dumplingsAI 0.1.1__tar.gz

dumplingsai-0.1.1/Agent_Base_.py

@@ -0,0 +1,659 @@
+import json
+import os
+import platform
+import re
+import threading
+import time
+import uuid
+from typing import Any
+
+import requests
+from bs4 import BeautifulSoup
+
+try:
+    from .agent_tool import _builtin_promote_overrides, builtin_tool, tool_registry
+    from .logging_config import logger  # 配置日志
+except Exception:
+    raise ImportError("不可单独执行")
+
+
+class Agent():
+    """
+    所有具体 Dumplings 必须实现四个属性：
+        api_key
+        api_provider
+        model_name
+        prompt
+    """
+    prompt = None
+    api_provider = None
+    model_name = None
+    api_key = None
+    fc_model = True # 现在xml工具调用改为下位支持，如有bug修复优先级降低
+    stream = True
+    description = None
+
+    # ---------------- 类层级钩子：子类覆盖 @builtin_tool 方法时自动复用父类的 meta ----------------
+    def __init_subclass__(cls, **kwargs):
+        super().__init_subclass__(**kwargs)
+        _builtin_promote_overrides(cls)
+    # ---------------- 通用构造 ----------------
+    def __init__(self,new_load=True):
+        self.uuid=self.__class__.uuid
+        self.name=self.__class__.name
+        self.stream_run=False
+        self.current_task_id = None  # 当前任务 ID
+        self.tool_call_hooks = []     # 工具调用钩子列表
+        agent_name = getattr(self.__class__, 'name', None) or getattr(self.__class__, '__name__', None)
+        if agent_name and self.uuid:
+            tool_registry.register_agent_uuid(self.uuid, agent_name)
+
+        # 获取该 agent 有权限的所有工具信息
+        tools_info = tool_registry.get_all_tools_info(self.uuid)
+        tools_prompt = ""
+        tools_list = list(tools_info.keys())
+
+        # 通过自动收集器获取 @builtin_tool 装饰的内置工具 schema
+        builtin_schemas = tool_registry.collect_builtin_tools(self)
+        builtin_names = {s["function"]["name"] for s in builtin_schemas}
+        for name in builtin_names:
+            if name not in tools_list:
+                tools_list.append(name)
+
+        # 生成工具提示
+        if tools_list:
+            tools_prompt = "\n\n你可以使用以下工具：\n"
+            # 注册的工具
+            for tool_name, tool_info in tools_info.items():
+                tools_prompt += f"- {tool_name}: {tool_info['description']}\n"
+            # 内置工具（自动来自 @builtin_tool 装饰器）
+            for s in builtin_schemas:
+                n = s["function"]["name"]
+                if n in tools_list and n not in tools_info:
+                    tools_prompt += f"- {n}: {s['function']['description']}\n"
+            if not self.fc_model:
+                tools_prompt += "在使用xml格式的工具时应采用（无参数调用）<工具名></工具名>（含参数调用）<工具名><参数1>放入你想传入的内容</参数1>...</工具名>"
+
+        # 注入 Skills 信息
+        try:
+            from .skill import skill_registry
+            tools_prompt += skill_registry.get_skills_prompt_text(self.uuid)
+        except ImportError:
+            pass
+
+
+        prompt = self.prompt + tools_prompt + ", 你的uuid " + str(self.uuid)
+        logger.debug(f"Agent {self.name} 初始化，系统提示词长度：{len(prompt)}")
+        # print(prompt)
+        if new_load:
+            self.history = [{"role": "system", "content": prompt}]
+        else:
+            self.history[0] = {"role": "system", "content": prompt}
+        self.headers = {
+            "Authorization": f"Bearer {self.api_key}",
+            "Content-Type": "application/json"
+        }
+        self.os_name = platform.system()
+        self.conversations_folder = os.getcwd()
+        if self.os_name == "Windows":
+            self.os_main_folder = os.getenv("USERPROFILE")
+        elif self.os_name == "Linux":
+            self.os_main_folder = os.path.expanduser("~")
+        elif self.os_name == "Darwin":
+            self.os_main_folder = os.getenv("HOME")
+
+        threading.Thread(target=self.Connectivity, daemon=True).start()
+
+    # ---------------- 辅助方法 ----------------
+    def _generate_task_id(self):
+        """生成唯一任务 ID"""
+        return str(uuid.uuid4())
+
+    def _get_timestamp(self):
+        """获取当前时间戳（毫秒级）"""
+        return int(time.time() * 1000)
+
+    def register_tool_hook(self, hook_func):
+        """
+        注册工具调用钩子
+        hook_func 签名：hook_func(event_type, tool_name, tool_args, tool_result, task_id)
+        event_type: 'before' | 'after' | 'error'
+        """
+        self.tool_call_hooks.append(hook_func)
+
