chatccc 0.2.184 → 0.2.186

package/README.md

@@ -144,12 +144,12 @@ npm install -g chatccc
 chatccc
 ```
 
-要求 Node.js >= 20。安装完成后，在任意目录执行 `chatccc` 即可启动。配置、日志和状态文件会保存在用户目录的 `.chatccc` 下：
-
-- Windows：`C:\Users\<用户名>\.chatccc\`
-- macOS / Linux：`~/.chatccc/`
-
-旧版本留在仓库或包目录下的 `config.json`、`logs/`、`state/` 会在首次启动时自动迁移到用户目录。
+要求 Node.js >= 20。安装完成后，在任意目录执行 `chatccc` 即可启动。配置、日志和状态文件会保存在用户目录的 `.chatccc` 下：
+
+- Windows：`C:\Users\<用户名>\.chatccc\`
+- macOS / Linux：`~/.chatccc/`
+
+旧版本留在仓库或包目录下的 `config.json`、`logs/`、`state/` 会在首次启动时自动迁移到用户目录。
 
 首次启动时，如果还没有有效配置，ChatCCC 会自动打开本地 Web 配置向导（默认 `http://127.0.0.1:18080`）。
 
@@ -202,7 +202,7 @@ chatccc
 # 在微信里找到机器人，发送 /new 开始对话
 ```
 
-微信登录信息会保存到 `~/.chatccc/state/ilink-auth.json`。token 过期后重新扫码即可。
+微信登录信息会保存到 `~/.chatccc/state/ilink-auth.json`。token 过期后重新扫码即可。
 
 ### 3. AI 工具配置
 
@@ -237,9 +237,20 @@ codex --version
 
 Codex 的默认模型和推理强度可继续由 `~/.codex/config.toml` 管理，也可以在 `config.json` 中覆盖。
 
+#### 可选：Chrome CDP
+
+常驻 Chrome CDP 开关用于维护本机 Chrome DevTools Protocol 端口；Codex `/usage` 可通过它读取 ChatGPT 订阅到期时间和剩余天数。
+
+依赖：
+
+- 本机已安装 **Google Chrome**。
+- 查询 ChatGPT 订阅到期时间时，需要在这个 CDP 专用 Chrome 窗口里登录 ChatGPT。
+
+默认端口是 `15166`，健康检查地址是 `http://127.0.0.1:15166/json/version`。Windows 下留空 `chromeDevtools.chromePath` 时会自动查找 Google Chrome。
+
 ### 4. `config.json`
 
-`~/.chatccc/config.json` 不存在时，ChatCCC 会从 `config.sample.json` 复制一份。常用结构如下：
+`~/.chatccc/config.json` 不存在时，ChatCCC 会从 `config.sample.json` 复制一份。常用结构如下：
 
 ```json
 {
@@ -247,32 +258,37 @@ Codex 的默认模型和推理强度可继续由 `~/.codex/config.toml` 管理
     "appId": "cli_xxxxxxxxxxxx",
     "appSecret": "xxxxxxxxxxxxxxxxxxxxxxxxxx"
   },
-  "platforms": {
-    "feishu": { "enabled": true, "platformType": "feishu" },
-    "ilink": { "enabled": true, "reuseTokenOnStart": true }
-  },
-  "port": 18080,
-  "gitTimeoutSeconds": 180,
-  "allowInterrupt": false,
-  "claude": {
-    "enabled": false,
-    "defaultAgent": true,
-    "model": "claude-sonnet-4-6",
-    "subagentModel": "",
-    "effort": "",
-    "apiKey": "",
-    "baseUrl": "",
-    "maxTurn": 0
-  },
-  "cursor": {
-    "enabled": false,
-    "defaultAgent": false,
-    "path": "",
-    "model": "",
-    "avatarBatteryMode": "apiPercent",
-    "onDemandMonthlyBudget": 1000
-  },
-  "codex": {
+  "platforms": {
+    "feishu": { "enabled": true, "platformType": "feishu" },
+    "ilink": { "enabled": true, "reuseTokenOnStart": true }
+  },
+  "chromeDevtools": {
+    "enabled": false,
+    "port": 15166,
+    "chromePath": ""
+  },
+  "port": 18080,
