mac-use
由 kekejun 提供
Browser & Automation简介
Control macOS GUI apps visually — take screenshots, click, scroll, type.
描述
该技能的功能
Controls macOS GUI applications through a screenshot-then-interact loop. Takes annotated screenshots using Apple Vision OCR to detect and number text elements, then clicks, types, scrolls, or presses keys based on what's visible on screen. Works with any app that has a graphical interface.
Unlike AppleScript or accessibility APIs, this works on any app regardless of whether it exposes automation hooks, since it operates purely on what's visible on screen.
何时使用
- Automating repetitive clicks in desktop apps without scripting APIs
- Filling out forms in apps that lack command-line access
- Navigating WeChat mini-programs or other sandboxed GUI environments
- Scraping data visible on screen from apps with no export option
- Testing desktop app UI flows end-to-end
示例工作流
User asks: open Safari, go to example.com, and click the first link
- 1
Run
open -a Safariand wait 2 seconds - 2
Run screenshot command on Safari, read annotated image at /tmp/mac_use.png and element list JSON
- 3
Identify address bar element number from OCR output, click it with clicknum
- 4
Type the URL and press return key, wait for page to load
- 5
Screenshot again, identify first link by element number, click it with clicknum
Safari navigates to example.com and opens the first link on the page
安装
clawhub install kekejun/mac-use运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。