{"repo": "chasen-liao/lanobot", "files": {"config.json": {"5ab2c91813e035a444f9a51d9778402f9d6ed946": {"t": 1785746769.0461369, "n": 401, "h": "56e38729bd56d0bc", "c": "{\n  \"llm\": {\n    \"provider\": \"siliconflow\",\n    \"model\": \"deepseek-ai/DeepSeek-V3.2\",\n    \"api_key\": \"sk-wlyxwslviscdgdgsfqelmsyvhjbsjloxdvvoritvjnremboi\",\n    \"base_url\": \"https://api.siliconflow.cn/v1\",\n    \"temperature\": 0.1,\n    \"max_tokens\": 4096\n  },\n  \"channels\": {\n    \"feishu_enabled\": false,\n    \"feishu_app_id\": \"\",\n    \"feishu_app_secret\": \"\"\n  },\n  \"logging\": {\n    \"level\": \"INFO\"\n  }\n}\n"}, "5256bc5495634f4de1a9f0cdb886ba412b5012b8": {"t": 1785767628.2883065, "n": 401, "h": "56e38729bd56d0bc", "c": "{\n  \"llm\": {\n    \"provider\": \"siliconflow\",\n    \"model\": \"deepseek-ai/DeepSeek-V3.2\",\n    \"api_key\": \"sk-wlyxwslviscdgdgsfqelmsyvhjbsjloxdvvoritvjnremboi\",\n    \"base_url\": \"https://api.siliconflow.cn/v1\",\n    \"temperature\": 0.1,\n    \"max_tokens\": 4096\n  },\n  \"channels\": {\n    \"feishu_enabled\": false,\n    \"feishu_app_id\": \"\",\n    \"feishu_app_secret\": \"\"\n  },\n  \"logging\": {\n    \"level\": \"INFO\"\n  }\n}\n"}, "9fd5b3cf984fd747e434f2e271b666e5a8476cb6": {"t": 1785767629.5318918, "n": 405, "h": "6ce613680ed1d585", "c": "{\n  \"llm\": {\n    \"provider\": \"siliconflow\",\n    \"model\": \"Pro/deepseek-ai/DeepSeek-V3.2\",\n    \"api_key\": \"sk-wlyxwslviscdgdgsfqelmsyvhjbsjloxdvvoritvjnremboi\",\n    \"base_url\": \"https://api.siliconflow.cn/v1\",\n    \"temperature\": 0.1,\n    \"max_tokens\": 4096\n  },\n  \"channels\": {\n    \"feishu_enabled\": false,\n    \"feishu_app_id\": \"\",\n    \"feishu_app_secret\": \"\"\n  },\n  \"logging\": {\n    \"level\": \"INFO\"\n  }\n}\n"}}, "framework/README.md": {"5ab2c91813e035a444f9a51d9778402f9d6ed946": {"t": 1785747232.4867244, "n": 11902, "h": "84b0062b2a605be4", "c": "# Lanobot 架构设计文档\n\n## 1. 整体架构\n\n```mermaid\nflowchart TB\n    subgraph Channels[\"渠道层 (channels/)\"]\n        TG[Telegram]\n        SL[Slack]\n        DD[钉钉]\n        FS[飞书]\n        DC[Discord]\n        WA[WhatsApp]\n        QQ[QQ]\n        EM[Email]\n    end\n\n    subgraph Bus[\"消息总线 (bus/)\"]\n        MB[MessageBus]\n        BE[事件定义]\n    end\n\n    subgraph Core[\"核心Agent层 (agent/)\"]\n        AG[AgentGraph<br/>LangGraph StateGraph]\n        SN[State Nodes]\n        ED[Edges]\n    end\n\n    subgraph Tools[\"工具层 (tools/)\"]\n        FS[文件系统]\n        SH[Shell执行]\n        WB[Web搜索/抓取]\n        MS[Message发送]\n        SP[Spawn子代理]\n        CR[Cron定时]\n        MC[MCP工具]\n    end\n\n    subgraph Memory[\"记忆层 (memory/)\"]\n        CP[Checkpointer<br/>短期记忆]\n        LS[Long-term Store<br/>长期记忆]\n        HM[History Manager]\n    end\n\n    subgraph Providers[\"LLM提供商 (providers/)\"]\n        OP[OpenAI]\n        AN[Anthropic]\n        AZ[Azure OpenAI]\n        DS[DeepSeek]\n        OR[OpenRouter]\n        LM[本地模型<br/>Ollama/vLLM]\n    end\n\n    subgraph Config[\"配置层 (config/)\"]\n        SC[Schema<br/>Pydantic]\n        LD[Loader]\n    end\n\n    subgraph Session[\"会话管理 (session/)\"]\n        SM[SessionManager]\n        SH[Session历史]\n    end\n\n    Channels --> MB\n    MB --> AG\n    AG --> Tools\n    AG --> AN\n    Tools -->|结果| AG\n    AG <--> CP\n    AG <--> LS\n    AG <--> HM\n    AG <--> MB\n    MB --> Channels\n```\n\n## 2. 模块职责\n\n### 2.1 渠道层 (channels/)\n\n**职责**：对接各种IM/通讯平台的API，统一消息格式\n\n**保留原因**：与业务无关，纯粹适配外部接口\n\n```python\n# channels/base.py\nclass ChannelAdapter(ABC):\n    \"\"\"渠道适配器基类\"\"\"\n\n    @abstractmethod\n    async def start(self):\n        \"\"\"启动渠道监听\"\"\"\n\n    @abstractmethod\n    async def send(self, chat_id: str, content: str, metadata: dict = None):\n        \"\"\"发送消息\"\"\"\n\n    @abstractmethod\n    async def stop(self):\n        \"\"\"停止渠道\"\"\"\n```\n\n**支持的渠道**：\n- Telegram, Slack, Discord\n- 钉钉, 飞书, 企业微信\n- WhatsApp, QQ, Email, Matrix\n\n### 2.2 消息总线 (bus/)\n\n**职责**：异步消息队列，解耦各模块\n\n**保留原因**：架构清晰，状态无关\n\n```python\n# bus/queue.py\nclass MessageBus:\n    \"\"\"Async message queue for decoupled channel-agent communication\"\"\"\n\n    def __init__(self):\n        self.inbound: asyncio.Queue[InboundMessage] = asyncio.Queue()\n        self.outbound: asyncio.Queue[OutboundMessage] = asyncio.Queue()\n\n    async def publish_inbound(self, msg: InboundMessage):\n        await self.inbound.put(msg)\n\n    async def consume_inbound(self) -> InboundMessage:\n        return await self.inbound.get()\n\n    async def publish_outbound(self, msg: OutboundMessage):\n        await self.outbound.put(msg)\n```\n\n### 2.3 Agent核心层 (agent/) - LangChain 1.x重构\n\n**职责**：Agent推理循环，工具调用协调\n\n**重构方案**：使用LangChain 1.x的 `create_agent` API（基于LangGraph）\n\n> ⚠️ **重要更新**: LangGraph v1 废弃了 `create_react_agent`，推荐使用 LangChain 1.x 的 `create_agent`\n\n```python\n# agent/graph.py\nfrom langchain.agents import create_agent\nfrom langgraph.checkpoint.memory import InMemorySaver\n\nclass AgentGraph:\n    \"\"\"LangChain 1.x Agent构建器\"\"\"\n\n    def __init__(self, model, tools: list, checkpointer=None, middleware=None):\n        self.graph = create_agent(\n            model,                      # 模型实例或模型名\n            tools,                      # 工具列表\n            system_prompt=\"你是lanobot，一个超轻量级AI助手。