@wu529778790/open-im 1.11.1-beta.8 → 1.11.1

package/README.md

@@ -2,66 +2,45 @@
 
 **English** · [中文](./README.zh-CN.md)
 
-Multi-platform IM bridge for AI CLI tools. Connect Telegram, Feishu, WeCom, DingTalk, QQ, WeChat (WorkBuddy), and WeChat (ClawBot) to Claude Code, Codex, and CodeBuddy — use your AI coding assistant from any phone or chat window.
+> Your AI coding assistant, in every chat app.
+
+open-im bridges Claude Code, Codex, and CodeBuddy to Telegram, Feishu, WeCom, DingTalk, QQ, WeChat (WorkBuddy), and WeChat (ClawBot). Send a message from your phone, get code written on your server.
 
 ## Architecture
 
 ![Open-IM Architecture](./diagram/architecture/open-im-architecture.svg)
 
-## Features
-
-- **Seven IM platforms** — Telegram, Feishu, WeCom, DingTalk, QQ, WorkBuddy, ClawBot
-- **Three AI backends** — Claude (Agent SDK), Codex, CodeBuddy (per-platform override supported)
-- **Streaming, media, sessions** — live output where supported; `/new` for a fresh AI session
-- **Web UI** — dashboard bundled in the package; default **`http://127.0.0.1:39282`**
-
-## Requirements
-
-- Node.js ≥ 20
-- At least one IM platform configured + credentials for your AI tool
-
-## Quick start
-
-```bash
-npx @wu529778790/open-im start
-```
-
-Or: `npm install -g @wu529778790/open-im` then `open-im start`.
+## Why
 
-Config: **`~/.open-im/config.json`**
+- **Work from anywhere** — trigger Claude Code from your phone while commuting, waiting in line, or on the couch
+- **Seamless handoff** — open-im shares sessions with the Claude Code CLI; pick up on your computer exactly where you left off on your phone
+- **Full power, simple interface** — stream responses, manage sessions, switch models — all through chat commands
+- **One bridge, many platforms** — same bot works on Telegram, Feishu, DingTalk, WeChat, and more
 
-## CLI
-
-| Command | Description |
-| --- | --- |
-| `open-im init` | Interactive setup (does not start the bridge) |
-| `open-im start` | Run the bridge in the background |
-| `open-im stop` | Stop the background bridge |
-| `open-im restart` | Stop then start |
-| `open-im dashboard` | Web config server only (no bridge) |
-
-After `start`, the CLI prints the dashboard URL (default **`http://127.0.0.1:39282`**).
+## Features
 
-## Chat commands
+### Chat commands
 
 | Command | Description |
 | --- | --- |
-| `/help` | Help |
-| `/new` | New AI session |
-| `/sessions` | Session history with preview |
-| `/resume [N]` | Resume session (no arg = most recent) |
-| `/status` | AI + session info |
-| `/cd` / `/pwd` | Switch work dir (auto-resumes that dir's session) |
-| `/allow` / `/y`, `/deny` / `/n` | Permission prompts |
-
-## Session continuity
+| `/help` | Show all commands |
+| `/new` | Start a fresh AI session |
+| `/sessions` | Browse session history with previews |
+| `/resume [N]` | Resume a session (no arg = most recent) |
+| `/history [N]` | View conversation messages in a session |
+| `/delete <N>` | Delete a session |
+| `/rename <title>` | Rename the current session |
+| `/fork [N]` | Fork a session (create a branch) |
+| `/models` | List available AI models |
+| `/context` | Show context window usage |
+| `/status` | Show AI tool, account, and session info |
+| `/cd <path>` / `/pwd` | Switch work directory (auto-resumes that dir's session) |
+| `/allow` / `/y`, `/deny` / `/n` | Respond to permission prompts |
+
+### Session continuity
 
 open-im and Claude Code CLI share the same session storage. In the same directory, you can seamlessly switch between phone and computer.
 