+  "gitTimeoutSeconds": 180,
+  "allowInterrupt": false,
+  "claude": {
+    "enabled": false,
+    "defaultAgent": true,
+    "model": "claude-sonnet-4-6",
+    "subagentModel": "",
+    "effort": "",
+    "apiKey": "",
+    "baseUrl": "",
+    "maxTurn": 0
+  },
+  "cursor": {
+    "enabled": false,
+    "defaultAgent": false,
+    "path": "",
+    "model": "",
+    "avatarBatteryMode": "apiPercent",
+    "onDemandMonthlyBudget": 1000
+  },
+  "codex": {
     "enabled": false,
     "defaultAgent": false,
     "path": "",
@@ -285,21 +301,24 @@ Codex 的默认模型和推理强度可继续由 `~/.codex/config.toml` 管理
 | 字段 | 说明 |
 | --- | --- |
 | `feishu.appId` / `feishu.appSecret` | 飞书应用凭证 |
-| `platforms.feishu.enabled` | 是否启用飞书 |
-| `platforms.feishu.platformType` | 飞书平台类型，默认 `feishu` |
-| `platforms.ilink.enabled` | 是否启用微信 iLink |
-| `platforms.ilink.reuseTokenOnStart` | 启动时是否复用已有微信登录 token |
-| `port` | 本地 Web 配置面板和中继服务端口 |
-| `gitTimeoutSeconds` | `/git` 命令超时时间，默认 180 秒 |
-| `allowInterrupt` | 是否允许新消息中断正在运行的任务；默认 false |
-| `*.enabled` | 是否启用对应 AI Agent |
-| `*.defaultAgent` | `/new` 未指定 Agent 时使用哪个工具 |
-| `cursor.path` / `codex.path` | CLI 可执行文件路径；留空时自动探测或使用 PATH |
-| `cursor.avatarBatteryMode` | Cursor 头像电量显示来源：`apiPercent` 或 `onDemandUse` |
-| `cursor.onDemandMonthlyBudget` | `avatarBatteryMode=onDemandUse` 时用于计算电量的月预算 |
-| `claude.model` / `claude.subagentModel` / `claude.effort` | 选填；设置后传给 Claude Agent SDK，留空以 `~/.claude/settings.json` 为准 |
-| `claude.apiKey` / `claude.baseUrl` | 选填；设置后传给 Claude Agent SDK，留空以 `~/.claude/settings.json` 为准 |
-| `claude.maxTurn` | 选填；Claude 最大对话轮数，默认 0（无限制），可在 Web UI 编辑 |
+| `platforms.feishu.enabled` | 是否启用飞书 |
+| `platforms.feishu.platformType` | 飞书平台类型，默认 `feishu` |
+| `platforms.ilink.enabled` | 是否启用微信 iLink |
+| `platforms.ilink.reuseTokenOnStart` | 启动时是否复用已有微信登录 token |
+| `chromeDevtools.enabled` | 是否启用常驻 Chrome CDP；默认 false |
+| `chromeDevtools.port` | Chrome CDP 端口；默认 15166 |
+| `chromeDevtools.chromePath` | Chrome 可执行文件路径；留空时自动探测 |
+| `port` | 本地 Web 配置面板和中继服务端口 |
+| `gitTimeoutSeconds` | `/git` 命令超时时间，默认 180 秒 |
+| `allowInterrupt` | 是否允许新消息中断正在运行的任务；默认 false |
+| `*.enabled` | 是否启用对应 AI Agent |
+| `*.defaultAgent` | `/new` 未指定 Agent 时使用哪个工具 |
+| `cursor.path` / `codex.path` | CLI 可执行文件路径；留空时自动探测或使用 PATH |
+| `cursor.avatarBatteryMode` | Cursor 头像电量显示来源：`apiPercent` 或 `onDemandUse` |
+| `cursor.onDemandMonthlyBudget` | `avatarBatteryMode=onDemandUse` 时用于计算电量的月预算 |
+| `claude.model` / `claude.subagentModel` / `claude.effort` | 选填；设置后传给 Claude Agent SDK，留空以 `~/.claude/settings.json` 为准 |