\",  # 注意：不再是prompt\n            checkpointer=checkpointer,\n            middleware=middleware,      # 可选：中间件列表\n        )\n\n    async def invoke(self, message: str, thread_id: str):\n        config = {\"configurable\": {\"thread_id\": thread_id}}\n        return await self.graph.ainvoke(\n            {\"messages\": [{\"role\": \"user\", \"content\": message}]},\n            config=config\n        )\n```\n\n#### Middleware系统\n\n```python\n# agent/middleware.py\nfrom langchain.agents.middleware import HumanInTheLoopMiddleware\n\n# 人机协作中间件 - 敏感操作需要审批\nhuman_middleware = HumanInTheLoopMiddleware(\n    interrupt_on={\n        \"send_message\": True,    # 发送消息需要审批\n        \"exec_command\": True,    # 执行命令需要审批\n        \"write_file\": True,      # 写文件需要审批\n    }\n)\n\n# 创建带中间件的Agent\nagent = create_agent(\n    model,\n    tools,\n    system_prompt=\"...\",\n    middleware=[human_middleware],\n    checkpointer=InMemorySaver(),\n)\n```\n\n### 2.4 工具层 (tools/) - LangChain化\n\n**职责**：Agent可调用的外部能力\n\n**重构方案**：使用`@tool`装饰器或`StructuredTool`\n\n```python\n# tools/filesystem.py\nfrom langchain_core.tools import tool, StructuredTool\nfrom pydantic import BaseModel, Field\n\n# 方式1：使用@tool装饰器（简单场景）\n@tool\ndef read_file(path: str) -> str:\n    \"\"\"读取文件内容\n\n    Args:\n        path: 文件路径（绝对路径或相对于workspace的路径）\n\n    Returns:\n        文件内容字符串\n    \"\"\"\n    # 实现逻辑\n    pass\n\n@tool\ndef write_file(path: str, content: str) -> str:\n    \"\"\"写入文件内容\n\n    Args:\n        path: 文件路径\n        content: 要写入的内容\n    \"\"\"\n    pass\n\n@tool\ndef list_directory(path: str = \".\") -> str:\n    \"\"\"列出目录内容\"\"\"\n    pass\n\n@tool\ndef edit_file(path: str, old: str, new: str) -> str:\n    \"\"\"编辑文件内容（替换指定文本）\"\"\"\n    pass\n```\n\n```python\n# tools/shell.py\n@tool\ndef exec_command(command: str, timeout: int = 60) -> str:\n    \"\"\"执行Shell命令\n\n    Args:\n        command: 要执行的命令\n        timeout: 超时时间（秒），默认60秒\n    \"\"\"\n    pass\n```\n\n```python\n# tools/web.py\n@tool\ndef web_search(query: str, max_results: int = 5) -> str:\n    \"\"\"Web搜索\"\"\"\n    pass\n\n@tool\ndef web_fetch(url: str) -> str:\n    \"\"\"抓取网页内容\"\"\"\n    pass\n\n@tool\ndef message_send(channel: str, chat_id: str, content: str) -> str:\n    \"\"\"发送消息到指定渠道\"\"\"\n    pass\n\n@tool\ndef spawn_subagent(task: str, model: str = None) -> str:\n    \"\"\"Spawn a sub-agent to handle subtasks\"\"\"\n    pass\n```\n\n### 2.5 记忆层 (memory/) - LangChain化\n\n**职责**：短期会话记忆 + 长期记忆\n\n**重构方案**：LangGraph Checkpointer + LangChain Store\n\n> ⚠️ **注意**: LangChain 1.x 使用 `InMemorySaver` 替代了 `MemorySaver`\n\n```python\n# memory/checkpointer.py\nfrom langgraph.checkpoint.memory import InMemorySaver\n\n# 开发环境\ncheckpointer = InMemorySaver()\n\n# 生产环境\n# from langgraph.checkpoint.postgres import PostgresSaver\n# checkpointer = PostgresSaver.from_conn_string(conn_string)\n```\n\n```python\n# memory/store.py\nfrom langchain.storage import InMemoryStore\nfrom langgraph.store.postgres import PostgresStore\n\nclass MemoryStore:\n    \"\"\"长期记忆存储\"\"\"\n\n    def __init__(self, use_postgres: bool = False):\n        if use_postgres:\n            self.store = PostgresStore.from_conn_string(conn_string)\n        else:\n            self.store = InMemoryStore()\n\n    async def search_memory(self, query: str, k: int = 5) -> list:\n        \"\"\"语义检索相关记忆\"\"\"\n        pass\n\n    async def save_memory(self, key: str, value: str, metadata: dict = None):\n        \"\"\"保存记忆\"\"\"\n        pass\n```\n\n### 2.6 Providers层 (providers/)\n\n**职责**：LLM接口适配\n\n**保留原因**：对接各种API，但包装为LangChain接口\n\n```python\n# providers/langchain_wrapper.py\nfrom langchain_anthropic import ChatAnthropic\nfrom langchain_openai import ChatOpenAI\nfrom langchain_community.chat_models import ChatLiteLLM\n\ndef create_llm(provider: str, model: str, api_key: str = None, api_base: str = None):\n    \"\"\"创建LangChain LLM实例\"\"\"\n\n    if provider == \"anthropic\":\n        return ChatAnthropic(\n            model=model,\n            api_key=api_key,\n            base_url=api_base\n        )\n    elif provider == \"openai\":\n        return ChatOpenAI(\n            model=model,\n            api_key=api_key,\n            base_url=api_base\n        )\n    # LiteLLM支持所有OpenAI兼容的API\n    elif provider == \"litellm\":\n        return ChatLiteLLM(\n            model=model,\n            api_key=api_key,\n            base_url=api_base\n        )\n```\n\n### 2.7 配置层 (config/)\n\n**职责**：YAML/Env配置解析\n\n**保留**：Pydantic配置系统已经很完善\n\n```python\n# config/schema.py\nclass Config(BaseSettings):\n    agents: AgentsConfig\n    channels: ChannelsConfig\n    providers: ProvidersConfig\n    gateway: GatewayConfig\n    tools: ToolsConfig\n```\n\n### 2.8 会话管理 (session/)\n\n**职责**：会话历史持久化\n\n**保留**：Session管理逻辑与具体Agent实现解耦\n\n```python\n# session/manager.py\nclass SessionManager:\n    \"\"\"会话管理器\"\"\"\n\n    def __init__(self, workspace: Path):\n        self.sessions_dir = workspace / \"sessions\"\n\n    def get_or_create(self, session_key: str) -> Session:\n        \"\"\"获取或创建会话\"\"\"\n\n    def save(self, session: Session):\n        \"\"\"保存会话\"\"\"\n\n    def get_history(self, session_key: str, max_messages: int = None) -> list:\n        \"\"\"获取会话历史\"\"\"\n```\n\n## 3. 