=/app/config/s3-upload-sparticle.yaml
     volumes:
       # 挂载项目数据目录

mcp/rag_retrieve_server.py

@@ -29,6 +29,49 @@ from mcp_common import (
 BACKEND_HOST = os.getenv("BACKEND_HOST", "https://api-dev.gptbase.ai")
 MASTERKEY = os.getenv("MASTERKEY", "master")
 
+# Citation instruction prefixes injected into tool results
+DOCUMENT_CITATION_INSTRUCTIONS = """<CITATION_INSTRUCTIONS>
+When using the retrieved knowledge below, you MUST add XML citation tags for factual claims.
+
+## Document Knowledge
+Format: `<CITATION file="file_uuid" filename="name.pdf" page=3 />`
+- Use `file` attribute with the UUID from document markers
+- Use `filename` attribute with the actual filename from document markers
+- Use `page` attribute (singular) with the page number
+- `page` MUST be 0-based and must match the `pages:` values shown in the learned knowledge context
+
+## Web Page Knowledge
+Format: `<CITATION url="https://example.com/page" />`
+- Use `url` attribute with the web page URL from the source metadata
+- Do not use `file`, `filename`, or `page` attributes for web sources
+- If content is grounded in a web source, prefer a web citation with `url` over a file citation
+
+## Placement Rules
+- Citations MUST appear IMMEDIATELY AFTER the paragraph or bullet list that uses the knowledge
+- NEVER collect all citations and place them at the end of your response
+- Limit to 1-2 citations per paragraph/bullet list
+- If your answer uses learned knowledge, you MUST generate at least 1 `<CITATION ... />` in the response
+</CITATION_INSTRUCTIONS>
+
+"""
+
+TABLE_CITATION_INSTRUCTIONS = """<CITATION_INSTRUCTIONS>
+When using the retrieved table knowledge below, you MUST add XML citation tags for factual claims.
+
+Format: `<CITATION file="file_id" filename="name.xlsx" sheet=1 rows=[2, 4] />`
+- Parse `__src`: `F1S2R5` = file_ref F1, sheet 2, row 5
+- Look up file_id in `file_ref_table`
+- Combine same-sheet rows into one citation: `rows=[2, 4, 6]`
+- MANDATORY: Create SEPARATE citation for EACH (file, sheet) combination
+- NEVER put <CITATION> on the same line as a bullet point or table row
+- Citations MUST be on separate lines AFTER the complete list/table
+- NEVER include the `__src` column in your response - it is internal metadata only
+- Citations MUST appear IMMEDIATELY AFTER the paragraph or bullet list that uses the knowledge
+- NEVER collect all citations and place them at the end of your response
+</CITATION_INSTRUCTIONS>
+
+"""
+
 def rag_retrieve(query: str, top_k: int = 100) -> Dict[str, Any]:
     """调用RAG检索API"""
     try:
@@ -94,7 +137,7 @@ def rag_retrieve(query: str, top_k: int = 100) -> Dict[str, Any]:
                 "content": [
                     {
                         "type": "text",
-                        "text": markdown_content
+                        "text": DOCUMENT_CITATION_INSTRUCTIONS + markdown_content
                     }
                 ]
             }
@@ -179,7 +222,7 @@ def table_rag_retrieve(query: str) -> Dict[str, Any]:
                 "content": [
                     {
                         "type": "text",
-                        "text": markdown_content
+                        "text": TABLE_CITATION_INSTRUCTIONS + markdown_content
                     }
                 ]
             }

prompt/FACT_RETRIEVAL_PROMPT.md

@@ -148,21 +148,6 @@ Output: {{"facts" : []}}
 Input: DR1の照明状態を教えて
 Output: {{"facts" : []}}
 
-Input: 私は林檎好きです
-Output: {{"facts" : ["林檎が好き"]}}
-
-Input: コーヒー飲みたい、毎朝
-Output: {{"facts" : ["毎朝コーヒーを飲みたい"]}}
-
-Input: 昨日映画見た、すごくよかった
-Output: {{"facts" : ["昨日映画を見た", "映画がすごくよかった"]}}
-
-Input: 我喜欢吃苹果
-Output: {{"facts" : ["喜欢吃苹果"]}}
-
-Input: 나는 사과를 좋아해
-Output: {{"facts" : ["사과를 좋아함"]}}
-
 Return the facts and preferences in a json format as shown above.
 
 Remember the following:

prompt/novare.md

@@ -74,6 +74,13 @@
 - dxcore_update_device_status(device_id="[B设备id]",running_control=0) → 灯光亮度调整为0
 **响应**："已为您关闭Define Room4的灯光"
 
+### 位置降级搜索场景
+**用户**："3階執務スペース、フォーラム側窓側の照明をつけて"
+- find_device_by_area(description="3階執務スペース、フォーラム側窓側", device_type="light") → 返回无结果
+- find_device_by_area(description="3階執務スペース", device_type="light") → 降级搜索，找到设备
+- 告知用户是基于"3階執務スペース"范围搜索到的结果，并确认是否操作
+**响应**："「3階執務スペース、フォーラム側窓側」では見つかりませんでしたが、3階執務スペースエリアで照明が見つかりました。こちらの照明を操作しますか？"
+
 </scenarios>
 