-**Phone → Computer:** open-im auto-resumes the latest CLI session — no configuration needed.
-
-**Computer → Phone:** use `claude --continue` (or `claude -c`) to pick up the phone conversation.
-
 ```
 # On computer
 cd /my-project && claude        # work as usual, then Ctrl+C
@@ -75,16 +54,43 @@ claude -c                       # continues the phone conversation
 
 > Only one side can be active at a time. Exit the CLI before sending messages from the phone, and vice versa.
 
-## Git co-authors
+### Platform support
 
-`Co-authored-by` is appended by default on AI-driven commits. **Disable:** set **`OPEN_IM_GIT_COAUTHOR=0`** in the environment and restart the bridge.
+Seven IM platforms, three AI backends, per-platform override:
 
-## Minimal config
+| Platform | Streaming | Media | Notes |
+| --- | --- | --- | --- |
+| Telegram | Yes | Images | Full bot support |
+| Feishu | Yes | Images | Streaming card |
+| WeCom | Yes | Images | Streaming card |
+| DingTalk | Partial | Images | Fallback to text |
+| QQ | Yes | Images | |
+| WorkBuddy | Yes | Images | WeChat-based |
+| ClawBot | Yes | Images | WeChat-based |
+
+Set `platforms.<name>.aiCommand` (`claude` / `codex` / `codebuddy`) per channel. Default: `claude`.
+
+### Web dashboard
+
+`open-im start` serves a built-in SPA and API at **`http://127.0.0.1:39282`** (configurable via `OPEN_IM_WEB_PORT`). For LAN access: `export OPEN_IM_WEB_HOST=0.0.0.0`.
+
+## Quick start
+
+```bash
+npx @wu529778790/open-im init    # interactive setup
+npx @wu529778790/open-im start   # run the bridge
+```
+
+Or install globally: `npm install -g @wu529778790/open-im` then `open-im start`.
+
+Config file: **`~/.open-im/config.json`**
+
+### Minimal config
 
 ```json
 {
   "tools": {
-    "claude": { "workDir": "/path/to/project", "skipPermissions": true, "timeoutMs": 600000 }
+    "claude": { "workDir": "/path/to/project", "skipPermissions": true }
   },
   "platforms": {
     "telegram": { "enabled": true, "botToken": "YOUR_TELEGRAM_BOT_TOKEN" }
@@ -92,11 +98,11 @@ claude -c                       # continues the phone conversation
 }
 ```
 
-Add other platforms under **`platforms`** as needed. Run **`open-im init`** for a full template.
+Run `open-im init` for a full template with all platforms.
 
 ### Claude (Agent SDK)
 
-No local `claude` binary required. Third-party / compatible API:
+No local `claude` binary required. Supports third-party / compatible APIs:
 
