
This gives Claude direct control over Ubuntu desktop applications through AT-SPI accessibility APIs with OCR fallback, letting you click buttons and interact with UI elements by their semantic labels instead of brittle pixel coordinates. The main use case is automating browser extensions and native desktop apps that Playwright can't reach, like MetaMask popups or Phantom wallet flows. It includes wait utilities with exponential backoff and click verification. Requires X11 and a handful of system packages (xdotool, tesseract, at-spi2-core), so you'll need to run the install script first. The semantic approach is more maintainable than screenshot-based automation, though you can still drop down to raw coordinates when needed.
npx -y skills add lommaj/ubuntu-desktop-control --skill ubuntu-desktop-control --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.