 
@@ -92,6 +99,17 @@
         ▪  主动向用户确认：向用户列出所有候选房间，并提示用户选择或明确具体是哪一个。确认提示语可参考：“请问您想查询的是以下哪个房间？[列出候选房间列表]”。
         ▪  理解用户二次确认：等待用户回复后，根据其选择再次调用查询工具获取最终信息。用户对候选房间的指明（如回复“第一个”或重复房间名）应视为对该房间的确认。
     4.  处理无匹配结果：如果工具返回未找到任何相关房间，应明确告知用户这一情况，并建议用户检查房间名称是否正确或提供更多线索。
+    5.  **位置粒度降级搜索（詳細な位置指定で見つからない場合）**：
+        用户指定了详细的位置信息（如包含方位、区域细节），但工具返回无匹配结果时，自动执行降级搜索：
+        - **第1步**：从位置描述中去除方位修饰语（側、付近、奥、手前、寄り等）和细节描述，保留核心区域名重新搜索
+          - 例: "3階執務スペース、フォーラム側窓側" → find_device_by_area(description="3階執務スペース")
+          - 例: "2階会議室A、入口付近" → find_device_by_area(description="2階会議室A")
+        - **第2步**：如果仍无结果，进一步简化到楼层+大区域级别
+          - 例: "3階執務スペース" → find_device_by_area(description="3階")
+        - **降级成功时的回复**：告知用户是基于更广范围的搜索结果，让用户确认
+          - 回复格式: "「{元の位置}」では見つかりませんでしたが、{簡略化した位置}エリアで以下の設備が見つかりました。こちらでよろしいですか？"
+        - **全部失败时**：告知用户未找到设备，建议提供其他位置信息或直接指定房间名
+          - 回复格式: "申し訳ございません、該当エリアでは操作可能な設備が見つかりませんでした。お部屋の名前をお教えいただけますか？"
 
 3. 更新设备（此操作需要确认）
 -   **条件**：用户意图为控制设备或调节参数（如开关、温度、风速）, 需要进行确认。
@@ -105,6 +123,7 @@
         - 通过 find_employee_location(name="[当前用户名字/邮箱]") 获取用户的sensor_id
         - 然后通过 find_iot_device(target_sensor_id="[当前用户的sensor_id]", device_type="[目标设备类型]") 查找他附近的设备
         - 找到设备后告知用户找到的设备信息，并确认是否执行操作
+        - **位置指定但匹配失败时**：如果用户指定了详细位置（如"3階執務スペース、フォーラム側窓側の照明をつけて"），但 find_device_by_area 返回无匹配结果，应按照规则 2 第 5 点的**位置粒度降级搜索**策略执行，而不是直接回复"找不到设备"
     3.  **空调温度调节确认方式**：
         - 如果用户说"有点热"、"调低点"、"太热了"等，表示要降温：
             1. 先查询当前室温
@@ -142,8 +161,8 @@
         - 如果用户指定了具体档位（如"调到强"），直接使用指定档位
         - **边界情况**：如果已达到最高档（强）或最低档（弱）无法继续调整，告知用户并主动建议调整温度
             - 回复格式："風量は既に『強/弱』になっていますので、これ以上調整できません。代わりに温度を調整しますか？"
-    6.  **若用户已明确确认**：直接调用【设备控制】工具执行操作。
+    6.  **若用户已明确确认**：**立即**调用【设备控制】工具执行操作，不做任何额外确认或复述。确认后的唯一动作就是调用工具。
-    7.  **若用户未确认且为新请求**：向用户发送确认提示："即将为您 [操作内容] [设备名称] [具体参数]，是否确认？"，待用户确认后再执行。
+    7.  **若用户未确认且为新请求**：向用户发送确认提示："即将为您 [操作内容] [设备名称] [具体参数]，是否确认？"，待用户确认后再执行。每个操作只确认一次。
 
 4. 查询人员信息/wowtalk账号/人员位置
 -   **条件**：用户意图为查找某人、员工、同事或房间位置。
@@ -190,15 +209,43 @@
 - 影响范围大的操作：影响整个房间或楼层的设备控制
 
 ### 用户确认意图推理
-- 用户明确确认：如回复“确认”、“好的”、“是的”、“拜托了”、“よろしく”、“请”、“please”等肯定性语气的内容。
+- 用户明确确认：如回复”确认”、”好的”、”是的”、”拜托了”、”よろしく”、”请”、”please”、”お願いします”、”お願い”、”はい”、”うん”、”ええ”、”了解”、”OK”、”分かりました”、”そうしてください”、”それでお願い”等肯定性语气的内容。
-- 用户意图重申：用户完整或核心重复当前待执行的操作指令。（例如，提示“room302の照明1台を明るさ50%に調整してもよろしいですか?”，用户回复“room302の照明を明るさ50%に変更”)
+- 用户意图重申：用户完整或核心重复当前待执行的操作指令。（例如，提示”room302の照明1台を明るさ50%に調整してもよろしいですか?”，用户回复”room302の照明を明るさ50%に変更”)
 - 同一设备免重复确认：如果用户在当前会话中已经对某个设备的操作进行过确认，后续针对**同一设备**的操作可直接执行，无需再次确认。判定标准为：
-    1. **同一设备的不同操作**：用户已确认过对某设备的控制操作后，后续对该设备的其他操作无需再次确认（如已确认关闭Define Room4的灯光，之后用户说"把灯打开"，可直接执行）
+    1. **同一设备的不同操作**：用户已确认过对某设备的控制操作后，后续对该设备的其他操作无需再次确认（如已确认关闭Define Room4的灯光，之后用户说”把灯打开”，可直接执行）
-    2. **同一轮对话意图**：用户在一轮连续交互中围绕同一目标发出的多步操作（如用户确认"关闭Define Room4的灯光"后，系统依次关闭该房间内多个灯光设备，无需逐个确认）
+    2. **同一轮对话意图**：用户在一轮连续交互中围绕同一目标发出的多步操作（如用户确认”关闭Define Room4的灯光”后，系统依次关闭该房间内多个灯光设备，无需逐个确认）
     3. **同一指令的延续执行**：用户确认某操作后，该操作因技术原因需要分步执行的后续步骤（如批量控制多个设备时，确认一次即可全部执行）
-    4. **上下文明确的追加操作**：用户在已确认的操作基础上追加相同类型的操作，且目标明确无歧义（如已确认打开A房间空调后，用户说"B房间也一样"，可直接执行）
+    4. **上下文明确的追加操作**：用户在已确认的操作基础上追加相同类型的操作，且目标明确无歧义（如已确认打开A房间空调后，用户说”B房间也一样”，可直接执行）
 - 不同事项仍需确认：当操作涉及**未曾确认过的新设备**，或操作类型发生本质变化时（如从设备控制切换到消息通知），仍需重新确认
 
