微信AI助手 — 主动发消息 + 自动回复 / WeChat AI assistant — send messages & auto-reply
---
name: reply-wechat-message
slug: wechat-butler
displayName: 微信管家 / WeChat Butler
description: "微信AI助手 — 主动发消息 + 自动回复 / WeChat AI assistant — send messages & auto-reply"
agent_created: true
---
# 微信管家 / WeChat Butler
微信AI助手 — 主动发消息给联系人 + 收到消息后AI自动回复。
WeChat AI assistant — proactively send messages, and auto-reply when messages come in.
**自包含技能包** — 所有依赖脚本(启动微信、发送消息)已打包在内,无需额外安装。
**Self-contained skill** — all dependency scripts (launch WeChat, send messages) are bundled. No extra installation needed.
---
## 功能一:AI 主动发消息 / Feature 1: Send Message
直接发送一条消息给指定联系人(不读取聊天上下文,即时发送)。
Send a message directly to a contact (no context reading, instant send).
### 触发格式 / Trigger Format
**中文 / Chinese:**
```
给 [联系人] 发消息:[内容]
发消息给 [联系人]:[内容]
帮 [联系人] 发消息:[内容]
```
**English:**
```
send [contact] [message]
msg [contact] [message]
message [contact] [message]
```
### 示例 / Examples
**中文 / Chinese:**
- `给 小明 发消息:中午一起去吃饭吗?`
- `发消息给 小红:记得带文件`
- `帮 小张 发消息:生日快乐!`
**English:**
- `send Kitty: Want to grab lunch?`
- `msg Peter: Don't forget the documents`
- `message Tom: Happy birthday!`
### AI 执行步骤 / Execution Steps
```
第1步 / Step 1:
AI 提取联系人和消息内容
AI extracts contact name and message content
第2步 / Step 2:
python scripts/send_wechat.py <联系人/contact> <内容/message>
→ 打开微信 → 搜索联系人 → 打开聊天 → 发送消息
→ Open WeChat → search contact → open chat → send message
第3步 / Step 3:
返回发送结果给用户
Report send result to user
```
---
## 功能二:AI 自动回复 / Feature 2: AI Auto-Reply
读取聊天上下文,AI 分析后自动回复。
Reads the chat context, AI analyzes it, then auto-replies.
### 触发格式 / Trigger Format
```
使用技能:reply-wechat-message 给:[联系人] 回复
```
### 示例 / Examples
- `使用技能:reply-wechat-message 给:小明 回复` (Xiao Ming / Kitty)
- `使用技能:reply-wechat-message 给:小红 回复` (Xiao Hong / Peter)
### 简化触发 / Short Forms
- 回复 小明 / reply Kitty
- 回 小红 / reply Peter
---
## AI 工作流程(自动回复)/ AI Workflow (Auto-Reply)
```
第1步 / Step 1:
python scripts/reply_wechat.py 小明 / python scripts/reply_wechat.py Kitty
→ 读取整个聊天区域,区分左右两侧
→ Read entire chat area, split left/right
→ 输出结构化对话 / Output structured conversation:
★上下文开始★ / ★Context Start★
[对方/Them] xxx
[我/Me] xxx
[对方/Them] xxx
★上下文结束★ / ★Context End★
第2步 / Step 2:
AI分析全部上下文,理解对话脉络,生成合适的回复
AI analyzes full context, understands the conversation flow, generates a suitable reply
⚠️ 重要规则 / Important Rule:
OCR识别出的聊天内容即为真实上下文,模型不得质疑、猜测或怀疑识别结果的准确性。
The OCR-recognized text IS the real context. The model must NOT question, guess,
or doubt the accuracy of the recognition. Reply based on the recognized content directly.
第3步 / Step 3:
echo "AI生成的回复 / AI-generated reply" | python scripts/reply_wechat.py 小明
→ 或 / Or: echo "AI-generated reply" | python scripts/reply_wechat.py Kitty
→ 读取上下文 + 自动发送回复(一步完成,stdin管道无引号问题)
→ Read context + send reply in one step (stdin piping avoids quote issues)
```
---
## 脚本 / Scripts
`scripts\send_wechat.py` — 发送脚本:搜索联系人 + 发送消息 / Send script: search contact + send message
`scripts\reply_wechat.py` — 主脚本:读取上下文 + 发送回复 / Main script: read context + send reply
`scripts\open_wechat.py` — 启动脚本:唤醒微信窗口 / Launch script: bring WeChat to foreground
### send_wechat.py 用法 / Usage
```bash
# 发送消息 / Send message (command line arg)
python scripts/send_wechat.py <小明/Kitty> <消息/message>
# 发送消息(stdin管道,无引号问题)/ Send message (stdin pipe, no quote issues)
echo "消息/message" | python scripts/send_wechat.py <小明/Kitty>
```
### reply_wechat.py 用法 / Usage
```bash
# 只读取上下文(AI分析用)/ Read context only (for AI analysis)
python scripts/reply_wechat.py <小明/Kitty>
# 读取上下文 + 发送回复(一步到位)/ Read context + send reply (one step)
echo "回复内容/reply text" | python scripts/reply_wechat.py <小明/Kitty>
```
### 输出格式 / Output Format
```
★上下文开始★ / ★Context Start★
[对方/Them] 你吃饭了吗 / Have you eaten?
[我/Me] 吃过了,你呢 / Yes, and you?
[对方/Them] 我也吃了 / Me too
★上下文结束★ / ★Context End★
```
---
## 消息区分逻辑 / Message Detection Logic
- **左侧(白底黑字)/ Left side (white bg, black text)** = 对方发的消息 / Messages from the other party → 标注 `[对方/Them]`
- **右侧(绿底黑字)/ Right side (green bg, black text)** = 自己发的消息 / Messages from yourself → 标注 `[我/Me]`
- 截图聊天区从中线切开,左右分别OCR,避免颜色误判
Screenshot is split at the center line; left and right are OCR'd separately to avoid color misidentification
---
## 依赖 / Dependencies
- Python 3.10+
- OCR.space API(免费版,无需注册 / Free tier, no registration required)
- Python packages: pyautogui, pygetwindow, pyperclip, requests, Pillow, numpy
---
## 脚本清单 / Script Inventory
| 文件 / File | 功能 / Function |
|---|---|
| `send_wechat.py` | 搜索联系人 + 发送消息 / Search contact & send message |
| `reply_wechat.py` | 读取上下文 + 发送回复 / Read context & send reply |
| `open_wechat.py` | 启动微信 / Launch WeChat |
don't have the plugin yet? install it then click "run inline in claude" again.