+| `claude.apiKey` / `claude.baseUrl` | 选填；设置后传给 Claude Agent SDK，留空以 `~/.claude/settings.json` 为准 |
+| `claude.maxTurn` | 选填；Claude 最大对话轮数，默认 0（无限制），可在 Web UI 编辑 |
 
 > **权限控制**：普通消息以 `bypassPermissions` 模式运行，跳过 Agent 操作确认。使用 `/plan` 或 `/ask` 前缀时，ChatCCC 自动切换为只读模式：Claude SDK 仅放行 Read + stop-stuck-loop 网络请求，Codex 使用 `--sandbox read-only`，Cursor 使用 `--mode plan/ask`。请只在可信环境中使用。
 
@@ -315,26 +334,26 @@ Codex 的默认模型和推理强度可继续由 `~/.codex/config.toml` 管理
 | --- | --- |
 | `/new` | 使用默认 Agent 创建新会话 |
 | `/new claude` | 创建 Claude Code 会话 |
-| `/new cursor` | 创建 Cursor 会话 |
-| `/new codex` | 创建 Codex 会话 |
-| `/newh` | 重置当前会话，保留工作目录 |
-| `/model` | 查看或切换当前会话的模型 |
-| `/stop` | 停止当前回复 |
-| `/cancel` | 取消当前会话里排队等待处理的消息 |
-| `/state` | 查看当前会话状态 |
-| `/cd` | 查看或设置当前会话工作目录 |
-| `/sessions` | 查看所有会话状态 |
-| `/session <数字>` | 将当前群聊切换到 `/sessions` 列表中的指定会话 |
-| `/usage` | 查看当前会话对应 Agent 的用量；Codex 显示 5h/周用量，Cursor 显示当前周期用量 |
-| `/git <子命令>` | 在当前会话工作目录执行 `git ...` 并回传输出 |
-| `/abd<内容>` | 去掉 `/abd` 前缀后把内容发给 Agent，并在消息末尾追加第一性原理需求澄清提示 |
-| `/plan <内容>` | 只读计划模式：仅允许读文件和 stop-stuck-loop 请求，不执行任何写操作 |
-| `/ask <内容>` | 只读问答模式：与 /plan 相同，仅允许读文件和 stop-stuck-loop 请求 |
-| `/restart` | 重启机器人进程 |
-| `/update` | 更新 npm 全局包并重启（仅限 `npm install -g chatccc` 安装的全局进程） |
-| `/deleteg` | 解散当前飞书会话群；Agent 会话记录保留 |
-
-> **模型切换**：`/model` 查看当前会话 Agent 的可选模型清单，`/model <名称>` 模糊匹配切换，`/model clear` 恢复默认。可选模型来自当前 Agent 的配置：Claude 使用 `claude.model` / `claude.subagentModel`，Cursor 使用 `cursor.model`，Codex 使用 `codex.model`。
+| `/new cursor` | 创建 Cursor 会话 |
+| `/new codex` | 创建 Codex 会话 |
+| `/newh` | 重置当前会话，保留工作目录 |
+| `/model` | 查看或切换当前会话的模型 |
+| `/stop` | 停止当前回复 |
+| `/cancel` | 取消当前会话里排队等待处理的消息 |
+| `/state` | 查看当前会话状态 |
+| `/cd` | 查看或设置当前会话工作目录 |
+| `/sessions` | 查看所有会话状态 |
+| `/session <数字>` | 将当前群聊切换到 `/sessions` 列表中的指定会话 |
+| `/usage` | 查看当前会话对应 Agent 的用量；Codex 显示 5h/周用量，若 Chrome CDP 可用且 ChatGPT 已登录，会额外显示订阅到期时间；Cursor 显示当前周期用量 |
+| `/git <子命令>` | 在当前会话工作目录执行 `git ...` 并回传输出 |
+| `/abd<内容>` | 去掉 `/abd` 前缀后把内容发给 Agent，并在消息末尾追加第一性原理需求澄清提示 |
+| `/plan <内容>` | 只读计划模式：仅允许读文件和 stop-stuck-loop 请求，不执行任何写操作 |
+| `/ask <内容>` | 只读问答模式：与 /plan 相同，仅允许读文件和 stop-stuck-loop 请求 |
+| `/restart` | 重启机器人进程 |
+| `/update` | 更新 npm 全局包并重启（仅限 `npm install -g chatccc` 安装的全局进程） |
+| `/deleteg` | 解散当前飞书会话群；Agent 会话记录保留 |
+
+> **模型切换**：`/model` 查看当前会话 Agent 的可选模型清单，`/model <名称>` 模糊匹配切换，`/model clear` 恢复默认。可选模型来自当前 Agent 的配置：Claude 使用 `claude.model` / `claude.subagentModel`，Cursor 使用 `cursor.model`，Codex 使用 `codex.model`。
 
