AI Skill MartBrowse skillsPromptsProgramsLearnWhat is a Skill?
 
← Back to list

huashu-mac-use (macOS Native App Automation)

Drives GUI-only macOS apps that have no API or CLI without stealing your focus, and leaves a reproducible screenshot trail for every step.

AutomationAdvanced★ 338⑂ 51AI score 9/10Last updated: Nov 7, 2026

What it does

  • Operates GUI-only apps: clicks menus, fills inputs, reads the UI and switches windows in native macOS apps that expose neither a CLI nor an API (desktop AI clients, closed-source tools).
  • Four-layer routing: L0 structural interfaces (CDP, AppleScript dictionaries, URL schemes, local ports) → L1 Accessibility semantics via agent-desktop → L2 window coordinates → L3 raw pixels, picking the right channel for your actual desktop topology.
  • Focus-protecting gates: writes default to headless; borrowing focus must clear three gates — terminal-window block, a machine-wide focus lock, and "if the user touched keyboard/mouse in the last 2s, wait instead of grabbing".
  • Real verification: a tool returning success is not proof; it reads state back, screenshots, checks app state indicators and side effects, and captures per-window background screenshots as forensic evidence.
  • Hard stop lines: irreversible actions (publish, pay, delete, OAuth consent), Enter in terminals/IDEs, system permission dialogs, and banking/medical apps are handed back to the user.

Who it's for

  • Anyone who must repeatedly operate a Mac app with no automation API
  • Engineers capturing clean per-window screenshots for bug reports or docs
  • Multi-Space users who don't want an agent yanking focus mid-task
  • People running the same prompt across models inside a desktop client and keeping evidence

Examples

  1. "Send this question in my Electron-based desktop client and capture the answer" → probe.sh detects the architecture → mac open --cdp 9333 → cdp.js insert/press → cdp.js shot.
  2. "Grab a screenshot of an app window sitting on another desktop" → mac windows <app> --all shows on=0 → mac shot captures it in the background with zero focus change.
  3. "Can this unfamiliar app be automated at all?" → probe.sh <app> reports bundle, architecture, URL scheme, real sdef, local ports and AX control count.

· · · Install guide · · ·

Try it now, no install

Paste this into Claude to use the skill without installing anything.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/alchaincyf/huashu-mac-use/HEAD/SKILL.md

What I want: (describe your task here)

If Claude can't open the link, open it yourself and paste the contents instead.

↓ If it works for you, download the ZIP below and install it. Then it runs on its own — no pasting each time.

Install in the Claude app (no terminal)
  1. Download the ZIP with the button below.
  2. In Claude, open Settings → Capabilities and turn on 'Code execution and file creation'. (one time)
  3. Go to Customize → Skills → + → 'Upload a skill' and upload the ZIP.
↓ Download ZIP
Install in Claude Code

Let Claude do it — paste this into Claude Code

Install the skill I found on AI Skill Mart.
Copy the SKILL.m folder from the GitHub repo alchaincyf/huashu-mac-use into my ~/.claude/skills/huashu-mac-use/.
When it's done, tell me in one line what this skill can do.

Install with a command instead

git clone https://github.com/alchaincyf/huashu-mac-use.git ~/.claude/skills/huashu-mac-use && bash ~/.claude/skills/huashu-mac-use/scripts/build.sh

⚠ This is a third-party skill. Check the source repository before installing.

  1. Open Terminal on macOS and confirm Git and Node.js are available: git --version, node -v.
  2. Clone the repo into your skills folder: git clone https://github.com/alchaincyf/huashu-mac-use.git ~/.claude/skills/huashu-mac-use
  3. Build the kernel once (~20 seconds): bash ~/.claude/skills/huashu-mac-use/scripts/build.sh — it also reports whether agent-desktop is installed.
  4. (Recommended) Install the semantic driver: npm install -g agent-desktop. Without it the skill falls back to its own AX and coordinate paths.
  5. In System Settings → Privacy & Security, grant your terminal app both Accessibility and Screen Recording. Restart the terminal after granting permissions for the first time.
  6. Restart Claude Code and try a request such as "screenshot this Mac app's window" or "type this text into that app's input field".
  7. For any write action, run it with --dry first to see how coordinates are resolved and how each safety gate rules.
View source on GitHub ↗License: MIT