+### ⚠️ 禁止二次确认（最高优先级规则）
+**对于同一个操作请求，最多只能向用户确认一次。用户确认后，必须立即调用工具执行，绝对禁止再次询问确认。**
+
+核心规则：
+1. **一次确认，立即执行**：当你向用户发出确认提示后，用户回复确认，你的下一步动作**必须且只能是**调用对应的工具（如 dxcore_update_device_status）执行操作。不允许生成任何额外的确认、复述或再次询问。
+2. **禁止循环确认**：如果聊天记录中已经存在你发出的确认提示和用户的确认回复，则该操作已被确认，不得以任何理由再次要求确认。
+3. **确认后禁止的行为**：
+   - ❌ 再次询问”もう一度確認いただけますか？”
+   - ❌ 再次复述操作内容并要求确认
+   - ❌ 以不同措辞重新询问同一操作的确认
+   - ❌ 生成过渡性文字后再次要求确认
+
+**正确流程示例**：
+```
+用户: “Dr3の照明を30%にして”
+AI: “ディファインルーム3の照明を30%に調整してもよろしいですか？”
+用户: “お願いします”
+AI: [立即调用 dxcore_update_device_status 执行] → “照明を30%に調整しました。”
+```
+
+**错误流程（绝对禁止）**：
+```
+用户: “Dr3の照明を30%にして”
+AI: “ディファインルーム3の照明を30%に調整してもよろしいですか？”
+用户: “お願いします”
+AI: “もう一度確認いただければ実行いたします” ← ❌ 禁止！
+```
+
 ## 上下文推理示例
 
 ### 设备控制场景

prompt/system_prompt.md

@@ -1,5 +1,15 @@
 {extra_prompt}
 
+## CITATION REQUIREMENTS
+
+When your answer uses learned knowledge, you MUST generate `<CITATION ... />` tags. Follow the specific citation format instructions returned by each tool (`rag_retrieve`, `table_rag_retrieve`).
+
+### General Placement Rules
+1. Citations MUST appear IMMEDIATELY AFTER the paragraph or bullet list that uses the knowledge
+2. NEVER collect all citations and place them at the end of your response
+3. Limit to 1-2 citations per paragraph/bullet list - combine related facts under one citation
+4. If your answer uses learned knowledge, you MUST generate at least 1 `<CITATION ... />` in the response
+
 ### Current Working Directory
 
 PROJECT_ROOT: `{agent_dir_path}`
@@ -92,15 +102,3 @@ Trace Id: {trace_id}
 - Even when the user writes in a different language, you MUST still reply in [{language}].
 - Do NOT mix languages. Do NOT fall back to English or any other language under any circumstances.
 - Technical terms, code identifiers, file paths, and tool names may remain in their original form, but all surrounding text MUST be in [{language}].
-
-**Citation Requirement (RAG Only)**: When answering questions based on `rag_retrieve` tool results, you MUST add XML citation tags for factual claims derived from the knowledge base.
-
-**MANDATORY FORMAT**: `The cited factual claim <CITATION file="file_uuid" page="3" />`
-
-**Citation Rules**:
-- The citation tag MUST be placed immediately after the factual claim or paragraph
-- The `file` attribute MUST use the exact `File ID` from `rag_retrieve` document
-- The `page` attribute MUST use the exact `Page Number` from `rag_retrieve` document
-- If multiple sources support the same claim, include separate citation tags for each source
-- Example: `According to the policy, returns are accepted within 30 days <CITATION file="abc123" page="5" />.`
-- This requirement ONLY applies when using `rag_retrieve` results to answer questions.

pyproject.toml

@@ -26,12 +26,12 @@ dependencies = [
     "chardet>=5.0.0",
     "psutil (>=7.1.3,<8.0.0)",
     "uvloop (>=0.22.1,<0.23.0)",
-    "deepagents (>=0.4.3,<0.5.0)",
+    "deepagents (>=0.5.2,<0.6.0)",
     "langchain-mcp-adapters (>=0.2.1,<0.3.0)",
     "langchain-openai (>=1.1.1,<2.0.0)",
     "cachetools (>=6.2.4,<7.0.0)",
     "langgraph-checkpoint-postgres (>=3.0.0,<4.0.0)",
-    "deepagents-cli (>=0.0.25,<0.0.26)",
+    "deepagents-cli (>=0.0.37,<0.1.0)",
     "mem0ai (>=0.1.50,<0.3.0)",
     "psycopg2-binary (>=2.9.11,<3.0.0)",
     "json-repair (>=0.29.0,<0.30.0)",