 ```json
 {
@@ -112,26 +118,37 @@ No local `claude` binary required. Third-party / compatible API:
 }
 ```
 
-### Per-platform AI
+### CLI reference
 
-Set **`platforms.<name>.aiCommand`** (`claude` / `codex` / `codebuddy`) per channel. Default: `claude`.
+| Command | Description |
+| --- | --- |
+| `open-im init` | Interactive setup (does not start the bridge) |
+| `open-im start` | Run the bridge in the background |
+| `open-im stop` | Stop the background bridge |
+| `open-im restart` | Stop then start |
+| `open-im dashboard` | Web config server only (no bridge) |
 
-### Web dashboard
+### Environment variables
 
-`open-im start` serves the built-in SPA and **`/api/*`** on **`OPEN_IM_WEB_PORT`** (default **39282**). For LAN access: `export OPEN_IM_WEB_HOST=0.0.0.0`.
+**`ANTHROPIC_*`** (shell or `tools.claude.env`), **`TELEGRAM_BOT_TOKEN`**, **`OPEN_IM_WEB_PORT`**, **`OPEN_IM_WEB_HOST`**, plus platform-specific `*_APP_ID`, `*_SECRET`, `WORKBUDDY_*`, etc.
 
-### Environment variables
+### Git co-authors
 
-Typical keys: **`ANTHROPIC_*`** (shell or **`tools.claude.env`**), **`TELEGRAM_BOT_TOKEN`**, **`OPEN_IM_WEB_PORT`**, **`OPEN_IM_WEB_HOST`**, plus platform-specific `*_APP_ID`, `*_SECRET`, `WORKBUDDY_*`, etc.
+`Co-authored-by` is appended by default on AI-driven commits. Disable: set **`OPEN_IM_GIT_COAUTHOR=0`** in the environment and restart the bridge.
 
 ### Privacy
 
-**Anonymous** usage information may be collected to improve reliability (no chat or prompt content). To disable: **`OPEN_IM_TELEMETRY=false`** or **`"telemetry": { "enabled": false }`** in **`config.json`**.
+**Anonymous** usage information may be collected to improve reliability (no chat or prompt content). To disable: **`OPEN_IM_TELEMETRY=false`** or `"telemetry": { "enabled": false }` in `config.json`.
 
 ## Platform setup & troubleshooting
 
 See **[docs/platforms.md](./docs/platforms.md)** for detailed per-platform configuration, credential setup, and troubleshooting.
 
+## Requirements
+
+- Node.js >= 20
+- At least one IM platform configured + credentials for your AI tool
+
 ## License
 
 [MIT](LICENSE)

package/README.zh-CN.md

@@ -2,65 +2,44 @@
 
 [English](./README.md) · **中文**
 
-多平台 IM 桥接：把 Telegram、飞书、企业微信、钉钉、QQ、微信（WorkBuddy）、微信（ClawBot）接到 Claude Code、Codex、CodeBuddy，在手机或聊天里使用 AI 编程助手。
+> 你的 AI 编程助手，在每个聊天 App 里。
+
+open-im 把 Claude Code、Codex、CodeBuddy 接入 Telegram、飞书、企业微信、钉钉、QQ、微信（WorkBuddy）、微信（ClawBot）。手机发条消息，服务器上就写好代码。
 
 ## 架构
 
 ![Open-IM 架构图](./diagram/architecture/open-im-architecture.svg)
 
-## 功能特性
+## 为什么用 open-im
 
-- **七个 IM 平台** — Telegram、飞书、企业微信、钉钉、QQ、WorkBuddy、ClawBot
-- **三种 AI 后端** — Claude（Agent SDK）、Codex、CodeBuddy（可按平台覆盖）
-- **流式、多媒体、会话** — 视平台能力；`/new` 开启新 AI 会话
-- **Web 控制台** — 随包内置，默认 **`http://127.0.0.1:39282`**
+- **随时随地** — 通勤、排队、躺沙发上，手机发消息就能让 Claude Code 干活
+- **无缝接力** — open-im 和 Claude Code CLI 共享 session，在手机上聊到一半，电脑上接着来
+- **完整能力，简单界面** — 流式输出、会话管理、模型切换，全靠聊天命令搞定
+- **一个桥接，多个平台** — 同一个 bot 支持 Telegram、飞书、钉钉、微信等
 
-## 环境要求
+## 功能
 
-- Node.js ≥ 20
-- 至少配置一个 IM 平台 + 所选 AI 的凭证
-
-## 快速开始
-
-```bash
-npx @wu529778790/open-im start
-```
-
-或：`npm install -g @wu529778790/open-im` 后执行 `open-im start`。
-
-配置：**`~/.open-im/config.json`**
-
-## CLI 命令
+### 聊天命令
 
 | 命令 | 说明 |
 | --- | --- |
-| `open-im init` | 交互配置（不启动桥接） |
-| `open-im start` | 后台运行桥接 |
-| `open-im stop` | 停止后台服务 |
-| `open-im restart` | 重启 |
-| `open-im dashboard` | 仅 Web 配置服务（不启动桥接） |
-
-`start` 后会提示控制台地址（默认 **`http://127.0.0.1:39282`**）。
-
-## IM 内命令
-
-| 命令 | 说明 |
-| --- | --- |
-| `/help` | 帮助 |
-| `/new` | 新 AI 会话 |
-| `/sessions` | 历史会话（含摘要预览） |
+| `/help` | 显示所有命令 |
+| `/new` | 开启新 AI 会话 |
+| `/sessions` | 浏览历史会话（含摘要预览） |
 | `/resume [序号]` | 恢复会话（无参数恢复最近一条） |
-| `/status` | AI 与会话信息 |
-| `/cd` / `/pwd` | 切换工作目录（自动恢复该目录的历史会话） |
+| `/history [序号]` | 查看会话对话记录 |
+| `/delete <序号>` | 删除历史会话 |
+| `/rename <标题>` | 重命名当前会话 |
+| `/fork [序号]` | 分支会话（创建副本） |
+| `/models` | 查看可用 AI 模型 |
+| `/context` | 查看上下文窗口占用 |
+| `/status` | 显示 AI 工具、账号、会话信息 |
+| `/cd <路径>` / `/pwd` | 切换工作目录（自动恢复该目录的历史会话） |
 | `/allow` / `/y`、`/deny` / `/n` | 权限确认 |
 
-## 会话接力
-
-open-im 和 Claude Code CLI 共享同一份 session 存储。在同一个目录下，手机和电脑可以无缝切换。
+### 会话接力
 
-**手机接电脑：** open-im 自动恢复同目录下最新的 CLI session，无需配置。
-
-**电脑接手机：** 使用 `claude --continue`（或 `claude -c`）接上 open-im 端的对话。
+open-im 和 Claude Code CLI 共享同一份 session 存储。同一个目录下，手机和电脑可以无缝切换。
 