数据流\n\n### 3.1 消息处理流程\n\n```mermaid\nsequenceDiagram\n    participant C as Channel\n    participant B as MessageBus\n    participant G as AgentGraph\n    participant T as Tools\n    participant P as Provider\n    participant M as Memory\n    participant O as Outbound\n\n    C->>B: publish_inbound(msg)\n    B->>G: 消费消息\n    G->>P: LLM调用(带工具定义)\n    P-->>G: LLM响应(含tool_calls)\n    G->>T: 执行工具\n    T-->>G: 工具结果\n    G->>M: 保存checkpointer状态\n    G->>P: 继续LLM调用\n    P-->>G: 最终响应\n    G->>B: publish_outbound(response)\n    B->>O: 推送响应\n    O->>C: 发送消息\n```\n\n### 3.2 记忆系统流程\n\n```mermaid\nflowchart LR\n    A[用户消息] --> B{检查token限制}\n    B -->|超限| C[consolidate_messages]\n    B -->|正常| D[直接处理]\n    C --> E[LLM摘要]\n    E --> F[保存到MEMORY.md]\n    F --> D\n    D --> G[返回响应]\n```\n\n## 4. 目录结构\n\n```\nlanobot/\n├── main.py                    # 入口点\n├── pyproject.toml             # 项目配置\n├── CLAUDE.md                  # Claude Code指导\n│\n├── config/                    # 配置层\n│   ├── schema.py              # Pydantic配置模型\n│   ├── loader.py              # 配置加载\n│   └── paths.py               # 路径工具\n│\n├── bus/                       # 消息总线\n│   ├── queue.py               # MessageBus实现\n│   └── events.py              # 事件定义\n│\n├── channels/                  # 渠道层（保留）\n│   ├── base.py                # 渠道基类\n│   ├── manager.py             # 渠道管理\n│   ├── telegram.py\n│   ├── slack.py\n│   ├── feishu.py\n│   ├── dingtalk.py\n│   ├── discord.py\n│   ├── whatsapp.py\n│   ├── qq.py\n│   ├── wecom.py\n│   ├── email.py\n│   └── matrix.py\n│\n├── providers/                 # LLM提供商\n│   ├── base.py                # 基础接口（可保留）\n│   ├── registry.py            # 提供商注册\n│   ├── langchain_wrapper.py   # LangChain包装器\n│   ├── openai_provider.py\n│   ├── anthropic_provider.py\n│   ├── azure_openai_provider.py\n│   ├── litellm_provider.py\n│   └── custom_provider.py\n│\n├── agent/                     # Agent核心层（LangChain化）\n│   ├── __init__.py\n│   ├── state.py               # AgentState定义\n│   ├── graph.py               # LangGraph构建\n│   ├── nodes.py               # 节点函数\n│   ├── edges.py               # 边定义\n│   └── middleware.py          # 中间件\n│\n├── tools/                     # 工具层（LangChain化）\n│   ├── __init__.py\n│   ├── registry.py            # 工具注册\n│   ├── filesystem.py          # 文件操作\n│   ├── shell.py               # Shell执行\n│   ├── web.py                 # Web工具\n│   ├── message.py             # 消息发送\n│   ├── spawn.py               # 子代理\n│   ├── cron.py                # 定时任务\n│   └── mcp.py                 # MCP集成\n│\n├── memory/                    # 记忆层（LangChain化）\n│   ├── checkpointer.py        # LangGraph Checkpointer\n│   ├── store.py               # 长期记忆\n│   └── history.py             # 历史记录管理\n│\n├── session/                   # 会话管理\n│   ├── __init__.py\n│   ├── manager.py             # 会话管理器\n│   └── types.py               # 会话类型\n│\n├── cron/                      # 定时任务\n│   ├── service.py\n│   └── types.py\n│\n├── heartbeat/                 # 心跳服务\n│   └── service.py\n│\n├── cli/                       # CLI命令\n│   ├── commands.py\n│   └── __init__.py\n│\n└── utils/                     # 工具函数\n    └── helpers.py\n```\n\n## 5. 重构计划\n\n### Phase 1: 基础架构\n- [ ] 创建agent/state.py - AgentState定义\n- [ ] 创建agent/graph.py - LangGraph构建\n- [ ] 创建providers/langchain_wrapper.py - LLM包装\n- [ ] 集成现有MessageBus\n\n### Phase 2: 工具迁移\n- [ ] 将tools/filesystem.py转为@tool\n- [ ] 将tools/shell.py转为@tool\n- [ ] 将tools/web.py转为@tool\n- [ ] 集成MCP工具\n\n### Phase 3: 记忆系统\n- [ ] 集成LangGraph Checkpointer\n- [ ] 实现长期记忆Store\n- [ ] 迁移会话历史\n\n### Phase 4: 渠道适配\n- [ ] 适配现有渠道到新Agent接口\n- [ ] 测试各渠道连通性\n\n### Phase 5: CLI和配置\n- [ ] 更新CLI命令\n- [ ] 验证配置系统\n\n## 6. 关键设计决策\n\n### 6.1 为什么用混合架构?\n- **保留**：渠道、配置、业务无关的模块已经稳定\n- **重构**：Agent核心需要LangChain的标准化和生态\n\n### 6.2 LangGraph vs 纯LangChain\n- 使用`create_agent`快速构建\n- 后续可扩展为自定义StateGraph\n\n### 6.3 记忆策略\n- **短期**：LangGraph Checkpointer (thread_id级别)\n- **长期**：LangChain Store (跨会话)\n- **历史**：文件存储 (可grep搜索)\n\n### 6.4 Provider策略\n- 保留原有provider注册机制\n- 新增LangChain包装层\n- 渐进式迁移"}, "771d048f6878219b0a134244862568223f15fd53": {"t": 1785767627.4940884, "n": 11902, "h": "84b0062b2a605be4", "c": "# Lanobot 架构设计文档\n\n## 1. 整体架构\n\n```mermaid\nflowchart TB\n    subgraph Channels[\"渠道层 (channels/)\"]\n        TG[Telegram]\n        SL[Slack]\n        DD[钉钉]\n        FS[飞书]\n        DC[Discord]\n        WA[WhatsApp]\n        QQ[QQ]\n        EM[Email]\n    end\n\n    subgraph Bus[\"消息总线 (bus/)\"]\n        MB[MessageBus]\n        BE[事件定义]\n    end\n\n    subgraph Core[\"核心Agent层 (agent/)\"]\n        AG[AgentGraph<br/>LangGraph StateGraph]\n        SN[State Nodes]\n        ED[Edges]\n    end\n\n    subgraph Tools[\"工具层 (tools/)\"]\n        FS[文件系统]\n        SH[Shell执行]\n        WB[Web搜索/抓取]\n        MS[Message发送]\n        SP[Spawn子代理]\n        CR[Cron定时]\n        MC[MCP工具]\n    end\n\n    subgraph Memory[\"记忆层 (memory/)\"]\n        CP[Checkpointer<br/>短期记忆]\n        LS[Long-term Store<br/>长期记忆]\n        HM[History Manager]\n    end\n\n    subgraph Providers[\"LLM提供商 (providers/)\"]\n        OP[OpenAI]\n        AN[Anthropic]\n        AZ[Azure OpenAI]\n        DS[DeepSeek]\n        OR[OpenRouter]\n        LM[本地模型<br/>Ollama/vLLM]\n    end\n\n    subgraph Config[\"配置层 (config/)\"]\n        SC[Schema<br/>Pydantic]\n        LD[Loader]\n    end\n\n    subgraph Session[\"会话管理 (session/)\"]\n        SM[SessionManager]\n        SH[Session历史]\n    end\n\n    Channels --> MB\n    MB --> AG\n    AG --> Tools\n    AG --> AN\n    Tools -->|结果| AG\n    AG <--> CP\n    AG <--> LS\n    AG <--> HM\n    AG <--> MB\n    MB --> Channels\n```\n\n## 2. 