 ---
 

package/agent-prompts/cursor_specific.md

@@ -1,3 +1,14 @@
 ## 计划/问答模式权限规则
 
-如果用户消息以 `/plan` 或 `/ask` 开头，说明当前处于只读模式。**遇到任何需要用户同意的权限请求时，不要向用户申请权限，立即调用 stop-stuck-loop 接口提前结束本轮会话。**
+如果用户消息以 `/plan` 或 `/ask` 开头，说明当前处于只读模式。**遇到任何需要用户同意的权限请求时，不要向用户申请权限，立即调用 stop-stuck-loop 接口提前结束本轮会话。**
+
+## 用户问题澄清规则
+
+当需要在开始实现前向用户确认问题、让用户做选择，或用户要求“从第一性原理出发挖掘真实需求 / 判断需求是否合理 / 开始实现前有什么问题要问我”时，**不要调用 `AskQuestion`、多选表单或任何结构化选择工具**。
+
+必须直接用普通文本回复用户：
+
+- 先简要说明你理解到的真实需求和你对合理性的判断。
+- 再列出需要用户确认的问题；如果有推荐选项，就用文本说明“我建议 X，原因是 Y”。
+- 问题数量尽量少，只问会影响实现边界、风险或不可逆决策的问题。
+- 不要把这些问题放进工具调用参数里，也不要只在卡片/工具结果中呈现。

package/config.sample.json

@@ -7,10 +7,35 @@
     "feishu": { "enabled": true, "platformType": "feishu" },
     "ilink": { "enabled": true, "reuseTokenOnStart": true }
   },
-  "port": 18080,
-  "gitTimeoutSeconds": 180,
-  "allowInterrupt": false,
-  "claude": {
+  "chromeDevtools": {
+    "enabled": false,
+    "port": 15166,
+    "chromePath": ""
+  },
+  "port": 18080,
+  "gitTimeoutSeconds": 180,
+  "allowInterrupt": false,
+  "rawStreamLogs": {
+    "claude": {
+      "enabled": false,
+      "maxBytesPerTurn": 52428800,
+      "retentionDays": 7,
+      "keepCompleted": false
+    },
+    "cursor": {
+      "enabled": false,
+      "maxBytesPerTurn": 52428800,
+      "retentionDays": 7,
+      "keepCompleted": false
+    },
+    "codex": {
+      "enabled": false,
+      "maxBytesPerTurn": 52428800,
+      "retentionDays": 7,
+      "keepCompleted": false
+    }
+  },
+  "claude": {
     "enabled": false,
     "defaultAgent": true,
     "model": "",

package/im-skills/feishu-skill/skill.md

@@ -7,6 +7,6 @@ Current working directory: {{cwd}}
 
 Use local endpoints instead of calling Feishu Open Platform directly.
 
-- **Send images**: POST `{{send_image_url}}` with `{"session_id":"{{session_id}}","path":"<absolute path>","caption":"<optional>"}` — read `{{im_skills_cache_dir}}/feishu-skill/receive-send-image.md`
-- **Send files/videos**: POST `{{send_file_url}}` with `{"session_id":"{{session_id}}","path":"<absolute path>","caption":"<optional>"}` — read `{{im_skills_cache_dir}}/feishu-skill/receive-send-file.md`
-- **Delegate a task to a new agent conversation**: POST `{{delegate_task_url}}` with `{"tool":"codex|claude|cursor","cwd":"<absolute working directory>","open_id":"<Feishu user open_id>","prompt":"<first task>"}`. Use `open_ids` for multiple users. This uses the normal ChatCCC prompt flow, so project prompt injection and IM skills still apply.