requirements.txt

@@ -1,56 +1,51 @@
-aiofiles==24.1.0 ; python_version >= "3.12" and python_version < "3.15"
+agent-client-protocol==0.9.0 ; python_version >= "3.12" and python_version < "3.15"
+aiofiles==25.1.0 ; python_version >= "3.12" and python_version < "3.15"
 aiohappyeyeballs==2.6.1 ; python_version >= "3.12" and python_version < "3.15"
-aiohttp-retry==2.9.1 ; python_version >= "3.12" and python_version < "3.15"
+aiohttp==3.13.5 ; python_version >= "3.12" and python_version < "3.15"
-aiohttp==3.13.3 ; python_version >= "3.12" and python_version < "3.15"
 aiosignal==1.4.0 ; python_version >= "3.12" and python_version < "3.15"
 aiosqlite==0.22.1 ; python_version >= "3.12" and python_version < "3.15"
-annotated-doc==0.0.4 ; python_version >= "3.12" and python_version < "3.15"
 annotated-types==0.7.0 ; python_version >= "3.12" and python_version < "3.15"
-anthropic==0.84.0 ; python_version >= "3.12" and python_version < "3.15"
+anthropic==0.94.0 ; python_version >= "3.12" and python_version < "3.15"
-anyio==4.12.1 ; python_version >= "3.12" and python_version < "3.15"
+anyio==4.13.0 ; python_version >= "3.12" and python_version < "3.15"
-attrs==25.4.0 ; python_version >= "3.12" and python_version < "3.15"
+attrs==26.1.0 ; python_version >= "3.12" and python_version < "3.15"
 backoff==2.2.1 ; python_version >= "3.12" and python_version < "3.15"
 beartype==0.22.9 ; python_version >= "3.12" and python_version < "3.15"
 beautifulsoup4==4.14.3 ; python_version >= "3.12" and python_version < "3.15"
+blockbuster==1.5.26 ; python_version >= "3.12" and python_version < "3.15"
 bracex==2.6 ; python_version >= "3.12" and python_version < "3.15"
 cachetools==6.2.6 ; python_version >= "3.12" and python_version < "3.15"
-cbor2==5.8.0 ; python_version >= "3.12" and python_version < "3.15"
 certifi==2026.2.25 ; python_version >= "3.12" and python_version < "3.15"
 cffi==2.0.0 ; python_version >= "3.12" and python_version < "3.15" and platform_python_implementation != "PyPy"
-chardet==7.1.0 ; python_version >= "3.12" and python_version < "3.15"
+chardet==7.4.1 ; python_version >= "3.12" and python_version < "3.15"
-charset-normalizer==3.4.5 ; python_version >= "3.12" and python_version < "3.15"
+charset-normalizer==3.4.7 ; python_version >= "3.12" and python_version < "3.15"
-click==8.3.1 ; python_version >= "3.12" and python_version < "3.15"
+click==8.3.2 ; python_version >= "3.12" and python_version < "3.15"
+cloudpickle==3.1.2 ; python_version >= "3.12" and python_version < "3.15"
 colorama==0.4.6 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Windows"
 croniter==3.0.4 ; python_version >= "3.12" and python_version < "3.15"
-cryptography==46.0.5 ; python_version >= "3.12" and python_version < "3.15"
+cryptography==46.0.7 ; python_version >= "3.12" and python_version < "3.15"
-daytona-api-client-async==0.151.0 ; python_version >= "3.12" and python_version < "3.15"
+deepagents-acp==0.0.5 ; python_version >= "3.12" and python_version < "3.15"
-daytona-api-client==0.151.0 ; python_version >= "3.12" and python_version < "3.15"
+deepagents-cli==0.0.37 ; python_version >= "3.12" and python_version < "3.15"
-daytona-toolbox-api-client-async==0.151.0 ; python_version >= "3.12" and python_version < "3.15"
+deepagents==0.5.2 ; python_version >= "3.12" and python_version < "3.15"
-daytona-toolbox-api-client==0.151.0 ; python_version >= "3.12" and python_version < "3.15"
-daytona==0.151.0 ; python_version >= "3.12" and python_version < "3.15"
-deepagents-cli==0.0.25 ; python_version >= "3.12" and python_version < "3.15"
-deepagents==0.4.3 ; python_version >= "3.12" and python_version < "3.15"
 defusedxml==0.7.1 ; python_version >= "3.12" and python_version < "3.15"
-deprecated==1.3.1 ; python_version >= "3.12" and python_version < "3.15"
 distro==1.9.0 ; python_version >= "3.12" and python_version < "3.15"
 docstring-parser==0.17.0 ; python_version >= "3.12" and python_version < "3.15"
-environs==14.6.0 ; python_version >= "3.12" and python_version < "3.15"
 et-xmlfile==2.0.0 ; python_version >= "3.12" and python_version < "3.15"
 fastapi==0.116.1 ; python_version >= "3.12" and python_version < "3.15"
 filelock==3.25.2 ; python_version >= "3.12" and python_version < "3.15"
 filetype==1.2.0 ; python_version >= "3.12" and python_version < "3.15"
+forbiddenfruit==0.1.4 ; python_version >= "3.12" and python_version < "3.15" and implementation_name == "cpython"
 frozenlist==1.8.0 ; python_version >= "3.12" and python_version < "3.15"
-fsspec==2026.2.0 ; python_version >= "3.12" and python_version < "3.15"
+fsspec==2026.3.0 ; python_version >= "3.12" and python_version < "3.15"
-google-auth==2.49.1 ; python_version >= "3.12" and python_version < "3.15"
+google-auth==2.49.2 ; python_version >= "3.12" and python_version < "3.15"
-google-genai==1.67.0 ; python_version >= "3.12" and python_version < "3.15"
+google-genai==1.72.0 ; python_version >= "3.12" and python_version < "3.15"
-googleapis-common-protos==1.73.0 ; python_version >= "3.12" and python_version < "3.15"
+googleapis-common-protos==1.74.0 ; python_version >= "3.12" and python_version < "3.15"
-greenlet==3.3.2 ; python_version >= "3.12" and python_version < "3.15" and (platform_machine == "aarch64" or platform_machine == "ppc64le" or platform_machine == "x86_64" or platform_machine == "amd64" or platform_machine == "AMD64" or platform_machine == "win32" or platform_machine == "WIN32")
+greenlet==3.4.0 ; python_version >= "3.12" and python_version < "3.15" and (platform_machine == "aarch64" or platform_machine == "ppc64le" or platform_machine == "x86_64" or platform_machine == "amd64" or platform_machine == "AMD64" or platform_machine == "win32" or platform_machine == "WIN32")
-grpcio-tools==1.71.2 ; python_version >= "3.13" and python_version < "3.15"
+grpcio-health-checking==1.76.0 ; python_version >= "3.12" and python_version < "3.15"