 ```bash
 # 电脑端
@@ -75,16 +54,43 @@ claude -c                       # 接上手机端的对话
 
 > 同一时刻只能有一端活跃。从手机发消息前先退出 CLI，反之亦然。
 
-## Git 共同作者
+### 平台支持
+
+七个 IM 平台，三种 AI 后端，可按平台覆盖：
 
-默认在 AI 发起的提交里追加 `Co-authored-by`。**关闭**：设置环境变量 **`OPEN_IM_GIT_COAUTHOR=0`** 并重启桥接。
+| 平台 | 流式输出 | 备注 |
+| --- | --- | --- |
+| Telegram | 支持 | 完整 bot 支持 |
+| 飞书 | 支持 | 流式卡片 |
+| 企业微信 | 支持 | 流式卡片 |
+| 钉钉 | 部分 | 回退到纯文本 |
+| QQ | 支持 | |
+| WorkBuddy | 支持 | 微信生态 |
+| ClawBot | 支持 | 微信生态 |
 
-## 最小配置
+在每个平台上设置 `platforms.<name>.aiCommand`（`claude` / `codex` / `codebuddy`），默认 `claude`。
+
+### Web 控制台
+
+`open-im start` 在 **`http://127.0.0.1:39282`** 提供内置页面与 API（通过 `OPEN_IM_WEB_PORT` 配置端口）。局域网访问：`export OPEN_IM_WEB_HOST=0.0.0.0`。
+
+## 快速开始
+
+```bash
+npx @wu529778790/open-im init    # 交互式配置
+npx @wu529778790/open-im start   # 启动桥接
+```
+
+或全局安装：`npm install -g @wu529778790/open-im` 后执行 `open-im start`。
+
+配置文件：**`~/.open-im/config.json`**
+
+### 最小配置
 
 ```json
 {
   "tools": {
-    "claude": { "workDir": "/path/to/project", "skipPermissions": true, "timeoutMs": 600000 }
+    "claude": { "workDir": "/path/to/project", "skipPermissions": true }
   },
   "platforms": {
     "telegram": { "enabled": true, "botToken": "YOUR_TELEGRAM_BOT_TOKEN" }
@@ -92,11 +98,11 @@ claude -c                       # 接上手机端的对话
 }
 ```
 
-在 **`platforms`** 下按需补充其他平台。完整模板请用 **`open-im init`**。
+完整模板请用 `open-im init`。
 
 ### Claude（Agent SDK）
 
-无需本地 `claude` 可执行文件。第三方兼容接口示例：
+无需本地 `claude` 可执行文件。支持第三方兼容接口：
 
 ```json
 {
@@ -112,26 +118,37 @@ claude -c                       # 接上手机端的对话
 }
 ```
 
-### 按平台指定 AI
+### CLI 命令
 
-在每个平台上设置 **`platforms.<name>.aiCommand`**（`claude` / `codex` / `codebuddy`）。默认 `claude`。
+| 命令 | 说明 |
+| --- | --- |
+| `open-im init` | 交互配置（不启动桥接） |
+| `open-im start` | 后台运行桥接 |
+| `open-im stop` | 停止后台服务 |
+| `open-im restart` | 重启 |
+| `open-im dashboard` | 仅 Web 配置服务（不启动桥接） |
 
