openqa-labs/openqa ? reverse-engineered prompt

Build me an open source agentic testing harness for web apps that lets me write tests in plain English instead of selectors or step definitions.

I want a tool I can initialize in an existing project with one command, then create feature files in Gherkin style and run them against a real browser. The agent should read each step, figure out what to click or type, and use the same page the test is using. It should work with either Claude Code or OpenCode, use the user’s existing login session locally, and support CI with API keys if needed.

Please include a clean setup flow, example feature files, environment variable handling for things like base URL and credentials, and good test outputs like screenshots, HTML reports, and traces. Make it easy to customize the harness inside its own folder, and make sure it works well with Playwright and can run in headless mode or visibly for debugging. If you need current docs for the agent or browser tools, look them up online.