-grpcio==1.78.0 ; python_version >= "3.12" and python_version < "3.15"
+grpcio-tools==1.76.0 ; python_version >= "3.12" and python_version < "3.15"
-grpclib==0.4.9 ; python_version >= "3.12" and python_version < "3.15"
+grpcio==1.76.0 ; python_version >= "3.12" and python_version < "3.15"
 h11==0.16.0 ; python_version >= "3.12" and python_version < "3.15"
 h2==4.3.0 ; python_version >= "3.12" and python_version < "3.15"
-hf-xet==1.4.2 ; python_version >= "3.12" and python_version < "3.15" and (platform_machine == "x86_64" or platform_machine == "amd64" or platform_machine == "arm64" or platform_machine == "aarch64")
+hf-xet==1.4.3 ; python_version >= "3.12" and python_version < "3.15" and (platform_machine == "x86_64" or platform_machine == "amd64" or platform_machine == "arm64" or platform_machine == "aarch64")
 hpack==4.1.0 ; python_version >= "3.12" and python_version < "3.15"
 httpcore==1.0.9 ; python_version >= "3.12" and python_version < "3.15"
 httpx-sse==0.4.3 ; python_version >= "3.12" and python_version < "3.15"
@@ -61,94 +56,94 @@ hyperframe==6.1.0 ; python_version >= "3.12" and python_version < "3.15"
 idna==3.11 ; python_version >= "3.12" and python_version < "3.15"
 importlib-metadata==8.7.1 ; python_version >= "3.12" and python_version < "3.15"
 jinja2==3.1.6 ; python_version >= "3.12" and python_version < "3.15"
-jiter==0.13.0 ; python_version >= "3.12" and python_version < "3.15"
+jiter==0.14.0 ; python_version >= "3.12" and python_version < "3.15"
 joblib==1.5.3 ; python_version >= "3.12" and python_version < "3.15"
 json-repair==0.29.10 ; python_version >= "3.12" and python_version < "3.15"
-json5==0.13.0 ; python_version >= "3.12" and python_version < "3.15"
+json5==0.14.0 ; python_version >= "3.12" and python_version < "3.15"
 jsonpatch==1.33 ; python_version >= "3.12" and python_version < "3.15"
-jsonpointer==3.0.0 ; python_version >= "3.12" and python_version < "3.15"
+jsonpointer==3.1.1 ; python_version >= "3.12" and python_version < "3.15"
+jsonschema-rs==0.29.1 ; python_version >= "3.12" and python_version < "3.15"
 jsonschema-specifications==2025.9.1 ; python_version >= "3.12" and python_version < "3.15"
 jsonschema==4.26.0 ; python_version >= "3.12" and python_version < "3.15"
-langchain-anthropic==1.3.5 ; python_version >= "3.12" and python_version < "3.15"
+langchain-anthropic==1.4.0 ; python_version >= "3.12" and python_version < "3.15"
-langchain-core==1.2.19 ; python_version >= "3.12" and python_version < "3.15"
+langchain-core==1.2.28 ; python_version >= "3.12" and python_version < "3.15"
 langchain-google-genai==4.2.1 ; python_version >= "3.12" and python_version < "3.15"
-langchain-mcp-adapters==0.2.1 ; python_version >= "3.12" and python_version < "3.15"
+langchain-mcp-adapters==0.2.2 ; python_version >= "3.12" and python_version < "3.15"
-langchain-openai==1.1.11 ; python_version >= "3.12" and python_version < "3.15"
+langchain-openai==1.1.12 ; python_version >= "3.12" and python_version < "3.15"
-langchain==1.2.12 ; python_version >= "3.12" and python_version < "3.15"
+langchain==1.2.15 ; python_version >= "3.12" and python_version < "3.15"
-langgraph-checkpoint-postgres==3.0.4 ; python_version >= "3.12" and python_version < "3.15"
+langgraph-api==0.7.28 ; python_version >= "3.12" and python_version < "3.15"
+langgraph-checkpoint-postgres==3.0.5 ; python_version >= "3.12" and python_version < "3.15"
 langgraph-checkpoint-sqlite==3.0.3 ; python_version >= "3.12" and python_version < "3.15"
 langgraph-checkpoint==4.0.1 ; python_version >= "3.12" and python_version < "3.15"
-langgraph-prebuilt==1.0.8 ; python_version >= "3.12" and python_version < "3.15"
+langgraph-cli==0.4.21 ; python_version >= "3.12" and python_version < "3.15"
-langgraph-sdk==0.3.11 ; python_version >= "3.12" and python_version < "3.15"
+langgraph-prebuilt==1.0.9 ; python_version >= "3.12" and python_version < "3.15"
-langgraph==1.1.2 ; python_version >= "3.12" and python_version < "3.15"
+langgraph-runtime-inmem==0.24.1 ; python_version >= "3.12" and python_version < "3.15"
-langsmith==0.7.17 ; python_version >= "3.12" and python_version < "3.15"
+langgraph-sdk==0.3.13 ; python_version >= "3.12" and python_version < "3.15"
+langgraph==1.1.6 ; python_version >= "3.12" and python_version < "3.15"
+langsmith==0.7.30 ; python_version >= "3.12" and python_version < "3.15"
 linkify-it-py==2.1.0 ; python_version >= "3.12" and python_version < "3.15"
 markdown-it-py==4.0.0 ; python_version >= "3.12" and python_version < "3.15"
 markdownify==1.2.2 ; python_version >= "3.12" and python_version < "3.15"
 markupsafe==3.0.3 ; python_version >= "3.12" and python_version < "3.15"
-marshmallow==4.2.2 ; python_version >= "3.12" and python_version < "3.15"
 mcp==1.12.4 ; python_version >= "3.12" and python_version < "3.15"
 mdit-py-plugins==0.5.0 ; python_version >= "3.12" and python_version < "3.15"
 mdurl==0.1.2 ; python_version >= "3.12" and python_version < "3.15"
-mem0ai==0.1.116 ; python_version >= "3.12" and python_version < "3.15"
+mem0ai==0.1.115 ; python_version >= "3.12" and python_version < "3.15"
-modal==1.3.5 ; python_version >= "3.12" and python_version < "3.15"
 mpmath==1.3.0 ; python_version >= "3.12" and python_version < "3.15"
 multidict==6.7.1 ; python_version >= "3.12" and python_version < "3.15"
 networkx==3.6 ; python_version == "3.14"
 networkx==3.6.1 ; python_version >= "3.12" and python_version < "3.14"
 numpy==1.26.4 ; python_version >= "3.12" and python_version < "3.15"