模块职责\n\n### 2.1 渠道层 (channels/)\n\n**职责**：对接各种IM/通讯平台的API，统一消息格式\n\n**保留原因**：与业务无关，纯粹适配外部接口\n\n```python\n# channels/base.py\nclass ChannelAdapter(ABC):\n    \"\"\"渠道适配器基类\"\"\"\n\n    @abstractmethod\n    async def start(self):\n        \"\"\"启动渠道监听\"\"\"\n\n    @abstractmethod\n    async def send(self, chat_id: str, content: str, metadata: dict = None):\n        \"\"\"发送消息\"\"\"\n\n    @abstractmethod\n    async def stop(self):\n        \"\"\"停止渠道\"\"\"\n```\n\n**支持的渠道**：\n- Telegram, Slack, Discord\n- 钉钉, 飞书, 企业微信\n- WhatsApp, QQ, Email, Matrix\n\n### 2.2 消息总线 (bus/)\n\n**职责**：异步消息队列，解耦各模块\n\n**保留原因**：架构清晰，状态无关\n\n```python\n# bus/queue.py\nclass MessageBus:\n    \"\"\"Async message queue for decoupled channel-agent communication\"\"\"\n\n    def __init__(self):\n        self.inbound: asyncio.Queue[InboundMessage] = asyncio.Queue()\n        self.outbound: asyncio.Queue[OutboundMessage] = asyncio.Queue()\n\n    async def publish_inbound(self, msg: InboundMessage):\n        await self.inbound.put(msg)\n\n    async def consume_inbound(self) -> InboundMessage:\n        return await self.inbound.get()\n\n    async def publish_outbound(self, msg: OutboundMessage):\n        await self.outbound.put(msg)\n```\n\n### 2.3 Agent核心层 (agent/) - LangChain 1.x重构\n\n**职责**：Agent推理循环，工具调用协调\n\n**重构方案**：使用LangChain 1.x的 `create_agent` API（基于LangGraph）\n\n> ⚠️ **重要更新**: LangGraph v1 废弃了 `create_react_agent`，推荐使用 LangChain 1.x 的 `create_agent`\n\n```python\n# agent/graph.py\nfrom langchain.agents import create_agent\nfrom langgraph.checkpoint.memory import InMemorySaver\n\nclass AgentGraph:\n    \"\"\"LangChain 1.x Agent构建器\"\"\"\n\n    def __init__(self, model, tools: list, checkpointer=None, middleware=None):\n        self.graph = create_agent(\n            model,                      # 模型实例或模型名\n            tools,                      # 工具列表\n            system_prompt=\"你是lanobot，一个超轻量级AI助手。\",  # 注意：不再是prompt\n            checkpointer=checkpointer,\n            middleware=middleware,      # 可选：中间件列表\n        )\n\n    async def invoke(self, message: str, thread_id: str):\n        config = {\"configurable\": {\"thread_id\": thread_id}}\n        return await self.graph.ainvoke(\n            {\"messages\": [{\"role\": \"user\", \"content\": message}]},\n            config=config\n        )\n```\n\n#### Middleware系统\n\n```python\n# agent/middleware.py\nfrom langchain.agents.middleware import HumanInTheLoopMiddleware\n\n# 人机协作中间件 - 敏感操作需要审批\nhuman_middleware = HumanInTheLoopMiddleware(\n    interrupt_on={\n        \"send_message\": True,    # 发送消息需要审批\n        \"exec_command\": True,    # 执行命令需要审批\n        \"write_file\": True,      # 写文件需要审批\n    }\n)\n\n# 创建带中间件的Agent\nagent = create_agent(\n    model,\n    tools,\n    system_prompt=\"...\",\n    middleware=[human_middleware],\n    checkpointer=InMemorySaver(),\n)\n```\n\n### 2.4 工具层 (tools/) - LangChain化\n\n**职责**：Agent可调用的外部能力\n\n**重构方案**：使用`@tool`装饰器或`StructuredTool`\n\n```python\n# tools/filesystem.py\nfrom langchain_core.tools import tool, StructuredTool\nfrom pydantic import BaseModel, Field\n\n# 方式1：使用@tool装饰器（简单场景）\n@tool\ndef read_file(path: str) -> str:\n    \"\"\"读取文件内容\n\n    Args:\n        path: 文件路径（绝对路径或相对于workspace的路径）\n\n    Returns:\n        文件内容字符串\n    \"\"\"\n    # 实现逻辑\n    pass\n\n@tool\ndef write_file(path: str, content: str) -> str:\n    \"\"\"写入文件内容\n\n    Args:\n        path: 文件路径\n        content: 要写入的内容\n    \"\"\"\n    pass\n\n@tool\ndef list_directory(path: str = \".\") -> str:\n    \"\"\"列出目录内容\"\"\"\n    pass\n\n@tool\ndef edit_file(path: str, old: str, new: str) -> str:\n    \"\"\"编辑文件内容（替换指定文本）\"\"\"\n    pass\n```\n\n```python\n# tools/shell.py\n@tool\ndef exec_command(command: str, timeout: int = 60) -> str:\n    \"\"\"执行Shell命令\n\n    Args:\n        command: 要执行的命令\n        timeout: 超时时间（秒），默认60秒\n    \"\"\"\n    pass\n```\n\n```python\n# tools/web.py\n@tool\ndef web_search(query: str, max_results: int = 5) -> str:\n    \"\"\"Web搜索\"\"\"\n    pass\n\n@tool\ndef web_fetch(url: str) -> str:\n    \"\"\"抓取网页内容\"\"\"\n    pass\n\n@tool\ndef message_send(channel: str, chat_id: str, content: str) -> str:\n    \"\"\"发送消息到指定渠道\"\"\"\n    pass\n\n@tool\ndef spawn_subagent(task: str, model: str = None) -> str:\n    \"\"\"Spawn a sub-agent to handle subtasks\"\"\"\n    pass\n```\n\n### 2.5 记忆层 (memory/) - LangChain化\n\n**职责**：短期会话记忆 + 长期记忆\n\n**重构方案**：LangGraph Checkpointer + LangChain Store\n\n> ⚠️ **注意**: LangChain 1.x 使用 `InMemorySaver` 替代了 `MemorySaver`\n\n```python\n# memory/checkpointer.py\nfrom langgraph.checkpoint.memory import InMemorySaver\n\n# 开发环境\ncheckpointer = InMemorySaver()\n\n# 生产环境\n# from langgraph.checkpoint.postgres import PostgresSaver\n# checkpointer = PostgresSaver.from_conn_string(conn_string)\n```\n\n```python\n# memory/store.py\nfrom langchain.storage import InMemoryStore\nfrom langgraph.store.postgres import PostgresStore\n\nclass