- Fix adb exec-out binary pipe for screenshot capture - Add XAI, Azure, and any OpenAI-compatible endpoint support - Add rate limit retry logic - Document all supported provider configurations
Implements a computer-use agent loop for E2E testing: screenshot → vision LLM → action → repeat Supports OpenAI, Gemini, xAI via OpenAI-compatible endpoints. Inspired by openai/openai-cua-sample-app, MobileAgent, AppAgent.