+- **Send images**: POST `{{send_image_url}}` with `{"session_id":"{{session_id}}","path":"<absolute path>","caption":"<optional>"}` — read `{{im_skills_cache_dir}}/feishu-skill/receive-send-image.md`
+- **Send files/videos**: POST `{{send_file_url}}` with `{"session_id":"{{session_id}}","path":"<absolute path>","caption":"<optional>"}` — read `{{im_skills_cache_dir}}/feishu-skill/receive-send-file.md`
+- **Delegate a task to a new agent conversation**: POST `{{delegate_task_url}}` with `{"tool":"codex|claude|cursor","cwd":"<absolute working directory>","open_id":"<Feishu user open_id>","prompt":"<first task>"}`. Use `open_ids` for multiple users. This uses the normal ChatCCC prompt flow, so project prompt injection and IM skills still apply.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "chatccc",
-  "version": "0.2.184",
+  "version": "0.2.186",
   "description": "Feishu bot bridge for Claude Code",
   "license": "Apache-2.0",
   "type": "module",

package/src/__tests__/agent-delegate-task-rpc.test.ts

@@ -1,165 +1,165 @@
-import { Readable } from "node:stream";
-import { describe, expect, it, vi, beforeEach } from "vitest";
-
-import type { PlatformAdapter } from "../platform-adapter.ts";
-
-const mockDelegateAgentTask = vi.hoisted(() => vi.fn());
-
-vi.mock("../agent-delegate-task.ts", () => ({
-  delegateAgentTask: mockDelegateAgentTask,
-}));
-
-import {
-  AGENT_DELEGATE_TASK_PATH,
-  buildAgentDelegateTaskCapabilityPrompt,
-  handleAgentDelegateTaskRequest,
-} from "../agent-delegate-task-rpc.ts";
-import { ABD_APPEND_PROMPT } from "../shared-prefix.ts";
-
-function request(body: unknown, path = AGENT_DELEGATE_TASK_PATH, method = "POST"): Readable & {
-  url?: string;
-  method?: string;
-  headers: Record<string, string>;
-} {
-  const req = Readable.from([Buffer.from(JSON.stringify(body), "utf8")]) as Readable & {
-    url?: string;
-    method?: string;
-    headers: Record<string, string>;
-  };
-  req.url = path;
-  req.method = method;
-  req.headers = { "content-type": "application/json; charset=utf-8" };
-  return req;
-}
-
-function response() {
-  const res = {
-    statusCode: 0,
-    headers: {} as Record<string, string>,
-    body: "",
-    writeHead(status: number, headers: Record<string, string>) {
-      this.statusCode = status;
-      this.headers = headers;
-      return this;
-    },
-    end(chunk?: string) {
-      this.body += chunk ?? "";
-      return this;
-    },
-  };
-  return res;
-}
-
-function platform(kind: "feishu" | "wechat" = "feishu"): PlatformAdapter {
-  return {
-    kind,
-    sendText: vi.fn(async () => true),
-    sendCard: vi.fn(async () => true),
-    sendRawCard: vi.fn(async () => true),
-    createGroup: vi.fn(async () => "chat-id"),
-    updateChatInfo: vi.fn(async () => {}),
-    getChatInfo: vi.fn(async () => ({ name: "chat", description: "" })),
-    disbandChat: vi.fn(async () => {}),
-    setChatAvatar: vi.fn(async () => {}),
-    extractSessionInfo: vi.fn(() => null),
-    cardCreate: vi.fn(async () => "card-id"),
-    cardSend: vi.fn(async () => "message-id"),
-    cardUpdate: vi.fn(async () => {}),
-  };
-}
-
-describe("agent delegate task RPC", () => {
-  beforeEach(() => {
-    mockDelegateAgentTask.mockReset();
-    mockDelegateAgentTask.mockResolvedValue({
-      chatId: "chat-new",
-      sessionId: "sid-new",
-      tool: "codex",
-      cwd: "F:\\repo",
-    });
-  });
-
-  it("delegates a task through the local API", async () => {
-    const req = request({
-      tool: "codex",
-      cwd: "F:\\repo",
-      open_id: "ou-user",
-      prompt: "帮我分析日志",
-    });
-    const res = response();
-
-    await handleAgentDelegateTaskRequest(req as never, res as never, platform());
-
-    expect(res.statusCode).toBe(200);