MemoryStore:\n    \"\"\"长期记忆存储\"\"\"\n\n    def __init__(self, use_postgres: bool = False):\n        if use_postgres:\n            self.store = PostgresStore.from_conn_string(conn_string)\n        else:\n            self.store = InMemoryStore()\n\n    async def search_memory(self, query: str, k: int = 5) -> list:\n        \"\"\"语义检索相关记忆\"\"\"\n        pass\n\n    async def save_memory(self, key: str, value: str, metadata: dict = None):\n        \"\"\"保存记忆\"\"\"\n        pass\n```\n\n### 2.6 Providers层 (providers/)\n\n**职责**：LLM接口适配\n\n**保留原因**：对接各种API，但包装为LangChain接口\n\n```python\n# providers/langchain_wrapper.py\nfrom langchain_anthropic import ChatAnthropic\nfrom langchain_openai import ChatOpenAI\nfrom langchain_community.chat_models import ChatLiteLLM\n\ndef create_llm(provider: str, model: str, api_key: str = None, api_base: str = None):\n    \"\"\"创建LangChain LLM实例\"\"\"\n\n    if provider == \"anthropic\":\n        return ChatAnthropic(\n            model=model,\n            api_key=api_key,\n            base_url=api_base\n        )\n    elif provider == \"openai\":\n        return ChatOpenAI(\n            model=model,\n            api_key=api_key,\n            base_url=api_base\n        )\n    # LiteLLM支持所有OpenAI兼容的API\n    elif provider == \"litellm\":\n        return ChatLiteLLM(\n            model=model,\n            api_key=api_key,\n            base_url=api_base\n        )\n```\n\n### 2.7 配置层 (config/)\n\n**职责**：YAML/Env配置解析\n\n**保留**：Pydantic配置系统已经很完善\n\n```python\n# config/schema.py\nclass Config(BaseSettings):\n    agents: AgentsConfig\n    channels: ChannelsConfig\n    providers: ProvidersConfig\n    gateway: GatewayConfig\n    tools: ToolsConfig\n```\n\n### 2.8 会话管理 (session/)\n\n**职责**：会话历史持久化\n\n**保留**：Session管理逻辑与具体Agent实现解耦\n\n```python\n# session/manager.py\nclass SessionManager:\n    \"\"\"会话管理器\"\"\"\n\n    def __init__(self, workspace: Path):\n        self.sessions_dir = workspace / \"sessions\"\n\n    def get_or_create(self, session_key: str) -> Session:\n        \"\"\"获取或创建会话\"\"\"\n\n    def save(self, session: Session):\n        \"\"\"保存会话\"\"\"\n\n    def get_history(self, session_key: str, max_messages: int = None) -> list:\n        \"\"\"获取会话历史\"\"\"\n```\n\n## 3. 数据流\n\n### 3.1 消息处理流程\n\n```mermaid\nsequenceDiagram\n    participant C as Channel\n    participant B as MessageBus\n    participant G as AgentGraph\n    participant T as Tools\n    participant P as Provider\n    participant M as Memory\n    participant O as Outbound\n\n    C->>B: publish_inbound(msg)\n    B->>G: 消费消息\n    G->>P: LLM调用(带工具定义)\n    P-->>G: LLM响应(含tool_calls)\n    G->>T: 执行工具\n    T-->>G: 工具结果\n    G->>M: 保存checkpointer状态\n    G->>P: 继续LLM调用\n    P-->>G: 最终响应\n    G->>B: publish_outbound(response)\n    B->>O: 推送响应\n    O->>C: 发送消息\n```\n\n### 3.2 记忆系统流程\n\n```mermaid\nflowchart LR\n    A[用户消息] --> B{检查token限制}\n    B -->|超限| C[consolidate_messages]\n    B -->|正常| D[直接处理]\n    C --> E[LLM摘要]\n    E --> F[保存到MEMORY.md]\n    F --> D\n    D --> G[返回响应]\n```\n\n## 4. 目录结构\n\n```\nlanobot/\n├── main.py                    # 入口点\n├── pyproject.toml             # 项目配置\n├── CLAUDE.md                  # Claude Code指导\n│\n├── config/                    # 配置层\n│   ├── schema.py              # Pydantic配置模型\n│   ├── loader.py              # 配置加载\n│   └── paths.py               # 路径工具\n│\n├── bus/                       # 消息总线\n│   ├── queue.py               # MessageBus实现\n│   └── events.py              # 事件定义\n│\n├── channels/                  # 渠道层（保留）\n│   ├── base.py                # 渠道基类\n│   ├── manager.py             # 渠道管理\n│   ├── telegram.py\n│   ├── slack.py\n│   ├── feishu.py\n│   ├── dingtalk.py\n│   ├── discord.py\n│   ├── whatsapp.py\n│   ├── qq.py\n│   ├── wecom.py\n│   ├── email.py\n│   └── matrix.py\n│\n├── providers/                 # LLM提供商\n│   ├── base.py                # 基础接口（可保留）\n│   ├── registry.py            # 提供商注册\n│   ├── langchain_wrapper.py   # LangChain包装器\n│   ├── openai_provider.py\n│   ├── anthropic_provider.py\n│   ├── azure_openai_provider.py\n│   ├── litellm_provider.py\n│   └── custom_provider.py\n│\n├── agent/                     # Agent核心层（LangChain化）\n│   ├── __init__.py\n│   ├── state.py               # AgentState定义\n│   ├── graph.py               # LangGraph构建\n│   ├── nodes.py               # 节点函数\n│   ├── edges.py               # 边定义\n│   └── middleware.py          # 中间件\n│\n├── tools/                     # 工具层（LangChain化）\n│   ├── __init__.py\n│   ├── registry.py            # 工具注册\n│   ├── filesystem.py          # 文件操作\n│   ├── shell.py               # Shell执行\n│   ├── web.py                 # Web工具\n│   ├── message.py             # 消息发送\n│   ├── spawn.py               # 子代理\n│   ├── cron.py                # 定时任务\n│   └── mcp.py                 # MCP集成\n│\n├── memory/                    # 记忆层（LangChain化）\n│   ├── checkpointer.py        # LangGraph Checkpointer\n│   ├── store.py               # 长期记忆\n│   └── history.py             # 历史记录管理\n│\n├── session/                   # 会话管理\n│   ├── __init__.py\n│   ├── manager.py             # 会话管理器\n│   └── types.py               # 会话类型\n│\n├── cron/                      # 定时任务\n│   ├── service.py\n│   └── types.py\n│\n├── heartbeat/                 # 心跳服务\n│   └── service.py\n│\n├── cli/                       # CLI命令\n│   ├── commands.py\n│   └── __init__.py\n│\n└── utils/                     # 工具函数\n    └── helpers.py\n```\n\n## 5. 