+    def _execute_hooks(self, event_type, tool_name, tool_args, tool_result=None):
+        """执行所有注册的钩子"""
+        for hook in self.tool_call_hooks:
+            try:
+                hook(
+                    event_type=event_type,
+                    tool_name=tool_name,
+                    tool_args=tool_args,
+                    tool_result=tool_result,
+                    task_id=self.current_task_id
+                )
+            except Exception as e:
+                logger.error(f"钩子执行失败：{e}")
+
+    # ---------------- 连通性测试 ----------------
+    def Connectivity(self):
+        payload = {
+            "model": self.model_name,
+            "messages": [{"role": "user", "content": "你好"}],
+            "stream": self.stream,
+            "stream_options": {"include_usage": True},
+            "max_tokens": 1
+        }
+        rsp = requests.post(self.api_provider,
+                            headers=self.headers,
+                            json=payload)
+        if rsp.status_code == 200:
+            logger.info(f"{self.name} 连接正常")
+        else:
+            logger.error(f"{self.name} 连接测试未通过，可能存在配置错误")
+        return rsp.status_code == 200
+
+    # ---------------- 主对话函数 ----------------
+    def conversation_with_tool(self, messages=None, tool=False, images=None):
+        """
+        进行对话，支持多模态输入（文本 + 图片）
+
+        Args:
+            messages: 文本消息
+            tool: 是否是工具调用后的继续对话
+            images: 图片列表，可以是 base64 字符串或图片 URL
+        """
+        work_history = self.history
+
+        if messages:
+            # 如果有图片，构建多模态内容
+            if images:
+                content_list = [{"type": "text", "text": messages}]
+                for img in images:
+                    if img.startswith("http"):
+                        content_list.append({
+                            "type": "image_url",
+                            "image_url": {"url": img}
+                        })
+                    else:
+                        # 假设是 base64
+                        content_list.append({
+                            "type": "image_url",
+                            "image_url": {"url": f"data:image/png;base64,{img}"}
+                        })
+                work_history.append({"role": "user", "content": content_list})
+            else:
+                work_history.append({"role": "user", "content": messages})
+
+        if self.fc_model:
+            # Function Calling 模式
+            tools_schema = tool_registry.get_all_tools_schema(self.uuid)
+
+            # 通过自动收集器添加内置工具（@builtin_tool 装饰）到 schema
+            for s in tool_registry.collect_builtin_tools(self):
+                tools_schema.append(s)
+
+            # 添加 Skills 到 Function Calling schema
+            try:
+                from .skill import skill_registry
+                tools_schema.extend(skill_registry.get_all_tool_schemas())
+            except ImportError:
+                pass
+
+            payload = {
+                "model": self.model_name,
+                "messages": work_history,
+                "stream": self.stream,
+                "tools": tools_schema,
+                "tool_choice": "auto"
+            }
+            if self.stream:
+                payload["stream_options"] = {"include_usage": True},
+        else:
+            # XML 模式（原有逻辑）
+            payload = {
+                "model": self.model_name,
+                "messages": work_history,
+                "stream": self.stream,
+                "stream_options": {"include_usage": True}
+            }
+            if self.stream:
+                payload["stream_options"] = {"include_usage": True},
+
+        rsp = requests.post(
+            self.api_provider,
+            headers={**self.headers,
+                     "Accept-Charset": "utf-8",
+                     "Accept": "text/event-stream"},
+            json=payload,
+            stream=self.stream
+        )
+        rsp.encoding = 'utf-8'
+
+        full_content = ""
+        self.stream_run = True
+        tool_calls_list = []  # 存储所有 tool_calls
+
+        if self.stream:
+            # 流式响应处理
+            for line in rsp.iter_lines(decode_unicode=True):
+                logger.trace(line)
+                if not line or not line.startswith('data: '):
+                    continue
+                data = line[6:]