-    expect(JSON.parse(res.body)).toMatchObject({
-      ok: true,
-      chat_id: "chat-new",
-      session_id: "sid-new",
-      tool: "codex",
-    });
-    expect(mockDelegateAgentTask).toHaveBeenCalledWith(expect.objectContaining({
-      tool: "codex",
-      cwd: expect.stringContaining("repo"),
-      promptText: "帮我分析日志",
-      openIds: ["ou-user"],
-      chatNamePrefix: "帮我分析日志",
-    }));
-  });
-
-  it("applies the shared ABD prefix to API prompts", async () => {
-    const req = request({
-      tool: "claude",
-      cwd: "F:\\repo",
-      openIds: ["ou-user"],
-      prompt: "/abd帮我分析",
-    });
-    const res = response();
-
-    await handleAgentDelegateTaskRequest(req as never, res as never, platform());
-
-    expect(res.statusCode).toBe(200);
-    expect(mockDelegateAgentTask).toHaveBeenCalledWith(expect.objectContaining({
-      promptText: `帮我分析\n\n---\n${ABD_APPEND_PROMPT}`,
-      chatNamePrefix: "帮我分析",
-    }));
-  });
-
-  it("rejects non-Feishu platforms", async () => {
-    const req = request({
-      tool: "codex",
-      cwd: "F:\\repo",
-      open_id: "ou-user",
-      prompt: "task",
-    });
-    const res = response();
-
-    await handleAgentDelegateTaskRequest(req as never, res as never, platform("wechat"));
-
-    expect(res.statusCode).toBe(409);
-    expect(mockDelegateAgentTask).not.toHaveBeenCalled();
-  });
-
-  it("rejects missing required parameters", async () => {
-    const req = request({ tool: "codex", cwd: "F:\\repo", prompt: "task" });
-    const res = response();
-
-    await handleAgentDelegateTaskRequest(req as never, res as never, platform());
-
-    expect(res.statusCode).toBe(400);
-    expect(JSON.parse(res.body).error).toContain("open_id");
-    expect(mockDelegateAgentTask).not.toHaveBeenCalled();
-  });
-
-  it("builds prompt instructions for the delegate endpoint", () => {
-    const prompt = buildAgentDelegateTaskCapabilityPrompt({
-      url: `http://127.0.0.1:18080${AGENT_DELEGATE_TASK_PATH}`,
-      cwd: "F:/repo",
-    });
-
-    expect(prompt).toContain("POST http://127.0.0.1:18080/api/agent/delegate-task");
-    expect(prompt).toContain('"tool":"codex|claude|cursor"');
-    expect(prompt).toContain('"cwd":"absolute working directory"');
-    expect(prompt).toContain("project prompt injection and IM skills still apply");
-    expect(prompt).toContain("Current working directory: F:/repo");
-  });
-});
+import { Readable } from "node:stream";
+import { describe, expect, it, vi, beforeEach } from "vitest";
+
+import type { PlatformAdapter } from "../platform-adapter.ts";
+
+const mockDelegateAgentTask = vi.hoisted(() => vi.fn());
+
+vi.mock("../agent-delegate-task.ts", () => ({
+  delegateAgentTask: mockDelegateAgentTask,
+}));
+
+import {
+  AGENT_DELEGATE_TASK_PATH,
+  buildAgentDelegateTaskCapabilityPrompt,
+  handleAgentDelegateTaskRequest,
+} from "../agent-delegate-task-rpc.ts";
+import { ABD_APPEND_PROMPT } from "../shared-prefix.ts";
+
+function request(body: unknown, path = AGENT_DELEGATE_TASK_PATH, method = "POST"): Readable & {
+  url?: string;
+  method?: string;
+  headers: Record<string, string>;
+} {
+  const req = Readable.from([Buffer.from(JSON.stringify(body), "utf8")]) as Readable & {
+    url?: string;
+    method?: string;
+    headers: Record<string, string>;
+  };
+  req.url = path;
+  req.method = method;
+  req.headers = { "content-type": "application/json; charset=utf-8" };
+  return req;
+}
+
+function response() {
+  const res = {
+    statusCode: 0,
+    headers: {} as Record<string, string>,
+    body: "",
+    writeHead(status: number, headers: Record<string, string>) {
+      this.statusCode = status;
+      this.headers = headers;
+      return this;
+    },
+    end(chunk?: string) {
+      this.body += chunk ?? "";