重构计划\n\n### Phase 1: 基础架构\n- [ ] 创建agent/state.py - AgentState定义\n- [ ] 创建agent/graph.py - LangGraph构建\n- [ ] 创建providers/langchain_wrapper.py - LLM包装\n- [ ] 集成现有MessageBus\n\n### Phase 2: 工具迁移\n- [ ] 将tools/filesystem.py转为@tool\n- [ ] 将tools/shell.py转为@tool\n- [ ] 将tools/web.py转为@tool\n- [ ] 集成MCP工具\n\n### Phase 3: 记忆系统\n- [ ] 集成LangGraph Checkpointer\n- [ ] 实现长期记忆Store\n- [ ] 迁移会话历史\n\n### Phase 4: 渠道适配\n- [ ] 适配现有渠道到新Agent接口\n- [ ] 测试各渠道连通性\n\n### Phase 5: CLI和配置\n- [ ] 更新CLI命令\n- [ ] 验证配置系统\n\n## 6. 关键设计决策\n\n### 6.1 为什么用混合架构?\n- **保留**：渠道、配置、业务无关的模块已经稳定\n- **重构**：Agent核心需要LangChain的标准化和生态\n\n### 6.2 LangGraph vs 纯LangChain\n- 使用`create_agent`快速构建\n- 后续可扩展为自定义StateGraph\n\n### 6.3 记忆策略\n- **短期**：LangGraph Checkpointer (thread_id级别)\n- **长期**：LangChain Store (跨会话)\n- **历史**：文件存储 (可grep搜索)\n\n### 6.4 Provider策略\n- 保留原有provider注册机制\n- 新增LangChain包装层\n- 渐进式迁移"}}, "tests/test_providers/test_registry.py": {"5ab2c91813e035a444f9a51d9778402f9d6ed946": {"t": 1785747510.7227652, "n": 6737, "h": "808c7305dbb7dd54", "c": "\"\"\"Provider 注册表测试.\"\"\"\n\nimport pytest\n\nfrom lanobot.providers.registry import (\n    PROVIDERS,\n    find_by_model,\n    find_by_name,\n    find_gateway,\n    get_default_provider,\n    get_provider,\n    list_providers,\n)\n\n\nclass TestProviderSpec:\n    \"\"\"ProviderSpec 数据类测试.\"\"\"\n\n    def test_create_spec(self):\n        \"\"\"测试创建 Provider 规范.\"\"\"\n        from lanobot.providers.registry import ProviderSpec\n\n        spec = ProviderSpec(\n            name=\"test\",\n            keywords=(\"test\",),\n            env_key=\"TEST_API_KEY\",\n            display_name=\"Test Provider\",\n            litellm_prefix=\"test\",\n        )\n        assert spec.name == \"test\"\n        assert spec.litellm_prefix == \"test\"\n        assert spec.label == \"Test Provider\"\n\n    def test_label_property(self):\n        \"\"\"测试 label 属性.\"\"\"\n        spec = get_provider(\"siliconflow\")\n        assert spec.label == \"SiliconFlow\"\n\n\nclass TestProviderRegistry:\n    \"\"\"Provider 注册表测试.\"\"\"\n\n    def test_providers_not_empty(self):\n        \"\"\"测试注册表不为空.\"\"\"\n        assert len(PROVIDERS) > 0\n\n    def test_list_providers(self):\n        \"\"\"测试列出所有 Provider.\"\"\"\n        providers = list_providers()\n        assert isinstance(providers, list)\n        assert len(providers) > 0\n\n    def test_gateway_providers(self):\n        \"\"\"测试网关 Provider.\"\"\"\n        # OpenRouter - 通过 API Key 前缀检测\n        spec = get_provider(\"openrouter\")\n        assert spec.name == \"openrouter\"\n        assert spec.is_gateway is True\n        assert spec.detect_by_key_prefix == \"sk-or-\"\n        assert spec.default_api_base == \"https://openrouter.ai/api/v1\"\n\n        # SiliconFlow - 通过 URL 关键字检测\n        spec = get_provider(\"siliconflow\")\n        assert spec.name == \"siliconflow\"\n        assert spec.is_gateway is True\n        assert \"siliconflow\" in spec.detect_by_base_keyword\n\n        # VolcEngine\n        spec = get_provider(\"volcengine\")\n        assert spec.name == \"volcengine\"\n        assert spec.is_gateway is True\n        assert \"volces\" in spec.detect_by_base_keyword\n\n    def test_standard_providers(self):\n        \"\"\"测试标准 Provider.\"\"\"\n        # DeepSeek\n        spec = get_provider(\"deepseek\")\n        assert spec.name == \"deepseek\"\n        assert spec.is_gateway is False\n        assert spec.litellm_prefix == \"deepseek\"\n        assert spec.env_key == \"DEEPSEEK_API_KEY\"\n\n        # OpenAI\n        spec = get_provider(\"openai\")\n        assert spec.name == \"openai\"\n        assert spec.litellm_prefix == \"\"\n\n        # Anthropic\n        spec = get_provider(\"anthropic\")\n        assert spec.name == \"anthropic\"\n        assert spec.supports_prompt_caching is True\n\n    def test_local_providers(self):\n        \"\"\"测试本地 Provider.\"\"\"\n        # Ollama\n        spec = get_provider(\"ollama\")\n        assert spec.name == \"ollama\"\n        assert spec.is_local is True\n        assert spec.default_api_base == \"http://localhost:11434\"\n\n        # vLLM\n        spec = get_provider(\"vllm\")\n        assert spec.name == \"vllm\"\n        assert spec.is_local is True\n\n    def test_direct_providers(self):\n        \"\"\"测试直接调用 Provider.\"\"\"\n        spec = get_provider(\"custom\")\n        assert spec.name == \"custom\"\n        assert spec.is_direct is True\n\n        spec = get_provider(\"azure_openai\")\n        assert spec.name == \"azure_openai\"\n        assert spec.is_direct is True\n\n    def test_oauth_providers(self):\n        \"\"\"测试 OAuth Provider.\"\"\"\n        spec = get_provider(\"openai_codex\")\n        assert spec.name == \"openai_codex\"\n        assert spec.is_oauth is True\n        assert spec.env_key == \"\"\n\n    def test_chinese_providers(self):\n        \"\"\"测试中文 Provider.