Skip to main content
Games often lack a useful accessibility tree. Yoqa’s agent is a strong fit because autonomous runs are vision-first.

Put rules in app context

Describe once in app context:
  • Win / lose / pause conditions
  • What important HUD controls look like
  • Gestures that are easy to misread (hold vs tap)
Keep case instructions at goal level: “Complete level 1” rather than pixel-perfect move lists — refine only when the agent drifts.

Prefer screenshots over trees

For coding-agent debug loops, yoqa screen may be sparse. Fall back to yoqa screenshot and description grounding more often than in native form-based apps.

Timing

Games animate quickly. Expect occasional misses on short-lived UI; add explicit waits in instructions (“wait until the reward chest appears”) or break flows into smaller goals. See also Non-native UI.