What problem does it solve? When no purpose-built tool, API, or MCP exists for a task, automation stalls. This Skill lets an AI operate the computer the way a person does — capturing the screen, locating targets visually, and acting through mouse and keyboard — so canvas apps, games, remote desktops, and installers remain automatable. ## Core Features & Use Cases - Cross-platform input scripts: Unified command interface (capture, crop, move, click, drag, scroll, type, key) backed by native tooling on Windows, WSL, and Linux (X11/Wayland via spectacle, grim, xdotool, ydotool, wtype). - Capture-locate-act-verify loop: Every action is verified with a fresh screenshot, with crop-based zooming for targets smaller than ~40px and explicit stop conditions after repeated misses. - Safety guardrails: Confirmation before first capture, non-destructive typing rules, and mandatory stops before deleting, overwriting, or acting on payment and account surfaces. - Use Case: Automate a GUI-only installer on a Linux desktop: capture the screen, locate the Next button, click it, and verify each dialog advances — no accessibility APIs required. ## Quick Start Ask the agent to open an application and complete a GUI task using the computah desktop-control skill, confirming the first screen capture.
