github-featured/wechat-gggithub--wechat-claude-code
2
0
Fork 0
Code Issues Pull Requests Deployments Models Agent Notes Packages Projects Releases Wiki Activity

docs: 重写 README，中文为默认，英文同步更新

Browse Source 
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
Wechat-ggGitHub
2026-06-10 18:02:33 +08:00
parent da02bd0fd4
commit ac3d53dcd8
3 changed files with 239 additions and 217 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+95 -84
View File
@@ -1,131 +1,142 @@
# wechat-claude-code
# WeChat Claude Code Bridge
**English** | [中文](README_zh.md)
<p align="center">
<strong>Chat with Claude Code in WeChat, just like texting a friend</strong>
</p>
A [Claude Code](https://claude.ai/claude-code) Skill that bridges personal WeChat to your local Claude Code. Chat with Claude from your phone via WeChat — text, images, slash commands, all supported.
<p align="center">
<a href="https://github.com/Wechat-ggGitHub/wechat-claude-code/blob/main/LICENSE"><img src="https://img.shields.io/badge/License-MIT-green?style=flat-square" alt="License: MIT"></a>
<a href="https://skills.sh/Wechat-ggGitHub/wechat-claude-code"><img src="https://img.shields.io/badge/skills.sh-view_page-blue?style=flat-square" alt="skills.sh"></a>
<a href="README_en.md"><img src="https://img.shields.io/badge/Lang-English-lightgrey?style=flat-square" alt="English"></a>
</p>
## Features
扫码绑定微信后，你的微信里会多出一个"好友"。给它发消息，消息会自动转发给你电脑上运行的 Claude Code，回复也会实时推送到微信。支持文字、图片、语音、文件的收发。
- **Clean output** — only Claude's text responses are shown in WeChat, tool calls run silently in the background
- **Typing indicator** — WeChat shows "typing..." while Claude is processing
- **Interrupt support** — send a new message mid-query to abort and redirect Claude
- **System prompt** — set a persistent prompt via `/prompt` (e.g. "Reply in Chinese")
- **Auto-approved tool access** — all tools run without manual approval for seamless operation
- Text conversation with Claude Code through WeChat
- Voice messages — speech-to-text automatically via WeChat
- File support — download, decrypt, and forward files for Claude to read and analyze
- Image recognition — send photos for Claude to analyze
- Slash commands — `/help`, `/clear`, `/model`, `/prompt`, `/status`, `/skills`, and more
- Launch any installed Claude Code skill from WeChat
- Cross-platform — macOS (launchd), Linux (systemd + nohup fallback)
- Session persistence — resume conversations across messages
- Rate-limit safe — proactive per-user send throttling with automatic retry
---
## Prerequisites
## 核心亮点
- Node.js >= 18
- macOS or Linux
- Personal WeChat account (QR code binding required)
- [Claude Code](https://docs.anthropic.com/en/docs/claude-code) CLI installed and authenticated
> **Note:** Claude Code supports third-party API providers (e.g. OpenRouter, AWS Bedrock, custom OpenAI-compatible endpoints) — set `ANTHROPIC_BASE_URL` and `ANTHROPIC_API_KEY` accordingly.
| | |
|---|---|
| **扫码即用** | 不用注册账号，不用部署服务器。微信扫码绑定，一分钟搞定。数据全在本地，隐私有保障。 |
| **消息不刷屏** | 只推送核心信息——进度、结果、关键决策。工具调用、中间过程等噪音自动过滤，阅读体验清爽。 |
| **"对方正在输入中..."** | Claude 在处理任务时，微信顶部会显示输入状态，随时感知它在干活。 |
| **电脑手机体验一致** | 手机端和电脑端 Claude Code 行为完全相同——同样的编排逻辑、同样的输出效果。不是两个割裂的 AI。 |
| **文件双向收发** | 发图片、Word、PDF 给 Claude 分析；Claude 生成的文件也会直接推送到微信，不用回到电脑前查看。 |
| **超时安抚** | 任务超过 5 分钟没响应？它会自动发一条消息告诉你还在干，不会让你对着空白聊天框干等。 |
## Installation
---
Clone into your Claude Code skills directory:
## 快速安装
**方式一：skills CLI（推荐）**
```bash
npx skills add Wechat-ggGitHub/wechat-claude-code
```
首次在对话中触发时，会自动克隆项目源码并安装依赖。
**方式二：手动克隆**
```bash
git clone https://github.com/Wechat-ggGitHub/wechat-claude-code.git ~/.claude/skills/wechat-claude-code
cd ~/.claude/skills/wechat-claude-code
npm install
cd ~/.claude/skills/wechat-claude-code && npm install
```
`postinstall` automatically compiles TypeScript via `tsc`.