-### Web 控制台
+### 环境变量
 
-`open-im start` 在 **`OPEN_IM_WEB_PORT`**（默认 **39282**）提供内置页面与 **`/api/*`**。局域网访问：`export OPEN_IM_WEB_HOST=0.0.0.0`。
+**`ANTHROPIC_*`**（shell 或 `tools.claude.env`）、**`TELEGRAM_BOT_TOKEN`**、**`OPEN_IM_WEB_PORT`**、**`OPEN_IM_WEB_HOST`**，以及各平台的 `*_APP_ID`、`*_SECRET`、`WORKBUDDY_*` 等。
 
-### 环境变量
+### Git 共同作者
 
-常见：**`ANTHROPIC_*`**（shell 或 **`tools.claude.env`**）、**`TELEGRAM_BOT_TOKEN`**、**`OPEN_IM_WEB_PORT`**、**`OPEN_IM_WEB_HOST`**，以及各平台的 `*_APP_ID`、`*_SECRET`、`WORKBUDDY_*` 等。
+默认在 AI 发起的提交里追加 `Co-authored-by`。关闭：设置环境变量 **`OPEN_IM_GIT_COAUTHOR=0`** 并重启桥接。
 
 ### 隐私
 
-为改进稳定性，可能记录**匿名**运行信息（不含聊天或 prompt 内容）。若需关闭：环境变量 **`OPEN_IM_TELEMETRY=false`**，或 **`config.json`** 中 **`"telemetry": { "enabled": false }`**。
+为改进稳定性，可能记录**匿名**运行信息（不含聊天或 prompt 内容）。若需关闭：环境变量 **`OPEN_IM_TELEMETRY=false`**，或 `config.json` 中 `"telemetry": { "enabled": false }`。
 
 ## 平台配置与故障排除
 
 详见 **[docs/platforms.zh-CN.md](./docs/platforms.zh-CN.md)**。
 
+## 环境要求
+
+- Node.js >= 20
+- 至少配置一个 IM 平台 + 所选 AI 的凭证
+
 ## License
 
 [MIT](LICENSE)

package/dist/adapters/claude-sdk-adapter.d.ts

@@ -6,7 +6,8 @@
  * - 支持 resume/cwd/model 等 options，无需 process.chdir hack
  * - SDK 内部管理 session 生命周期，无需手动维护 session pool
  */
-import type { SDKSessionInfo } from '@anthropic-ai/claude-agent-sdk';
+import { query } from '@anthropic-ai/claude-agent-sdk';
+import type { SDKSessionInfo, SessionMessage, ModelInfo, SDKControlGetContextUsageResponse, AccountInfo } from '@anthropic-ai/claude-agent-sdk';
 import type { ToolAdapter, RunCallbacks, RunOptions, RunHandle } from './tool-adapter.interface.js';
 interface ClaudeSessionMeta {
     sessionId: string;
@@ -28,6 +29,39 @@ export declare class ClaudeSDKAdapter implements ToolAdapter {
      * List sessions for a directory using the SDK's listSessions API.
      */
     static listSessionsForDir(workDir: string, limit?: number): Promise<SDKSessionInfo[]>;
+    /**
+     * Get session messages for a given session ID.
+     */
+    static getSessionMessagesForId(sessionId: string, workDir: string, limit?: number): Promise<SessionMessage[]>;
+    /**
+     * Delete a session by ID.
+     */
+    static deleteSessionById(sessionId: string, workDir?: string): Promise<boolean>;
+    /**
+     * Rename a session.
+     */
+    static renameSessionById(sessionId: string, title: string, workDir?: string): Promise<boolean>;
+    /**
+     * Fork a session.
+     */
+    static forkSessionById(sessionId: string, workDir?: string): Promise<string | undefined>;
+    /**
+     * Create a short-lived query for fetching session info (models, context, etc).
+     * The caller must close the returned query when done.
+     */
+    static createInfoQuery(workDir: string, model?: string): Promise<ReturnType<typeof query>>;
+    /**
+     * Get available models for a work directory.
+     */
+    static getSupportedModels(workDir: string, model?: string): Promise<ModelInfo[]>;
+    /**
+     * Get context usage for a work directory.
+     */
+    static getContextUsage(workDir: string, model?: string): Promise<SDKControlGetContextUsageResponse | undefined>;
+    /**
+     * Get account info for a work directory.
+     */
+    static getAccountInfo(workDir: string, model?: string): Promise<AccountInfo | undefined>;
     run(prompt: string, sessionId: string | undefined, workDir: string, callbacks: RunCallbacks, options?: RunOptions): RunHandle;
 }
 export {};

