microsoft/Webwright — reverse-engineered prompt

Build me a lightweight Python browser agent framework called Webwright.

I want it to let a coding model use a terminal to write and rerun Python scripts that open one or more Chromium browser sessions, inspect pages only when needed, and complete real web tasks from start to finish. Every run should leave behind a single reusable script, plus logs, screenshots, and other artifacts on disk so it is easy to debug, replay, and improve. Please add a simple CLI, support model backends for OpenAI, Anthropic, and OpenRouter, and keep the whole thing small and easy to read instead of hiding the logic inside a heavy framework.

It should also be usable from coding agents like Claude Code or Codex through skills or plugin style integration. If possible, include a mode that takes the finished task result and renders it as a small HTML app someone can view and reuse. Add a few tests and a clear README with setup and example tasks. Look up current docs online if you need to.