Summary

• Adds a new hyprland skill for controlling the Hyprland Wayland desktop and interacting with running applications on Linux
• Adds the skill to the README under Development & Code Tools in alphabetical order

What problem it solves

On a Linux Wayland desktop running Hyprland, Claude can now autonomously interact with any GUI application — clicking buttons, typing text, reading screen content via screenshots, and managing windows/workspaces. This is useful for automating GUI workflows, sending messages in IM apps, browser automation, and general desktop control.

Who uses this workflow

Linux developers and power users running Hyprland as their Wayland compositor who want Claude to help automate desktop tasks.

How it works

The skill uses a layered interaction strategy:

1. AT-SPI2 (accessibility API) for supported apps like Firefox and Chrome — semantic, no coordinates needed
2. ydotool for universal mouse/keyboard input across Wayland and XWayland apps
3. xdotool for XWayland-specific fallback
4. Screenshot + visual analysis as the final fallback

It also maintains a per-app knowledge table that gets updated after each task, so the right method is used immediately next time.

Example

Send a message to Alice in Slack saying "I'll be 5 minutes late"

Claude will: check Slack's AT-SPI2 tree → find the message input → type and send, or fall back to screenshot-based interaction if needed.