+                if data == '[DONE]':
+                    break
+                try:
+                    chunk = json.loads(data)
+                except json.JSONDecodeError:
+                    continue
+
+                delta = (chunk.get('choices') or [{}])[0].get('delta') or {}
+
+                # Function Calling: 处理 tool_calls
+                if self.fc_model:
+                    tool_calls = delta.get('tool_calls')
+                    if tool_calls:
+                        for call in tool_calls:
+                            # 初始化或更新 tool_calls_list
+                            if call.get('index') is not None:
+                                idx = call['index']
+                                if idx >= len(tool_calls_list):
+                                    tool_calls_list.append({
+                                        'id': call.get('id'),
+                                        'function': {
+                                            'name': call['function']['name'],
+                                            'arguments': call['function'].get('arguments') or ''
+                                        }
+                                    })
+                                else:
+                                    if 'arguments' in call['function'] and call['function']['arguments'] is not None:
+                                        tool_calls_list[idx]['function']['arguments'] += call['function']['arguments']
+                            continue
+
+                # 普通 content
+                content = delta.get('content', '')
+                if content:
+                    full_content += content
+                    self.pack(content, finish_task=False)
+
+                usage = chunk.get('usage')
+                if usage:
+                    self.stream_run = False
+                    self.pack(finish_task=True)
+                    self.pack(f"\n本次请求用量：提示 {usage['prompt_tokens']} tokens，"
+                          f"生成 {usage['completion_tokens']} tokens，"
+                          f"总计 {usage['total_tokens']} tokens。", other=True)
+        else:
+            # 非流式响应处理
+            self.stream_run = False
+            try:
+                response_json = rsp.json()
+                logger.trace(response_json)
+                message = response_json['choices'][0]['message']
+                full_content = message.get('content', '')
+
+                # Function Calling: 处理 tool_calls
+                if self.fc_model:
+                    tool_calls_list = message.get('tool_calls', [])
+
+                if full_content:
+                    self.pack(full_content, finish_task=False)
+                usage = response_json.get('usage', {})
+                if usage:
+                    self.pack(finish_task=True)
+                    self.pack(f"\n本次请求用量：提示 {usage['prompt_tokens']} tokens，"
+                          f"生成 {usage['completion_tokens']} tokens，"
+                          f"总计 {usage['total_tokens']} tokens。", other=True)
+            except Exception as e:
+                logger.error(f"非流式响应处理错误: {e}")
+                full_content = rsp.text
+
+        logger.trace(f"AI 回复内容长度：{len(full_content)}")
+
+        # Function Calling: 执行工具调用
+        if self.fc_model and tool_calls_list:
+            logger.debug(f"发现 Function Calling 工具调用: {tool_calls_list}")
+
+            # 添加 assistant message with tool_calls
+            work_history.append({
+                "role": "assistant",
+                "content": None,
+                "tool_calls": tool_calls_list
+            })
+
+            tool_results = []
+            for tool_call in tool_calls_list:
+                tool_name = tool_call['function']['name']
+                tool_id = tool_call['id']
+
+                # 查找工具
+                tool_func = None
+                if tool_registry.check_permission(self.uuid, tool_name):
+                    tool_info = tool_registry.get_tool_info(tool_name)
+                    if tool_info is not None:
+                        tool_func = tool_info['function']
+
+                if tool_func is None and hasattr(self, tool_name):
+                    method = getattr(self, tool_name)
+                    if callable(method):
+                        tool_func = method
+