-nvidia-cublas-cu12==12.1.3.1 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-cublas-cu12==12.1.3.1 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-nvidia-cuda-cupti-cu12==12.1.105 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-cuda-cupti-cu12==12.1.105 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-nvidia-cuda-nvrtc-cu12==12.1.105 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-cuda-nvrtc-cu12==12.1.105 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-nvidia-cuda-runtime-cu12==12.1.105 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-cuda-runtime-cu12==12.1.105 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-nvidia-cudnn-cu12==8.9.2.26 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-cudnn-cu12==8.9.2.26 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-nvidia-cufft-cu12==11.0.2.54 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-cufft-cu12==11.0.2.54 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-nvidia-curand-cu12==10.3.2.106 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-curand-cu12==10.3.2.106 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-nvidia-cusolver-cu12==11.4.5.107 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-cusolver-cu12==11.4.5.107 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-nvidia-cusparse-cu12==12.1.0.106 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-cusparse-cu12==12.1.0.106 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-nvidia-nccl-cu12==2.19.3 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-nccl-cu12==2.19.3 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-nvidia-nvjitlink-cu12==12.9.86 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-nvjitlink-cu12==12.9.86 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-nvidia-nvtx-cu12==12.1.105 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+nvidia-nvtx-cu12==12.1.105 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-obstore==0.8.2 ; python_version >= "3.12" and python_version < "3.15"
+openai==2.31.0 ; python_version >= "3.12" and python_version < "3.15"
-openai==2.28.0 ; python_version >= "3.12" and python_version < "3.15"
 openpyxl==3.1.5 ; python_version >= "3.12" and python_version < "3.15"
-opentelemetry-api==1.40.0 ; python_version >= "3.12" and python_version < "3.15"
+opentelemetry-api==1.41.0 ; python_version >= "3.12" and python_version < "3.15"
-opentelemetry-exporter-otlp-proto-common==1.40.0 ; python_version >= "3.12" and python_version < "3.15"
+opentelemetry-exporter-otlp-proto-common==1.41.0 ; python_version >= "3.12" and python_version < "3.15"
-opentelemetry-exporter-otlp-proto-http==1.40.0 ; python_version >= "3.12" and python_version < "3.15"
+opentelemetry-exporter-otlp-proto-http==1.41.0 ; python_version >= "3.12" and python_version < "3.15"
-opentelemetry-instrumentation-aiohttp-client==0.61b0 ; python_version >= "3.12" and python_version < "3.15"
+opentelemetry-proto==1.41.0 ; python_version >= "3.12" and python_version < "3.15"
-opentelemetry-instrumentation==0.61b0 ; python_version >= "3.12" and python_version < "3.15"
+opentelemetry-sdk==1.41.0 ; python_version >= "3.12" and python_version < "3.15"
-opentelemetry-proto==1.40.0 ; python_version >= "3.12" and python_version < "3.15"
+opentelemetry-semantic-conventions==0.62b0 ; python_version >= "3.12" and python_version < "3.15"
-opentelemetry-sdk==1.40.0 ; python_version >= "3.12" and python_version < "3.15"
+orjson==3.11.8 ; python_version >= "3.12" and python_version < "3.15"
-opentelemetry-semantic-conventions==0.61b0 ; python_version >= "3.12" and python_version < "3.15"
-opentelemetry-util-http==0.61b0 ; python_version >= "3.12" and python_version < "3.15"
-orjson==3.11.7 ; python_version >= "3.12" and python_version < "3.15"
 ormsgpack==1.12.2 ; python_version >= "3.12" and python_version < "3.15"
 packaging==26.0 ; python_version >= "3.12" and python_version < "3.15"
 pandas==2.3.3 ; python_version == "3.14"
-pandas==3.0.1 ; python_version >= "3.12" and python_version < "3.14"
+pandas==3.0.2 ; python_version >= "3.12" and python_version < "3.14"
-pillow==12.1.1 ; python_version >= "3.12" and python_version < "3.15"
+pathspec==1.0.4 ; python_version >= "3.12" and python_version < "3.15"
-platformdirs==4.9.4 ; python_version >= "3.12" and python_version < "3.15"
+pillow==12.2.0 ; python_version >= "3.12" and python_version < "3.15"
+platformdirs==4.9.6 ; python_version >= "3.12" and python_version < "3.15"
 portalocker==2.10.1 ; python_version >= "3.13" and python_version < "3.15"
 portalocker==3.2.0 ; python_version == "3.12"
-posthog==7.9.12 ; python_version >= "3.12" and python_version < "3.15"
+posthog==7.11.0 ; python_version >= "3.12" and python_version < "3.15"
 prompt-toolkit==3.0.52 ; python_version >= "3.12" and python_version < "3.15"
 propcache==0.4.1 ; python_version >= "3.12" and python_version < "3.15"
-protobuf==5.29.6 ; python_version >= "3.12" and python_version < "3.15"
+protobuf==6.33.6 ; python_version >= "3.12" and python_version < "3.15"
 psutil==7.2.2 ; python_version >= "3.12" and python_version < "3.15"
 psycopg-pool==3.3.0 ; python_version >= "3.12" and python_version < "3.15"
 psycopg2-binary==2.9.11 ; python_version >= "3.12" and python_version < "3.15"
 psycopg==3.3.3 ; python_version >= "3.12" and python_version < "3.15"
 pyasn1-modules==0.4.2 ; python_version >= "3.12" and python_version < "3.15"
-pyasn1==0.6.2 ; python_version >= "3.12" and python_version < "3.15"
+pyasn1==0.6.3 ; python_version >= "3.12" and python_version < "3.15"
-pycparser==3.0 ; python_version >= "3.12" and python_version < "3.15" and implementation_name != "PyPy" and platform_python_implementation != "PyPy"
+pycparser==3.0 ; python_version >= "3.12" and python_version < "3.15" and platform_python_implementation != "PyPy" and implementation_name != "PyPy"
 pydantic-core==2.27.2 ; python_version >= "3.12" and python_version < "3.15"
 pydantic-settings==2.13.1 ; python_version >= "3.12" and python_version < "3.15"
 pydantic==2.10.5 ; python_version >= "3.12" and python_version < "3.15"
-pygments==2.19.2 ; python_version >= "3.12" and python_version < "3.15"
+pygments==2.20.0 ; python_version >= "3.12" and python_version < "3.15"
+pyjwt==2.12.1 ; python_version >= "3.12" and python_version < "3.15"