\"\"\"\n        # Zhipu (智谱)\n        spec = get_provider(\"zhipu\")\n        assert spec.name == \"zhipu\"\n        assert \"glm\" in spec.keywords\n        assert \"zhi\" in spec.display_name.lower()\n\n        # DashScope (阿里云)\n        spec = get_provider(\"dashscope\")\n        assert spec.name == \"dashscope\"\n        assert \"qwen\" in spec.keywords\n\n        # Moonshot (月之暗面)\n        spec = get_provider(\"moonshot\")\n        assert spec.name == \"moonshot\"\n        assert \"kimi\" in spec.keywords\n\n        # MiniMax\n        spec = get_provider(\"minimax\")\n        assert spec.name == \"minimax\"\n        assert spec.default_api_base == \"https://api.minimax.io/v1\"\n\n    def test_unknown_provider_raises_error(self):\n        \"\"\"测试未知 Provider 抛出错误.\"\"\"\n        with pytest.raises(ValueError) as exc:\n            get_provider(\"unknown_provider\")\n        assert \"Unknown provider\" in str(exc.value)\n\n    def test_default_provider(self):\n        \"\"\"测试获取默认 Provider (第一个非 Direct).\"\"\"\n        default = get_default_provider()\n        # 第一个非 Direct 应该是 OpenRouter (Gateway)\n        assert default.is_gateway is True\n\n    def test_find_by_name(self):\n        \"\"\"测试按名称查找.\"\"\"\n        spec = find_by_name(\"deepseek\")\n        assert spec is not None\n        assert spec.name == \"deepseek\"\n\n        # 不存在的名称返回 None\n        assert find_by_name(\"nonexistent\") is None\n\n    def test_find_gateway_by_key_prefix(self):\n        \"\"\"测试通过 API Key 前缀检测网关.\"\"\"\n        # sk-or- -> OpenRouter\n        spec = find_gateway(\n            provider_name=None,\n            api_key=\"sk-or-v1xxxxx\",\n            api_base=None,\n        )\n        assert spec is not None\n        assert spec.name == \"openrouter\"\n\n    def test_find_gateway_by_base_keyword(self):\n        \"\"\"测试通过 URL 关键字检测网关.\"\"\"\n        # siliconflow in URL -> SiliconFlow\n        spec = find_gateway(\n            provider_name=None,\n            api_key=\"sk-xxxx\",\n            api_base=\"https://api.siliconflow.cn/v1\",\n        )\n        assert spec is not None\n        assert spec.name == \"siliconflow\"\n\n    def test_find_by_model(self):\n        \"\"\"测试通过模型名称查找 Provider.\"\"\"\n        # gpt-* -> OpenAI\n        spec = find_by_model(\"gpt-4o\")\n        assert spec is not None\n        assert spec.name == \"openai\"\n\n        # claude-* -> Anthropic\n        spec = find_by_model(\"claude-3-opus\")\n        assert spec is not None\n        assert spec.name == \"anthropic\"\n\n        # deepseek-chat -> DeepSeek\n        spec = find_by_model(\"deepseek-chat\")\n        assert spec is not None\n        assert spec.name == \"deepseek\"\n\n        # qwen-max -> DashScope\n        spec = find_by_model(\"qwen-max\")\n        assert spec is not None\n        assert spec.name == \"dashscope\"\n\n        # kimi-k2.5 -> Moonshot\n        spec = find_by_model(\"kimi-k2.5\")\n        assert spec is not None\n        assert spec.name == \"moonshot\"\n\n    def test_model_overrides(self):\n        \"\"\"测试模型参数覆盖.\"\"\"\n        spec = get_provider(\"moonshot\")\n        assert spec.model_overrides is not None\n        # kimi-k2.5 有 temperature 覆盖\n        has_override = any(m == \"kimi-k2.5\" for m, _ in spec.model_overrides)\n        assert has_override is True\n\n\nif __name__ == \"__main__\":\n    pytest.main([__file__, \"-v\"])"}, "9fd5b3cf984fd747e434f2e271b666e5a8476cb6": {"t": 1785767628.1130886, "n": 6737, "h": "808c7305dbb7dd54", "c": "\"\"\"Provider 注册表测试.\"\"\"\n\nimport pytest\n\nfrom lanobot.providers.registry import (\n    PROVIDERS,\n    find_by_model,\n    find_by_name,\n    find_gateway,\n    get_default_provider,\n    get_provider,\n    list_providers,\n)\n\n\nclass TestProviderSpec:\n    \"\"\"ProviderSpec 数据类测试.\"\"\"\n\n    def test_create_spec(self):\n        \"\"\"测试创建 Provider 规范.\"\"\"\n        from lanobot.providers.registry import ProviderSpec\n\n        spec = ProviderSpec(\n            name=\"test\",\n            keywords=(\"test\",),\n            env_key=\"TEST_API_KEY\",\n            display_name=\"Test Provider\",\n            litellm_prefix=\"test\",\n        )\n        assert spec.name == \"test\"\n        assert spec.litellm_prefix == \"test\"\n        assert spec.label == \"Test Provider\"\n\n    def test_label_property(self):\n        \"\"\"测试 label 属性.\"\"\"\n        spec = get_provider(\"siliconflow\")\n        assert spec.label == \"SiliconFlow\"\n\n\nclass TestProviderRegistry:\n    \"\"\"Provider 注册表测试.\"\"\"\n\n    def test_providers_not_empty(self):\n        \"\"\"测试注册表不为空.\"\"\"\n        assert len(PROVIDERS) > 0\n\n    def test_list_providers(self):\n        \"\"\"测试列出所有 Provider.\"\"\"\n        providers = list_providers()\n        assert isinstance(providers, list)\n        assert len(providers) > 0\n\n    def test_gateway_providers(self):\n        \"\"\"测试网关 Provider.\"\"\"\n        # OpenRouter - 通过 API Key 前缀检测\n        spec = get_provider(\"openrouter\")\n        assert spec.name == \"openrouter\"\n        assert spec.is_gateway is True\n        assert spec.detect_by_key_prefix == \"sk-or-\"\n        assert spec.default_api_base == \"https://openrouter.ai/api/v1\"\n\n        # SiliconFlow - 通过 URL 关键字检测\n        spec = get_provider(\"siliconflow\")\n        assert spec.name == \"siliconflow\"\n        assert spec.is_gateway is True\n        assert \"siliconflow\" in spec.detect_by_base_keyword\n\n        # VolcEngine\n        spec = get_provider(\"volcengine\")\n        assert spec.name == \"volcengine\"\n        assert spec.is_gateway is True\n        assert \"volces\" in spec.detect_by_base_keyword\n\n    def test_standard_providers(self):\n        \"\"\"测试标准 Provider.