+                if tool_func is None:
+                    error_msg = f"找不到工具 '{tool_name}'"
+                    logger.warning(error_msg)
+                    tool_results.append({
+                        'tool_call_id': tool_id,
+                        'name': tool_name,
+                        'content': error_msg
+                    })
+                    continue
+
+                # 调用工具（解析 arguments 为 dict）
+                try:
+                    args = json.loads(tool_call['function']['arguments'])
+                    # 生成任务 ID
+                    self.current_task_id = self._generate_task_id()
+                    # 执行 before 钩子
+                    self._execute_hooks('before', tool_name, args)
+                    logger.debug(f"调用工具 {tool_name}，参数: {args}")
+                    self.pack(tool_name=tool_name, tool_parameter=args)
+                    result = tool_func(**args)
+
+                    # 执行 after 钩子
+                    self._execute_hooks('after', tool_name, args, result)
+                    # 打包工具返回值
+                    self.pack(tool_result=result, tool_name=tool_name)
+                    tool_results.append({
+                        'tool_call_id': tool_id,
+                        'name': tool_name,
+                        'content': result
+                    })
+                except Exception as e:
+                    error_msg = f"执行工具 {tool_name} 时出错: {str(e)}"
+                    logger.error(error_msg)
+
+                    # 执行 error 钩子（如果 args 已定义）
+                    if 'args' in locals():
+                        self._execute_hooks('error', tool_name, args, error_msg)
+                    tool_results.append({
+                        'tool_call_id': tool_id,
+                        'name': tool_name,
+                        'content': error_msg
+                    })
+
+            # 添加 tool responses 到历史
+            for result in tool_results:
+                work_history.append({
+                    "role": "tool",
+                    "tool_call_id": result['tool_call_id'],
+                    "name": result['name'],
+                    "content": result['content']
+                })
+
+            # 继续对话
+            logger.debug("工具执行完成，继续对话")
+            return self.conversation_with_tool(tool=True)
+
+        # XML 模式：提取并执行工具
+        xml_pattern = re.compile(r'<(\w+)>.*?</\1>', flags=re.S)
+        clean_pattern = re.compile(r'</?(out_text|thinking)>', flags=re.S)
+        clean_content = clean_pattern.sub('', full_content)
+        xml_blocks = [m.group(0) for m in xml_pattern.finditer(clean_content)]
+
+        tool_results = []
+        tool_names = []
+        for block in xml_blocks:
+            logger.debug(f"发现工具块：{block[:50]}...")
+            soup = BeautifulSoup(block, "xml")
+            root = soup.find()
+            if root is None:
+                raise ValueError("空 XML")
+            tool_name = root.name
+
+            # 优先级查找工具：1. 工具注册器 2. 类方法 3. 返回无工具
+            tool_func = None
+            tool_source = None
+
+            # 优先级1: 在工具注册器中查找
+            if tool_registry.check_permission(self.uuid, tool_name):
+                tool_info = tool_registry.get_tool_info(tool_name)
+                if tool_info is not None:
+                    tool_func = tool_info['function']
+                    tool_source = "工具注册器"
+
+            # 优先级2: 在类方法中查找
+            if tool_func is None and hasattr(self, tool_name):
+                method = getattr(self, tool_name)
+                if callable(method):
+                    tool_func = method
+                    tool_source = "类方法"
+
+            # 优先级3: 都没有找到
+            if tool_func is None:
+                available_tools = self.get_all_available_tools()
+                tool_error = f"工具错误：找不到工具 '{tool_name}'。"
+                if available_tools:
+                    tool_error += f" 你可以使用以下工具：{', '.join(available_tools)}"
+
+                work_history.append({"role": "system", "content": tool_error})
+                tool_results.append({"error": tool_error})
+                logger.warning(f"工具 {tool_name} 未找到，可用工具: {available_tools}")
+                continue
+
+            logger.debug(f"从 {tool_source} 找到工具 {tool_name}")
+            print(block)
+
+            # 解析 XML 参数，与 Function Calling 模式统一