## 快速开始
## Quick Start
### 1. Setup (first time only)
Scan QR code to bind your WeChat account:
### 1. 扫码绑定
```bash
cd ~/.claude/skills/wechat-claude-code
npm run setup
```
A QR code image will open — scan it with WeChat. Then configure your working directory.
弹出二维码，用微信扫码。
### 2. Start the daemon
### 2. 启动服务
```bash
npm run daemon -- start
```
- **macOS**: registers a launchd agent for auto-start and auto-restart
- **Linux**: uses systemd user service (falls back to nohup if systemd unavailable)
macOS 下自动注册 launchd，开机自启、崩溃自动重启。
### 3. Chat in WeChat
### 3. 开始聊天
Send any message in WeChat to start chatting with Claude Code.
打开微信，给你新出现的那个"好友"发条消息试试。
### 4. Manage the service
### 管理服务
```bash
npm run daemon -- status # Check if running
npm run daemon -- stop # Stop the daemon
npm run daemon -- restart # Restart (after code updates)
npm run daemon -- logs # View recent logs
npm run daemon -- status # 查看运行状态
npm run daemon -- stop # 停止服务
npm run daemon -- restart # 重启服务（更新代码后使用）
npm run daemon -- logs # 查看日志
```
## WeChat Commands
---
| Command | Description |
|---------|-------------|
| `/help` | Show available commands |
| `/clear` | Clear current session (start fresh) |
| `/reset` | Full reset including working directory |
| `/stop` | Stop current query and discard queued messages |
| `/model <name>` | Switch Claude model |
| `/prompt [text]` | View or set a system prompt appended to every query |
| `/status` | View current session state |
| `/cwd [path]` | View or switch working directory |
| `/skills` | List installed Claude Code skills |
| `/history [n]` | View last N chat messages |
| `/compact` | Start a new CLI session (clear token context) |
| `/undo [n]` | Remove last N messages from history |
| `/version` | Show version info |
| `/<skill> [args]` | Trigger any installed skill |
## 微信端命令
## How It Works
直接在微信聊天中发送即可：
| 命令 | 说明 |
|------|------|
| `/help` | 显示帮助 |
| `/clear` | 清除当前会话，开始新对话 |
| `/stop` | 停止当前任务 |
| `/model <名称>` | 切换 Claude 模型 |
| `/prompt <内容>` | 设置系统提示词（如"用中文回答"） |
| `/cwd <路径>` | 切换工作目录 |
| `/skills` | 查看已安装的 Skill |
| `/status` | 查看当前会话状态 |
| `/history [数量]` | 查看最近对话记录 |
| `/compact` | 压缩上下文，开始新 CLI 会话 |
| `/reset` | 完全重置（包括工作目录等设置） |
| `/undo [数量]` | 撤销最近几条对话 |
| `/<skill> [参数]` | 触发任意已安装的 Skill |
---
## 工作原理
```
WeChat (phone) ←→ ilink bot API ←→ Node.js daemon ←→ Claude Code CLI (local)
微信（手机） ←→ ilink Bot API ←→ Node.js 守护进程 ←→ Claude Code CLI（本地）
```
- The daemon long-polls WeChat's ilink bot API for new messages
- Messages are forwarded to Claude Code by spawning the `claude` CLI with `--output-format stream-json`
- Tool calls run silently in the background while Claude works
- Claude's text output is streamed to WeChat in real-time; tool calls run silently
- Responses are sent back to WeChat with automatic rate-limit retry
- Platform-native service management keeps the daemon running (launchd on macOS, systemd/nohup on Linux)
守护进程通过长轮询监听微信消息，转发给本地 `claude` CLI 处理，回复实时流式推送回微信。全程跑在你自己电脑上。
## Data
---
All data is stored in `~/.wechat-claude-code/`:
