Skip to main content
WebViews, Flutter/React Native canvases, and custom-drawn controls often expose weak or empty accessibility trees.

Prefer vision

  • Autonomous Yoqa runs already use screenshots — ideal for these UIs.
  • For CLI agents, lean on yoqa screenshot + -d grounding when yoqa screen is empty or misleading.

Describe visually

Name what a human would see: color, icon shape, relative position (“blue circular FAB bottom-right”), not internal view IDs.

Hybrid screens

If only part of the screen is native, still write goal-level instructions; add detail only for the custom region that the agent mis-taps. Related: Games.