package/dist/adapters/claude-sdk-adapter.js

@@ -6,7 +6,7 @@
  * - 支持 resume/cwd/model 等 options，无需 process.chdir hack
  * - SDK 内部管理 session 生命周期，无需手动维护 session pool
  */
-import { query, listSessions } from '@anthropic-ai/claude-agent-sdk';
+import { query, listSessions, getSessionMessages, deleteSession, renameSession, forkSession } from '@anthropic-ai/claude-agent-sdk';
 import { existsSync, readFileSync, statSync, openSync, readSync, closeSync } from 'fs';
 import { execSync } from 'child_process';
 import { homedir } from 'os';
@@ -15,6 +15,17 @@ import { createLogger } from '../logger.js';
 import { refreshClaudeEnvToProcess } from '../config/file-io.js';
 import { toReplyPlainText } from '../shared/utils.js';
 const log = createLogger('ClaudeSDK');
+/**
+ * 注入交互式上下文指令，让 Claude 认为这是交互式终端会话。
+ * 解决 query() 非交互环境下 Claude 跳过用户选择、直接自主决策的问题。
+ */
+const INTERACTIVE_CONTEXT = `[SYSTEM: You are in an interactive chat session via instant messenger. The user CAN and WILL respond to your messages — treat this like an interactive terminal session. Rules:
+1. When you need to make a decision involving user preference (choosing an approach, selecting between options, deciding what to do next), ALWAYS present clearly numbered options and WAIT for the user's response.
+2. Do NOT proceed autonomously when user input would be valuable.
+3. Only proceed autonomously for obvious single-path tasks (e.g. reading a file, running a simple command).
+4. When presenting options, format them as a numbered list and end with a question.
+]
+`;
 function loadUserPluginSettings() {
     try {
         const settingsPath = join(homedir(), '.claude', 'settings.json');
@@ -199,7 +210,6 @@ export class ClaudeSDKAdapter {
      */
     static destroy() {
         // query() API 的 Query 对象通过 abortController.abort() 清理
-        // 不再需要手动管理 session pool
     }
     /**
      * List sessions for a directory using the SDK's listSessions API.
@@ -213,6 +223,138 @@ export class ClaudeSDKAdapter {
             return [];
         }
     }
+    /**
+     * Get session messages for a given session ID.
+     */
+    static async getSessionMessagesForId(sessionId, workDir, limit = 50) {
+        try {
+            return await getSessionMessages(sessionId, { dir: workDir, limit });
+        }
+        catch (err) {
+            log.warn(`Failed to get session messages for ${sessionId}: ${err}`);
+            return [];
+        }
+    }
+    /**
+     * Delete a session by ID.
+     */
+    static async deleteSessionById(sessionId, workDir) {
+        try {
+            await deleteSession(sessionId, { dir: workDir });
+            return true;
+        }
+        catch (err) {
+            log.warn(`Failed to delete session ${sessionId}: ${err}`);
+            return false;
+        }
+    }
+    /**
+     * Rename a session.
+     */
+    static async renameSessionById(sessionId, title, workDir) {
+        try {
+            await renameSession(sessionId, title, { dir: workDir });
+            return true;
+        }
+        catch (err) {
+            log.warn(`Failed to rename session ${sessionId}: ${err}`);
+            return false;
+        }
+    }
+    /**
+     * Fork a session.
+     */
+    static async forkSessionById(sessionId, workDir) {
+        try {
+            const result = await forkSession(sessionId, { dir: workDir });
+            return result.sessionId;
+        }
+        catch (err) {
+            log.warn(`Failed to fork session ${sessionId}: ${err}`);
+            return undefined;
+        }
+    }
+    /**
+     * Create a short-lived query for fetching session info (models, context, etc).
+     * The caller must close the returned query when done.
+     */
+    static async createInfoQuery(workDir, model) {
+        const resolvedModel = model?.trim() || process.env.ANTHROPIC_MODEL?.trim() || 'claude-opus-4-5';
+        return query({
+            prompt: '',
+            options: {
+                cwd: workDir,
+                model: resolvedModel,
+                permissionMode: 'default',
+            },