## 后续计划
- **消息队列优化** — 连续发多条指令时，回复容易串。正在研究更好的队列策略，也欢迎讨论。
- **电脑休眠不中断** — 利用 macOS 的 `caffeinate` 命令阻止系统睡眠，合上盖子也能响应微信消息。
- **接续电脑会话** — 在电脑上聊了很久，出门想接着聊。计划支持从当前电脑端的 Claude Code 会话直接续聊，工作空间和上下文保持一致。
---
## 前置条件
- Node.js >= 18
- macOS 或 Linux
- 个人微信账号
- 已安装 [Claude Code](https://docs.anthropic.com/en/docs/claude-code) CLI 并完成认证
> **提示：** Claude Code 支持第三方 API 提供商（OpenRouter、AWS Bedrock 等），设置 `ANTHROPIC_BASE_URL` 和 `ANTHROPIC_API_KEY` 即可。
## 数据目录
所有数据存储在 `~/.wechat-claude-code/`：
```
~/.wechat-claude-code/
├── accounts/ # WeChat account credentials (one JSON per account)
├── config.json # Global config (working directory, model, system prompt)
├── sessions/ # Session data (one JSON per account)
├── get_updates_buf # Message polling sync buffer
└── logs/ # Rotating logs (daily, 30-day retention)
```
## Development
```bash
npm run dev # Watch mode — auto-compile on TypeScript changes
npm run build # Compile TypeScript
├── accounts/ # 微信账号凭证
├── config.json # 全局配置
├── sessions/ # 会话数据
└── logs/ # 运行日志（每日轮转，保留 30 天）
```
## License
README_en.md
+144
View File
@@ -0,0 +1,144 @@
# WeChat Claude Code Bridge
<p align="center">
<strong>Chat with Claude Code in WeChat, just like texting a friend</strong>
</p>
<p align="center">
<a href="https://github.com/Wechat-ggGitHub/wechat-claude-code/blob/main/LICENSE"><img src="https://img.shields.io/badge/License-MIT-green?style=flat-square" alt="License: MIT"></a>
<a href="https://skills.sh/Wechat-ggGitHub/wechat-claude-code"><img src="https://img.shields.io/badge/skills.sh-view_page-blue?style=flat-square" alt="skills.sh"></a>
<a href="README.md"><img src="https://img.shields.io/badge/Lang-中文-lightgrey?style=flat-square" alt="中文"></a>
</p>
Scan a QR code to bind your WeChat, and a new "friend" appears in your contacts. Send it a message — it gets forwarded to Claude Code running on your computer, and the reply streams back to WeChat in real time. Supports text, images, voice, and files.
---
## Highlights
| | |
|---|---|
| **Scan and go** | No account signup, no server deployment. Scan a QR code and you're done in a minute. All data stays on your machine. |
| **Clean messages** | Only key info gets pushed — progress, results, key decisions. Tool calls and intermediate noise are filtered out automatically. |
| **"Typing..." indicator** | WeChat shows a typing indicator while Claude is working, so you always know it's on it. |
| **Consistent experience** | Mobile and desktop Claude Code behave identically — same orchestration, same output. Not two disconnected AIs. |
| **Two-way files** | Send images, Word docs, PDFs for Claude to analyze. Files Claude generates get pushed directly to WeChat — no need to go back to your computer. |
| **Timeout reassurance** | Task taking longer than 5 minutes? You'll get an automatic message letting you know it's still working. |
---
## Install
**Option 1: skills CLI (recommended)**
```bash
npx skills add Wechat-ggGitHub/wechat-claude-code
```
The first time you trigger the skill, it will automatically clone the source and install dependencies.