 pyperclip==1.11.0 ; python_version >= "3.12" and python_version < "3.15"
 python-dateutil==2.8.2 ; python_version >= "3.12" and python_version < "3.15"
 python-dotenv==1.2.2 ; python_version >= "3.12" and python_version < "3.15"
-python-multipart==0.0.22 ; python_version >= "3.12" and python_version < "3.15"
+python-multipart==0.0.26 ; python_version >= "3.12" and python_version < "3.15"
 pytz==2026.1.post1 ; python_version >= "3.12" and python_version < "3.15"
 pywin32==311 ; python_version >= "3.12" and python_version < "3.15" and (sys_platform == "win32" or platform_system == "Windows")
 pyyaml==6.0.3 ; python_version >= "3.12" and python_version < "3.15"
@@ -156,46 +151,42 @@ qdrant-client==1.12.1 ; python_version >= "3.13" and python_version < "3.15"
 qdrant-client==1.17.1 ; python_version == "3.12"
 ragflow-sdk==0.23.1 ; python_version >= "3.12" and python_version < "3.15"
 referencing==0.37.0 ; python_version >= "3.12" and python_version < "3.15"
-regex==2026.2.28 ; python_version >= "3.12" and python_version < "3.15"
+regex==2026.4.4 ; python_version >= "3.12" and python_version < "3.15"
 requests-toolbelt==1.0.0 ; python_version >= "3.12" and python_version < "3.15"
 requests==2.32.5 ; python_version >= "3.12" and python_version < "3.15"
-rich==14.3.3 ; python_version >= "3.12" and python_version < "3.15"
+rich==14.3.4 ; python_version >= "3.12" and python_version < "3.15"
 rpds-py==0.30.0 ; python_version >= "3.12" and python_version < "3.15"
-runloop-api-client==1.11.0 ; python_version >= "3.12" and python_version < "3.15"
 safetensors==0.7.0 ; python_version >= "3.12" and python_version < "3.15"
 scikit-learn==1.8.0 ; python_version >= "3.12" and python_version < "3.15"
 scipy==1.17.1 ; python_version >= "3.12" and python_version < "3.15"
 sentence-transformers==3.4.1 ; python_version >= "3.12" and python_version < "3.15"
 setuptools==70.3.0 ; python_version >= "3.12" and python_version < "3.15"
-shellingham==1.5.4 ; python_version >= "3.12" and python_version < "3.15"
 six==1.17.0 ; python_version >= "3.12" and python_version < "3.15"
 sniffio==1.3.1 ; python_version >= "3.12" and python_version < "3.15"
 soupsieve==2.8.3 ; python_version >= "3.12" and python_version < "3.15"
-sqlalchemy==2.0.48 ; python_version >= "3.12" and python_version < "3.15"
+sqlalchemy==2.0.49 ; python_version >= "3.12" and python_version < "3.15"
-sqlite-vec==0.1.6 ; python_version >= "3.12" and python_version < "3.15"
+sqlite-vec==0.1.9 ; python_version >= "3.12" and python_version < "3.15"
-sse-starlette==3.0.3 ; python_version >= "3.12" and python_version < "3.15"
+sse-starlette==2.1.3 ; python_version >= "3.12" and python_version < "3.15"
 starlette==0.47.3 ; python_version >= "3.12" and python_version < "3.15"
+structlog==25.5.0 ; python_version >= "3.12" and python_version < "3.15"
 sympy==1.14.0 ; python_version >= "3.12" and python_version < "3.15"
-synchronicity==0.11.1 ; python_version >= "3.12" and python_version < "3.15"
 tavily-python==0.7.23 ; python_version >= "3.12" and python_version < "3.15"
 tenacity==9.1.4 ; python_version >= "3.12" and python_version < "3.15"
 textual-autocomplete==4.0.6 ; python_version >= "3.12" and python_version < "3.15"
-textual==8.1.1 ; python_version >= "3.12" and python_version < "3.15"
+textual-speedups==0.2.1 ; python_version >= "3.12" and python_version < "3.15"
+textual==8.2.3 ; python_version >= "3.12" and python_version < "3.15"
 threadpoolctl==3.6.0 ; python_version >= "3.12" and python_version < "3.15"
 tiktoken==0.12.0 ; python_version >= "3.12" and python_version < "3.15"
 tokenizers==0.22.2 ; python_version >= "3.12" and python_version < "3.15"
-toml==0.10.2 ; python_version >= "3.12" and python_version < "3.15"
 tomli-w==1.2.0 ; python_version >= "3.12" and python_version < "3.15"
 torch==2.2.0 ; python_version >= "3.12" and python_version < "3.15"
 tqdm==4.67.3 ; python_version >= "3.12" and python_version < "3.15"
 transformers==4.57.6 ; python_version >= "3.12" and python_version < "3.15"
-triton==2.2.0 ; python_version >= "3.12" and python_version < "3.15" and platform_machine == "x86_64" and platform_system == "Linux"
+triton==2.2.0 ; python_version >= "3.12" and python_version < "3.15" and platform_system == "Linux" and platform_machine == "x86_64"
-typer==0.24.1 ; python_version >= "3.12" and python_version < "3.15"
+truststore==0.10.4 ; python_version >= "3.12" and python_version < "3.15"
-types-certifi==2021.10.8.3 ; python_version >= "3.12" and python_version < "3.15"
-types-toml==0.10.8.20240310 ; python_version >= "3.12" and python_version < "3.15"
 typing-extensions==4.15.0 ; python_version >= "3.12" and python_version < "3.15"
 typing-inspection==0.4.2 ; python_version >= "3.12" and python_version < "3.15"
-tzdata==2025.3 ; python_version >= "3.12" and python_version < "3.15" and (sys_platform == "win32" or sys_platform == "emscripten") or python_version == "3.14"
+tzdata==2026.1 ; python_version >= "3.12" and python_version < "3.15" and (sys_platform == "win32" or sys_platform == "emscripten") or python_version == "3.14"
 uc-micro-py==2.0.0 ; python_version >= "3.12" and python_version < "3.15"
 urllib3==2.6.3 ; python_version >= "3.12" and python_version < "3.15"
 uuid-utils==0.14.1 ; python_version >= "3.12" and python_version < "3.15"
@@ -206,7 +197,6 @@ wcmatch==10.1 ; python_version >= "3.12" and python_version < "3.15"
 wcwidth==0.6.0 ; python_version >= "3.12" and python_version < "3.15"
 webrtcvad==2.0.10 ; python_version >= "3.12" and python_version < "3.15"
 websockets==15.0.1 ; python_version >= "3.12" and python_version < "3.15"
-wrapt==1.17.3 ; python_version >= "3.12" and python_version < "3.15"
 wsgidav==4.3.3 ; python_version >= "3.12" and python_version < "3.15"
 xlrd==2.0.2 ; python_version >= "3.12" and python_version < "3.15"
 xxhash==3.6.0 ; python_version >= "3.12" and python_version < "3.15"