\"\"\"\n        # DeepSeek\n        spec = get_provider(\"deepseek\")\n        assert spec.name == \"deepseek\"\n        assert spec.is_gateway is False\n        assert spec.litellm_prefix == \"deepseek\"\n        assert spec.env_key == \"DEEPSEEK_API_KEY\"\n\n        # OpenAI\n        spec = get_provider(\"openai\")\n        assert spec.name == \"openai\"\n        assert spec.litellm_prefix == \"\"\n\n        # Anthropic\n        spec = get_provider(\"anthropic\")\n        assert spec.name == \"anthropic\"\n        assert spec.supports_prompt_caching is True\n\n    def test_local_providers(self):\n        \"\"\"测试本地 Provider.\"\"\"\n        # Ollama\n        spec = get_provider(\"ollama\")\n        assert spec.name == \"ollama\"\n        assert spec.is_local is True\n        assert spec.default_api_base == \"http://localhost:11434\"\n\n        # vLLM\n        spec = get_provider(\"vllm\")\n        assert spec.name == \"vllm\"\n        assert spec.is_local is True\n\n    def test_direct_providers(self):\n        \"\"\"测试直接调用 Provider.\"\"\"\n        spec = get_provider(\"custom\")\n        assert spec.name == \"custom\"\n        assert spec.is_direct is True\n\n        spec = get_provider(\"azure_openai\")\n        assert spec.name == \"azure_openai\"\n        assert spec.is_direct is True\n\n    def test_oauth_providers(self):\n        \"\"\"测试 OAuth Provider.\"\"\"\n        spec = get_provider(\"openai_codex\")\n        assert spec.name == \"openai_codex\"\n        assert spec.is_oauth is True\n        assert spec.env_key == \"\"\n\n    def test_chinese_providers(self):\n        \"\"\"测试中文 Provider.\"\"\"\n        # Zhipu (智谱)\n        spec = get_provider(\"zhipu\")\n        assert spec.name == \"zhipu\"\n        assert \"glm\" in spec.keywords\n        assert \"zhi\" in spec.display_name.lower()\n\n        # DashScope (阿里云)\n        spec = get_provider(\"dashscope\")\n        assert spec.name == \"dashscope\"\n        assert \"qwen\" in spec.keywords\n\n        # Moonshot (月之暗面)\n        spec = get_provider(\"moonshot\")\n        assert spec.name == \"moonshot\"\n        assert \"kimi\" in spec.keywords\n\n        # MiniMax\n        spec = get_provider(\"minimax\")\n        assert spec.name == \"minimax\"\n        assert spec.default_api_base == \"https://api.minimax.io/v1\"\n\n    def test_unknown_provider_raises_error(self):\n        \"\"\"测试未知 Provider 抛出错误.\"\"\"\n        with pytest.raises(ValueError) as exc:\n            get_provider(\"unknown_provider\")\n        assert \"Unknown provider\" in str(exc.value)\n\n    def test_default_provider(self):\n        \"\"\"测试获取默认 Provider (第一个非 Direct).\"\"\"\n        default = get_default_provider()\n        # 第一个非 Direct 应该是 OpenRouter (Gateway)\n        assert default.is_gateway is True\n\n    def test_find_by_name(self):\n        \"\"\"测试按名称查找.\"\"\"\n        spec = find_by_name(\"deepseek\")\n        assert spec is not None\n        assert spec.name == \"deepseek\"\n\n        # 不存在的名称返回 None\n        assert find_by_name(\"nonexistent\") is None\n\n    def test_find_gateway_by_key_prefix(self):\n        \"\"\"测试通过 API Key 前缀检测网关.\"\"\"\n        # sk-or- -> OpenRouter\n        spec = find_gateway(\n            provider_name=None,\n            api_key=\"sk-or-v1xxxxx\",\n            api_base=None,\n        )\n        assert spec is not None\n        assert spec.name == \"openrouter\"\n\n    def test_find_gateway_by_base_keyword(self):\n        \"\"\"测试通过 URL 关键字检测网关.\"\"\"\n        # siliconflow in URL -> SiliconFlow\n        spec = find_gateway(\n            provider_name=None,\n            api_key=\"sk-xxxx\",\n            api_base=\"https://api.siliconflow.cn/v1\",\n        )\n        assert spec is not None\n        assert spec.name == \"siliconflow\"\n\n    def test_find_by_model(self):\n        \"\"\"测试通过模型名称查找 Provider.\"\"\"\n        # gpt-* -> OpenAI\n        spec = find_by_model(\"gpt-4o\")\n        assert spec is not None\n        assert spec.name == \"openai\"\n\n        # claude-* -> Anthropic\n        spec = find_by_model(\"claude-3-opus\")\n        assert spec is not None\n        assert spec.name == \"anthropic\"\n\n        # deepseek-chat -> DeepSeek\n        spec = find_by_model(\"deepseek-chat\")\n        assert spec is not None\n        assert spec.name == \"deepseek\"\n\n        # qwen-max -> DashScope\n        spec = find_by_model(\"qwen-max\")\n        assert spec is not None\n        assert spec.name == \"dashscope\"\n\n        # kimi-k2.5 -> Moonshot\n        spec = find_by_model(\"kimi-k2.5\")\n        assert spec is not None\n        assert spec.name == \"moonshot\"\n\n    def test_model_overrides(self):\n        \"\"\"测试模型参数覆盖.\"\"\"\n        spec = get_provider(\"moonshot\")\n        assert spec.model_overrides is not None\n        # kimi-k2.5 有 temperature 覆盖\n        has_override = any(m == \"kimi-k2.5\" for m, _ in spec.model_overrides)\n        assert has_override is True\n\n\nif __name__ == \"__main__\":\n    pytest.main([__file__, \"-v\"])"}}}}