**Option 2: Manual clone**
```bash
git clone https://github.com/Wechat-ggGitHub/wechat-claude-code.git ~/.claude/skills/wechat-claude-code
cd ~/.claude/skills/wechat-claude-code && npm install
```
## Quick Start
### 1. Bind WeChat
```bash
cd ~/.claude/skills/wechat-claude-code
npm run setup
```
A QR code will pop up — scan it with WeChat.
### 2. Start the service
```bash
npm run daemon -- start
```
On macOS, this registers a launchd agent for auto-start on boot and auto-restart on crash.
### 3. Start chatting
Open WeChat and send a message to your new "friend".
### Manage the service
```bash
npm run daemon -- status # Check if running
npm run daemon -- stop # Stop the service
npm run daemon -- restart # Restart (after code updates)
npm run daemon -- logs # View recent logs
```
---
## WeChat Commands
Send these directly in the WeChat chat:
| Command | Description |
|---------|-------------|
| `/help` | Show available commands |
| `/clear` | Clear current session, start fresh |
| `/stop` | Stop current task |
| `/model <name>` | Switch Claude model |
| `/prompt <text>` | Set a system prompt (e.g. "reply in Chinese") |
| `/cwd <path>` | Switch working directory |
| `/skills` | List installed Skills |
| `/status` | View current session state |
| `/history [n]` | View recent chat history |
| `/compact` | Compact context, start a new CLI session |
| `/reset` | Full reset including working directory |
| `/undo [n]` | Remove last N messages from history |
| `/<skill> [args]` | Trigger any installed Skill |
---
## How It Works
```
WeChat (phone) ←→ ilink Bot API ←→ Node.js daemon ←→ Claude Code CLI (local)
```
The daemon long-polls WeChat for new messages, forwards them to the local `claude` CLI, and streams replies back to WeChat. Everything runs on your own machine.
---
## Roadmap
- **Message queue optimization** — Consecutive messages can produce mixed-up replies. Working on a better queuing strategy. Ideas welcome.
- **Prevent sleep** — Use macOS `caffeinate` to keep the system awake, so closing the lid doesn't interrupt the service.
- **Resume desktop session** — Chat on your computer for a while, then continue the same session from WeChat on the go. Same workspace, same context.
---
## Prerequisites
- Node.js >= 18
- macOS or Linux
- A personal WeChat account
- [Claude Code](https://docs.anthropic.com/en/docs/claude-code) CLI installed and authenticated
> **Note:** Claude Code supports third-party API providers (OpenRouter, AWS Bedrock, etc.) — set `ANTHROPIC_BASE_URL` and `ANTHROPIC_API_KEY` accordingly.
## Data Directory
All data is stored in `~/.wechat-claude-code/`:
```
~/.wechat-claude-code/
├── accounts/ # WeChat account credentials
├── config.json # Global config
├── sessions/ # Session data
└── logs/ # Rotating logs (daily, 30-day retention)
```
## License
[MIT](LICENSE)
README_zh.md
-133
View File
@@ -1,133 +0,0 @@
# wechat-claude-code
[English](README.md) | **中文**
一个 [Claude Code](https://claude.ai/claude-code) Skill，将个人微信桥接到本地 Claude Code。通过手机微信与 Claude 对话——文字、图片、斜杠命令，全部支持。
## 功能特性
- **纯净输出** — 微信只展示 Claude 的文字回答，工具调用在后台静默执行
- **输入指示器** — Claude 处理时微信顶部显示"正在输入..."