+      return this;
+    },
+  };
+  return res;
+}
+
+function platform(kind: "feishu" | "wechat" = "feishu"): PlatformAdapter {
+  return {
+    kind,
+    sendText: vi.fn(async () => true),
+    sendCard: vi.fn(async () => true),
+    sendRawCard: vi.fn(async () => true),
+    createGroup: vi.fn(async () => "chat-id"),
+    updateChatInfo: vi.fn(async () => {}),
+    getChatInfo: vi.fn(async () => ({ name: "chat", description: "" })),
+    disbandChat: vi.fn(async () => {}),
+    setChatAvatar: vi.fn(async () => {}),
+    extractSessionInfo: vi.fn(() => null),
+    cardCreate: vi.fn(async () => "card-id"),
+    cardSend: vi.fn(async () => "message-id"),
+    cardUpdate: vi.fn(async () => {}),
+  };
+}
+
+describe("agent delegate task RPC", () => {
+  beforeEach(() => {
+    mockDelegateAgentTask.mockReset();
+    mockDelegateAgentTask.mockResolvedValue({
+      chatId: "chat-new",
+      sessionId: "sid-new",
+      tool: "codex",
+      cwd: "F:\\repo",
+    });
+  });
+
+  it("delegates a task through the local API", async () => {
+    const req = request({
+      tool: "codex",
+      cwd: "F:\\repo",
+      open_id: "ou-user",
+      prompt: "帮我分析日志",
+    });
+    const res = response();
+
+    await handleAgentDelegateTaskRequest(req as never, res as never, platform());
+
+    expect(res.statusCode).toBe(200);
+    expect(JSON.parse(res.body)).toMatchObject({
+      ok: true,
+      chat_id: "chat-new",
+      session_id: "sid-new",
+      tool: "codex",
+    });
+    expect(mockDelegateAgentTask).toHaveBeenCalledWith(expect.objectContaining({
+      tool: "codex",
+      cwd: expect.stringContaining("repo"),
+      promptText: "帮我分析日志",
+      openIds: ["ou-user"],
+      chatNamePrefix: "帮我分析日志",
+    }));
+  });
+
+  it("applies the shared ABD prefix to API prompts", async () => {
+    const req = request({
+      tool: "claude",
+      cwd: "F:\\repo",
+      openIds: ["ou-user"],
+      prompt: "/abd帮我分析",
+    });
+    const res = response();
+
+    await handleAgentDelegateTaskRequest(req as never, res as never, platform());
+
+    expect(res.statusCode).toBe(200);
+    expect(mockDelegateAgentTask).toHaveBeenCalledWith(expect.objectContaining({
+      promptText: `帮我分析\n\n---\n${ABD_APPEND_PROMPT}`,
+      chatNamePrefix: "帮我分析",
+    }));
+  });
+
+  it("rejects non-Feishu platforms", async () => {
+    const req = request({
+      tool: "codex",
+      cwd: "F:\\repo",
+      open_id: "ou-user",
+      prompt: "task",
+    });
+    const res = response();
+
+    await handleAgentDelegateTaskRequest(req as never, res as never, platform("wechat"));
+
+    expect(res.statusCode).toBe(409);
+    expect(mockDelegateAgentTask).not.toHaveBeenCalled();
+  });
+
+  it("rejects missing required parameters", async () => {
+    const req = request({ tool: "codex", cwd: "F:\\repo", prompt: "task" });
+    const res = response();
+
+    await handleAgentDelegateTaskRequest(req as never, res as never, platform());
+
+    expect(res.statusCode).toBe(400);
+    expect(JSON.parse(res.body).error).toContain("open_id");
+    expect(mockDelegateAgentTask).not.toHaveBeenCalled();
+  });
+
+  it("builds prompt instructions for the delegate endpoint", () => {
+    const prompt = buildAgentDelegateTaskCapabilityPrompt({
+      url: `http://127.0.0.1:18080${AGENT_DELEGATE_TASK_PATH}`,
+      cwd: "F:/repo",
+    });
+
+    expect(prompt).toContain("POST http://127.0.0.1:18080/api/agent/delegate-task");
+    expect(prompt).toContain('"tool":"codex|claude|cursor"');
+    expect(prompt).toContain('"cwd":"absolute working directory"');
+    expect(prompt).toContain("project prompt injection and IM skills still apply");
+    expect(prompt).toContain("Current working directory: F:/repo");
+  });
+});