+            params = {}
+            for child in root.children:
+                if hasattr(child, 'name') and child.name:
+                    params[child.name] = child.text
+
+            # 生成任务 ID
+            self.current_task_id = self._generate_task_id()
+
+            # 执行 before 钩子
+            self._execute_hooks('before', tool_name, params)
+
+            self.pack(tool_name=tool_name, tool_parameter=params)
+
+            # 执行工具：根据函数签名决定传递 dict 还是解包参数
+            import inspect
+            sig = inspect.signature(tool_func)
+            param_count = len([p for p in sig.parameters.values()
+                             if p.default == inspect.Parameter.empty and p.kind not in (inspect.Parameter.VAR_POSITIONAL, inspect.Parameter.VAR_KEYWORD)])
+
+            # 如果函数接受 **kwargs 或单个参数，传递整个 dict
+            # 否则解包参数
+            has_kwargs = any(p.kind == inspect.Parameter.VAR_KEYWORD for p in sig.parameters.values())
+            if has_kwargs or param_count == 0:
+                result = tool_func(**params) if params else tool_func()
+            elif param_count == 1 and len(params) == 1:
+                # 单个参数函数，传递单个值
+                result = tool_func(list(params.values())[0])
+            else:
+                # 多参数函数，尝试解包
+                try:
+                    result = tool_func(**params)
+                except TypeError:
+                    # 如果解包失败，回退到传递整个 XML 块（向后兼容）
+                    result = tool_func(block)
+
+            # 执行 after 钩子（传入结果）
+            self._execute_hooks('after', tool_name, params, result)
+
+            # 打包工具返回值
+            self.pack(tool_result=result, tool_name=tool_name)
+
+            if not result:
+                result = f"no return for the tool{tool_name}"
+
+            tool_results.append(result)
+            tool_names.append(tool_name)
+
+        #如配置错误强制跳出避免堵塞
+        if "<attempt_completion>" in full_content:
+            self.pack("\n[系统] AI 已标记任务完成，程序退出。", tool_name="attempt_completion")
+            # sys.exit(0)
+
+        # 3. 若工具产生结果，继续对话
+        if tool_results:
+            logger.debug(f"工具执行成功，结果数：{len(tool_results)}")
+            n = 0
+            for i in tool_results:
+                try:
+                    work_history.append({"role": "system", "content": f"{tool_names[n]} results: {i}"})
+                    n += 1
+                except Exception:
+                    break
+            logger.debug(f"对话历史长度：{len(self.history)}")
+            return self.conversation_with_tool(tool=True)
+        if tool:
+            logger.debug(f"返回对话历史最后一条，长度：{len(work_history[-1].get('content')) if work_history[-1].get('content') else 0}")
+            return work_history[-1].get("content")
+        return full_content
+
+    def pack(self, message=None, tool_model=False, tool_name=None, tool_parameter=None,
+             finish_task=False, other=False, tool_result=None):
+        """
+        打包输出内容，包含 AI UUID、任务戳和时间戳
+
+        参数:
+            message: 普通消息内容
+            tool_model: 是否是工具模型调用
+            tool_name: 工具名称
+            tool_parameter: 工具参数
+            finish_task: 是否任务完成
+            other: 其他标记
+            tool_result: 工具执行结果
+        """
+        content = {}
+        task_id = self.current_task_id or self._generate_task_id()
+        timestamp = self._get_timestamp()
+
+        if finish_task:
+            content = {
+                "task": True,
+                "task_id": task_id,
+                "timestamp": timestamp,
+                "ai_uuid": self.uuid,
+                "ai_name": self.name
+            }
+        elif tool_model:
+            content = {
+                "tool_name": tool_name,
+                "tool_parameter": tool_parameter,
+                "ai_uuid": self.uuid,
+                "ai_name": self.name,
+                "task_id": task_id,
+                "timestamp": timestamp,
+                "task": False
+            }
+        elif tool_result is not None:
+            # 工具返回值
+            content = {
+                "tool_result": tool_result,
+                "tool_name": tool_name,
+                "ai_uuid": self.uuid,
+                "ai_name": self.name,
+                "task_id": task_id,
+                "timestamp": timestamp,
+                "task": False
+            }
