feat(mac-desktop-autopilot): add macOS desktop automation skill

README.md

@@ -27,6 +27,7 @@ Development skills for AI coding agents. Plug into your favorite AI coding tool
 | `minimax-music-gen` | Generate vocal songs, instrumentals, and covers using MiniMax Music API. Two modes: Basic (one-liner in, song out) and Advanced Control (edit lyrics, refine prompt, plan structure). Supports lyrics generation, style vocabulary, streaming playback, and iterative feedback. | Official |
 | `buddy-sings` | Let your Claude Code pet (/buddy) sing a personalized song. Interprets the pet's name and personality into a unique cached vocal identity, auto-gathers context (conversation, memory, git history) for themed lyrics, and generates music via minimax-music-gen. | Official |
 | `minimax-music-playlist` | Generate personalized playlists by analyzing your music taste. Builds a taste profile (genre, mood, language, vocal preferences), plans a themed tracklist, generates songs with album cover art, and refines the profile from feedback. | Official |
+| `mac-desktop-autopilot` | macOS local desktop automation: screenshot capture, mouse/keyboard control via pyautogui, and file upload to CDN for phone sharing. Triggers: 截图、截屏、截个图、控制电脑、发文件到手机。 | Community |
 
 ## Installation
 

README_zh.md

@@ -27,6 +27,7 @@
 | `minimax-music-gen` | 使用 MiniMax Music API 生成人声歌曲、纯音乐和翻唱。支持基础模式（一句话生成）和强控制模式（编辑歌词、调整 prompt、规划曲式）。内置歌词生成、风格词表、流式播放和迭代反馈。 | Official |
 | `buddy-sings` | 让你的 Claude Code 宠物（/buddy）唱一首专属歌曲。根据宠物名字和个性生成独特声线并缓存，自动采集上下文（对话、记忆、git 历史）生成主题歌词，调用 minimax-music-gen 完成创作。 | Official |
 | `minimax-music-playlist` | 分析用户音乐品味生成个性化歌单。构建音乐画像（曲风、情绪、语言、声线偏好），规划主题曲目，生成歌曲与专辑封面，根据反馈持续优化画像。 | Official |
+| `mac-desktop-autopilot` | macOS 本地桌面自动化：截图、鼠标键盘控制（pyautogui）、文件上传到 CDN 分享到手机。触发词：截图、截屏、控制电脑、发文件到手机。 | Community |
 
 ## 安装
 

skills/mac-desktop-autopilot/SKILL.md

@@ -0,0 +1,127 @@
+---
+name: mac-desktop-autopilot
+description: |
+  macOS local desktop automation: screenshot capture, mouse/keyboard control, and file upload to CDN for phone sharing.
+  Use when: user asks to take a screenshot, capture screen, control the desktop, click a button, open an app, see what's on screen,
+  or send a file from computer to phone. Trigger phrases: 截图、截屏、截个图、控制电脑、操控电脑、发文件到手机、点一下、帮我打开xxx、看看屏幕现在是什么。
+  Do NOT use for: web browser automation (use Playwright MCP instead) or remote desktop (SSH/VNC).
+metadata:
+  version: "1.0"
+  category: productivity
+  license: MIT
+  platform: macOS
+  required_permissions:
+    - Screen Recording
+    - Accessibility
+  dependencies:
+    - pyautogui
+    - screencapture (built-in macOS)
+  sources:
+    - macOS built-in screencapture
+    - pyautogui (pip)
+---
+
+# macOS Desktop Autopilot
+
+Local macOS desktop automation via `screencapture` (screenshot) + `pyautogui` (mouse/keyboard) + Matrix CDN (file sharing).
+
+## Prerequisites
+
+1. **Screen Recording permission**: System Settings → Privacy & Security → Screen Recording → authorize Terminal or MiniMax Agent
+2. **Accessibility permission**: System Settings → Privacy & Security → Accessibility → authorize Terminal or MiniMax Agent
+3. **pyautogui installed**: `pip3 install pyautogui --default-timeout=120`
+
+## Procedure
+
+### Screenshot
+
+```bash
+screencapture /tmp/screenshot.png
+```
+
+After capturing, use `describe_images` tool to read and describe the screenshot. For phone sharing, upload to CDN.
+
+### Mouse Operations
+
+```python
+import pyautogui, time
+pyautogui.FAILSAFE = True  # move to corner to abort
+
+# Move to coordinate (pixels, origin = top-left)
+pyautogui.moveTo(x, y, duration=0.3)
+
+# Left-click
+pyautogui.click(x, y)
+
+# Double-click
+pyautogui.doubleClick(x, y)
+
+# Right-click
+pyautogui.rightClick(x, y)
+
+# Drag
+pyautogui.dragTo(x2, y2, duration=0.5)
+
+# Scroll
+pyautogui.scroll(-3)  # scroll down 3 units
+```
+
+> Why FAILSAFE: moving the cursor to any screen corner triggers an emergency stop to prevent runaway automation.
+
+### Keyboard Operations
+
+```python
+# Type text (cursor must already be in text field)
+pyautogui.typewrite("Hello", interval=0.05)
+
+# Press a key
+pyautogui.press("enter")
+pyautogui.press("escape")
+pyautogui.press("cmd", "s")  # Cmd+S
+
+# Combo keys
+pyautogui.hotkey("cmd", "c")  # copy
+pyautogui.hotkey("cmd", "v")  # paste
+pyautogui.hotkey("cmd", "a")  # select all
+```
+
+### Send File to Phone
+
+1. Upload file via `matrix_upload_to_cdn` → get CDN URL
+2. Share the URL with user → download on phone
+
+### OCR / Read Screen Content
+
+```bash
+screencapture /tmp/ocr_screen.png
+```
+Then analyze with `describe_images` tool to extract text.
+
+## Output Contract
+
+- **Screenshot**: display image directly or share CDN link
+- **Mouse/keyboard**: take a confirmation screenshot immediately after action
+- **File upload**: return CDN download link
+
+## Failure Handling
+
+| Error | Cause | Fix |
+|-------|-------|-----|
+| "could not create image" | Missing Screen Recording permission | User authorizes in System Settings |
+| pyautogui runaway | FAILSAFE not enabled | Move cursor to screen corner to stop |
+| Permission denied | Missing Accessibility permission | User authorizes in System Settings → Privacy → Accessibility |
+| pyautogui not installed | pip install failed | Retry `pip3 install pyautogui --default-timeout=120` |
+
+## Examples
+
+**Screenshot to phone**
+User: "截个图发给我"
+→ `screencapture /tmp/page.png` → `matrix_upload_to_cdn(/tmp/page.png)` → return download link
+
+**Open an app**
+User: "帮我打开微信"
+→ Take screenshot → OCR locate Dock → click WeChat icon
+
+**What's on screen**
+User: "现在屏幕是什么"
+→ Screenshot → `describe_images` analyze → describe screen content