- **中断支持** — 在 Claude 处理中发送新消息可打断当前任务
- **系统提示词** — 通过 `/prompt` 设置持久化提示词（如"用中文回答"）
- **工具自动执行** — 所有工具无需手动审批，无缝运行
- 通过微信与 Claude Code 进行文字对话
- 语音消息——自动转文字（微信自带语音识别）
- 文件支持——下载解密文件，让 Claude 读取并分析
- 图片识别——发送照片让 Claude 分析
- 斜杠命令——`/help`、`/clear`、`/model`、`/prompt`、`/status`、`/skills` 等
- 在微信中触发任意已安装的 Claude Code Skill
- 跨平台——macOS（launchd）、Linux（systemd + nohup 回退）
- 会话持久化——跨消息恢复上下文
- 限频保护——per-user 主动限速 + 自动重试
## 前置条件
- Node.js >= 18
- macOS 或 Linux
- 个人微信账号（需扫码绑定）
- 已安装 [Claude Code](https://docs.anthropic.com/en/docs/claude-code) CLI 并完成认证
> **注意：** Claude Code 支持第三方 API 提供商（如 OpenRouter、AWS Bedrock、自定义 OpenAI 兼容接口）——按需设置 `ANTHROPIC_BASE_URL` 与 `ANTHROPIC_API_KEY` 即可。
## 安装
克隆到 Claude Code skills 目录：
```bash
git clone https://github.com/Wechat-ggGitHub/wechat-claude-code.git ~/.claude/skills/wechat-claude-code
cd ~/.claude/skills/wechat-claude-code
npm install
```
`postinstall` 脚本会自动编译 TypeScript。
## 快速开始
### 1. 首次设置
扫码绑定微信账号：
```bash
cd ~/.claude/skills/wechat-claude-code
npm run setup
```
会自动弹出二维码图片，用微信扫码后配置工作目录。
### 2. 启动服务
```bash
npm run daemon -- start
```
- **macOS**：注册 launchd 代理，实现开机自启和自动重启
- **Linux**：使用 systemd 用户服务（无 systemd 时回退到 nohup）
### 3. 在微信中聊天
直接在微信中发消息即可与 Claude Code 对话。
### 4. 管理服务
```bash
npm run daemon -- status # 查看运行状态
npm run daemon -- stop # 停止服务
npm run daemon -- restart # 重启服务（代码更新后使用）
npm run daemon -- logs # 查看最近日志
```
## 微信端命令
| 命令 | 说明 |
|------|------|
| `/help` | 显示帮助 |
| `/clear` | 清除当前会话（重新开始） |
| `/reset` | 完全重置（包括工作目录等设置） |
| `/stop` | 停止当前查询并清空排队消息 |
| `/model <名称>` | 切换 Claude 模型 |
| `/prompt [内容]` | 查看或设置系统提示词（全局生效） |
| `/status` | 查看当前会话状态 |
| `/cwd [路径]` | 查看或切换工作目录 |
| `/skills` | 列出已安装的 Claude Code Skill |
| `/history [数量]` | 查看最近 N 条对话记录 |
| `/compact` | 压缩上下文（开始新 CLI 会话，保留历史） |
| `/undo [数量]` | 撤销最近 N 条对话 |
| `/version` | 查看版本信息 |
| `/<skill> [参数]` | 触发任意已安装的 Skill |
## 工作原理
```
微信（手机） ←→ ilink bot API ←→ Node.js 守护进程 ←→ Claude Code CLI（本地）
```
- 守护进程通过长轮询监听微信 ilink bot API 的新消息
- 消息通过 spawn `claude` CLI（`--output-format stream-json`）转发给 Claude Code
- 工具调用在 Claude 工作时后台静默执行
- Claude 的文字输出实时流式发送到微信，工具调用静默执行
- 回复发送回微信，限频时自动重试
- 平台原生服务管理保持守护进程运行（macOS 使用 launchd，Linux 使用 systemd/nohup）
## 数据目录
所有数据存储在 `~/.wechat-claude-code/`：
```
~/.wechat-claude-code/
├── accounts/ # 微信账号凭证（每个账号一个 JSON）
├── config.json # 全局配置（工作目录、模型、系统提示词）
├── sessions/ # 会话数据（每个账号一个 JSON）
├── get_updates_buf # 消息轮询同步缓冲
└── logs/ # 运行日志（每日轮转，保留 30 天）
```
## 开发
```bash
npm run dev # 监听模式——TypeScript 文件变更时自动编译
npm run build # 编译 TypeScript
```
## License
[MIT](LICENSE)