What problem does it solve?
Controlling a remote Linux desktop through computer_use_remote is unreliable when relying on pixel clicks alone, especially under Wayland where focus and input routing behave differently. This Skill provides backend-specific guidance for using AT-SPI structural targeting, verifying actions with screenshots, and handling Linux-specific permission and focus failures.
Core Features & Use Cases
- AT-SPI Structural Targeting: Use
ax_snapshot to inspect the accessibility tree and ax_action to press, focus, or set values on named buttons, menus, and text fields instead of guessing coordinates.
- Wayland Input Handling: Apply target-guarded text injection, verified window focus via
element_action, and screenshot proof after every state-changing action.
- Permission and Fallback Rules: Detect errors like
COMPUTER_USE_AX_UNAVAILABLE or COMPUTER_USE_REARM_REQUIRED and stop for user re-arming, with a defined fallback order from app-native tooling to keyboard paths to coordinate clicks.
- Use Case: After
host-computer-use reports a Linux/Wayland backend with AT-SPI features, use this Skill to click a dialog's OK button by role and title, then confirm the result with a fresh screenshot before typing into the focused window.
Quick Start
After the connected A0 CLI reports a Linux Wayland backend with AT-SPI features, load this skill and use ax_snapshot followed by ax_action to operate the target window, verifying each step with a screenshot.