+        else:
+            content = {
+                "message": message,
+                "ai_uuid": self.uuid,
+                "ai_name": self.name,
+                "other": other,
+                "task_id": task_id,
+                "timestamp": timestamp,
+                "task": False
+            }
+        self.out(content)
+
+    def out(self, content):
+        if content.get("tool_name"):
+            print("调用工具:", content.get("tool_name"), "参数", content.get("tool_parameter"))
+            return
+        if not content.get("task"):
+            message = content.get("message")
+            if message is not None:
+                print(message, end="")
+        else:
+            print()
+
+
+    # ---------------- 内置工具（@builtin_tool 装饰器自动从签名推导 schema）----------------
+
+    @builtin_tool(
+        description="请求其他Agent帮助",
+        params={
+            "agent_id": "目标Agent的UUID或名称",
+            "message": "请求内容",
+        },
+    )
+    def ask_for_help(self, agent_id: str, message: str) -> str:
+        """请求另一个 Agent 协助完成子任务，并把对方的回复作为工具返回值返回。"""
+        try:
+            from Dumplings import agent_list
+            target = agent_list.get(agent_id)
+            if target is None:
+                return f"未找到 Agent：{agent_id}"
+            reply = target.conversation_with_tool(message)
+            return str(reply)
+        except Exception as e:
+            logger.error(f"ask_for_help 失败：{e}")
+            return f"协助请求失败：{e}"
+
+    @builtin_tool(
+        description="列出所有可用的Agent及其UUID和名称",
+    )
+    def list_agents(self) -> str:
+        """返回当前系统中所有已注册 Agent 的清单，便于发现协作对象。"""
+        from Dumplings import agent_list
+
+        unique_agents: dict = {}
+        for key, agent in agent_list.items():
+            uuid = getattr(agent, "uuid", None)
+            name = getattr(agent, "name", None)
+            desc = getattr(agent, "description", None)
+            if uuid and name and uuid not in unique_agents:
+                unique_agents[uuid] = {"uuid": uuid, "name": name, "description": desc}
+
+        if not unique_agents:
+            return "可用的Agent列表：（暂无）"
+
+        lines = []
+        for info in unique_agents.values():
+            lines.append(
+                f"- {info['name']} (UUID: {info['uuid']}) description: {info['description']}"
+            )
+        return "可用的Agent列表：" + "".join(lines)
+
+    @builtin_tool(
+        description="标记任务完成并退出",
+        params={"report_content": "汇报内容（可空）"},
+    )
+    def attempt_completion(self, report_content: str = "") -> str:
+        """告知框架当前任务已结束，传入汇报内容；该工具返回值即为整个任务的对外输出。"""
+        return report_content if report_content else "任务已完成"
+
+    @builtin_tool(
+        description="重新加载你自己：重新拉取当前可用的工具列表和Skills，重置系统提示词。",
+    )
+    def reload(self) -> str:
+        """重新构建内部状态（系统提示词 / 工具列表 / Skills），保留对话历史。"""
+        self.__init__(new_load=False)
+        return "successfully reloaded"
+
+    def get_all_available_tools(self) -> list:
+        tools: list[Any] = []
+        tools_info = tool_registry.get_all_tools_info(self.uuid)
+        if tools_info:
+            tools.extend(tools_info.keys())
+        for s in tool_registry.collect_builtin_tools(self):
+            tools.append(s["function"]["name"])
+        return tools
+

dumplingsai-0.1.1/Agent_list.py

@@ -0,0 +1,37 @@
+# 1. 准备一个“双键”容器
+agent_list = {}          # {key1: cls, key2: cls}
+
+def register_agent(uuid, name, description=None):
+    """
+    类装饰器：把被装饰的类同时注册到两个字典键下。
+    两个键指向**同一个类对象**，因此内存中只有一份。
+    """
+    def _decorator(cls):
+        cls.uuid = uuid
+        cls.name = name
+        cls.description = description
+        cls_instantiation = cls()
+        agent_list[uuid] = cls_instantiation
+        agent_list[name] = cls_instantiation
+        return cls              # 原样返回，不影响类本身
+    return _decorator
+
+
+if __name__ == '__main__':
+    # 2. 用法示例
+    @register_agent('foo', 'bar')
+    class Demo:
+        def __init__(self, name):
+            self.name = name
+
+        def greet(self):
+            return f'Hello {self.name}'
+
+    # 3. 通过任意键都能拿到类
+    Cls1 = agent_list['foo']   # 拿到 Demo
+    Cls2 = agent_list['bar']   # 拿到同一个 Demo
+    assert Cls1 is Cls2              # True
+
+    # 4. 像正常类一样用
+    obj = Cls1('world')
+    print(obj.greet())               # Hello world