huashu-mac-use skill
A skill that lets an agent drive API-less native Mac apps with background reads, focus-safe writes and proof for every step.
- 343 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent gets mouse, keyboard and window-screenshot control on a Mac, including windows with personal data.
- The skill asks for Screen Recording and Accessibility for the terminal; they apply to every process run from it.
- A wrong action in an app with no undo, such as deletion or overwrite, can change data irreversibly.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/huashu-mac-useDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent gets mouse, keyboard and window-screenshot control on a Mac, including windows with personal data.
- The skill asks for Screen Recording and Accessibility for the terminal; they apply to every process run from it.
- A wrong action in an app with no undo, such as deletion or overwrite, can change data irreversibly.
This is third-party code. Review the repository files before installing.
What it does
The `huashu-mac-use` skill picks a control channel from how the app is built. It first tries the app's own CLI, AppleScript dictionary, URL schemes and local ports (including the debug port of embedded Chromium), then the accessibility tree, window coordinates and pixels. Reads (window shots, window lists, the accessibility tree) run in the background without switching desktops. Writes go straight to the app process by default, and four checks run before anything is brought to the front. Every step is paired with a screenshot and originals are never deleted. The Swift core compiles once through `build.sh`.
Who it is for. Claude Code, Codex and Cursor users on macOS who automate apps without a public API while keeping their own work undisturbed.
Good fit when
- An agent builds a Blender scene while the user works in another window.
- You need background screenshots of progress in an app with no API.
- Before automating, you check whether the app has a CLI, port or AppleScript dictionary.
- Each agent action must be backed by a screenshot for a report.
Not a fit when
- A built-in API or CLI already covers the task.
- You want unattended clicks on payment, send and delete buttons.
- The computer is not running macOS.
Example request
Check whether System Settings can be automated and take a background window screenshot without switching my desktop.Limitations
macOS only, with Xcode Command Line Tools and Node.js, and the terminal needs Screen Recording and Accessibility permissions. The skill stops at irreversible actions, permission dialogs and the keychain, but Enter inside a terminal or IDE equals shell access. Documentation is in Chinese. For apps that cannot be captured in the background the skill borrows focus briefly. The author notes that built-in model computer use may soon cover part of it. Last commit in October 2026.
How to disable. Delete the `~/.claude/skills/huashu-mac-use` folder and revoke Screen Recording and Accessibility for the terminal in System Settings.
FAQ
What does probe.sh do?
It inspects an app before work: architecture, URL schemes, AppleScript dictionary, local ports and the number of editable elements in the accessibility tree.
Does the skill disturb the user's work?
By design, no. Reads run in the background, writes default to no focus, and when focus is borrowed it is returned within a fraction of a second.
Which actions does the skill leave to a human?
Publishing, payments, deletion, overwrite saves, permission dialogs and the keychain.
Are special macOS permissions needed?
Yes. The terminal that launches the agent needs Screen Recording and Accessibility. Switching terminals requires granting them again.
What are the four checks before taking focus?
Guard conditions before a window is brought forward, such as verifying the target app is active. The `--dry` mode shows each verdict without executing.
Discussion
No questions or reviews yet.
Related
SkillsA Swift MCP server that lets an agent drive any macOS app: click, type and read the interface tree
Mano skill
Mano skill (mano-cua)
A CLI and skill for desktop control from a text task: screenshots, clicks, typing and scrolling, in the cloud or locally on Apple Silicon
A macOS CLI and MCP server: screenshots, accessibility-based UI inspection and app control via clicks and typing
A CLI that drives a Mac to a goal from one phrase: code reads the screen, a cheap model picks actions.
A local macOS CLI that lets an agent read the screen via OCR, click and type, with no cloud and no vision model
MCP server for macOS desktop control through the Accessibility API: clicks, typing, windows, apps, shell commands and AppleScript
README in short
Four control layers: the app's structural interface, accessibility tree, window coordinates and pixels. The layer is chosen from what the preliminary probe shows.