+        });
+    }
+    /**
+     * Get available models for a work directory.
+     */
+    static async getSupportedModels(workDir, model) {
+        let q;
+        try {
+            q = await this.createInfoQuery(workDir, model);
+            return await q.supportedModels();
+        }
+        catch (err) {
+            log.warn(`Failed to get supported models: ${err}`);
+            return [];
+        }
+        finally {
+            if (q) {
+                try {
+                    await q.return(undefined);
+                }
+                catch { /* ignore */ }
+            }
+        }
+    }
+    /**
+     * Get context usage for a work directory.
+     */
+    static async getContextUsage(workDir, model) {
+        let q;
+        try {
+            q = await this.createInfoQuery(workDir, model);
+            return await q.getContextUsage();
+        }
+        catch (err) {
+            log.warn(`Failed to get context usage: ${err}`);
+            return undefined;
+        }
+        finally {
+            if (q) {
+                try {
+                    await q.return(undefined);
+                }
+                catch { /* ignore */ }
+            }
+        }
+    }
+    /**
+     * Get account info for a work directory.
+     */
+    static async getAccountInfo(workDir, model) {
+        let q;
+        try {
+            q = await this.createInfoQuery(workDir, model);
+            return await q.accountInfo();
+        }
+        catch (err) {
+            log.warn(`Failed to get account info: ${err}`);
+            return undefined;
+        }
+        finally {
+            if (q) {
+                try {
+                    await q.return(undefined);
+                }
+                catch { /* ignore */ }
+            }
+        }
+    }
     run(prompt, sessionId, workDir, callbacks, options) {
         // 刷新 Claude 环境变量（支持 cc switch 后无需重启即可生效）
         refreshClaudeEnvToProcess();
@@ -255,12 +397,18 @@ export class ClaudeSDKAdapter {
                     }
                 }
                 const q = query({
-                    prompt,
+                    prompt: INTERACTIVE_CONTEXT + prompt,
                     options: {
                         cwd: workDir,
                         model: resolvedModel,
                         permissionMode,
+                        // 启用完整工具集，与 cc 终端一致
+                        tools: { type: 'preset', preset: 'claude_code' },
+                        // 启用所有技能（superpowers, playwright 等），与 cc 终端一致
+                        skills: 'all',
                         ...(resumeId ? { resume: resumeId } : {}),
+                        ...(options?.fallbackModel ? { fallbackModel: options.fallbackModel } : {}),
+                        ...(options?.disallowedTools?.length ? { disallowedTools: options.disallowedTools } : {}),
                         abortController,
                         stderr: (data) => {
                             recentStderr = appendStderrSnippet(recentStderr, data);

package/dist/adapters/tool-adapter.interface.d.ts

@@ -30,6 +30,10 @@ export interface RunOptions {
     hookPort?: number;
     /** Codex 专用：HTTP/HTTPS 代理地址，如 http://127.0.0.1:7890 */
     proxy?: string;
+    /** 备用模型，主模型过载时自动切换 */
+    fallbackModel?: string;
+    /** 禁用的工具列表 */
+    disallowedTools?: string[];
 }
 export interface RunHandle {
     abort: () => void;

package/dist/clawbot/client.d.ts

@@ -10,5 +10,5 @@
 import type { Config } from '../config.js';
 import type { ClawBotState } from './types.js';
 export declare function getChannelState(): ClawBotState;
-export declare function initClawbot(config: Config, eventHandler: (chatId: string, msgId: string, content: string) => Promise<void>, onStateChange?: (state: ClawBotState) => void): Promise<void>;
+export declare function initClawbot(config: Config, eventHandler: (chatId: string, msgId: string, content: string, imagePaths?: string[]) => Promise<void>, onStateChange?: (state: ClawBotState) => void): Promise<void>;
 export declare function stopClawbot(): void;