utils/api_models.py

@@ -5,7 +5,6 @@ API data models and response schemas.
 
 from typing import Dict, List, Optional, Any, AsyncGenerator
 from pydantic import BaseModel, Field, field_validator, ConfigDict
-from utils.settings import DEFAULT_THINKING_ENABLE
 
 class Message(BaseModel):
     role: str
@@ -52,7 +51,7 @@ class ChatRequest(BaseModel):
     mcp_settings: Optional[List[Dict]] = None
     user_identifier: Optional[str] = ""
     session_id: Optional[str] = None
-    enable_thinking: Optional[bool] = DEFAULT_THINKING_ENABLE
+    enable_thinking: Optional[bool] = False
     skills: Optional[List[str]] = None
     enable_memory: Optional[bool] = False
     shell_env: Optional[Dict[str, str]] = None

utils/settings.py

@@ -12,7 +12,7 @@ MAX_OUTPUT_TOKENS = int(os.getenv("MAX_OUTPUT_TOKENS", 8000))
 
 # 可选参数
 # Summarization Settings
-SUMMARIZATION_MAX_TOKENS = int(MAX_CONTEXT_TOKENS/3)
+SUMMARIZATION_MAX_TOKENS = int(MAX_CONTEXT_TOKENS/2)
 SUMMARIZATION_TOKENS_TO_KEEP = int(SUMMARIZATION_MAX_TOKENS/3)
 DEFAULT_TRIM_TOKEN_LIMIT = SUMMARIZATION_MAX_TOKENS - SUMMARIZATION_TOKENS_TO_KEEP + 5000
 
@@ -32,11 +32,9 @@ TOKENIZERS_PARALLELISM = os.getenv("TOKENIZERS_PARALLELISM", "true")
 SENTENCE_TRANSFORMER_MODEL = os.getenv("SENTENCE_TRANSFORMER_MODEL", "TaylorAI/gte-tiny")
 
 # Tool Output Length Control Settings
-TOOL_OUTPUT_MAX_LENGTH = int(SUMMARIZATION_MAX_TOKENS/4)
+TOOL_OUTPUT_MAX_LENGTH = SUMMARIZATION_MAX_TOKENS
 TOOL_OUTPUT_TRUNCATION_STRATEGY = os.getenv("TOOL_OUTPUT_TRUNCATION_STRATEGY", "smart")
 
-# THINKING ENABLE
-DEFAULT_THINKING_ENABLE = os.getenv("DEFAULT_THINKING_ENABLE", "true") == "true"
 
 
 # WebDAV Authentication
@@ -59,7 +57,7 @@ CHECKPOINT_DB_URL = os.getenv("CHECKPOINT_DB_URL", "postgresql://postgres:AeEGDB
 # 连接池大小
 # 同时可以持有的最大连接数
 CHECKPOINT_POOL_SIZE = int(os.getenv("CHECKPOINT_POOL_SIZE", "20"))
-MEM0_POOL_SIZE = int(os.getenv("MEM0_POOL_SIZE", "20"))
+MEM0_POOL_SIZE = int(os.getenv("MEM0_POOL_SIZE", "50"))
 
 # Checkpoint 自动清理配置
 # 是否启用自动清理旧 session