mishushakov/llm-scraper — reverse-engineered prompt

Build me a TypeScript library that can open any webpage in a real browser session and turn what it sees into clean structured data with an LLM. I want to give it a page plus a schema and get back parsed results I can trust, with good TypeScript types. It should work with popular model providers like OpenAI, Anthropic, Google, local models, and other compatible APIs, and it should be easy to swap between them.

Please make it flexible about what gets sent to the model, like processed HTML, raw HTML, markdown, plain text, screenshots for vision models, or fully custom content. I also want a streaming option so partial objects can come back as they are generated, plus a way to generate reusable browser side scraping code from the schema.

Use browser automation for the page loading part, include a solid example like pulling top stories from a news site, and add tests and simple docs so someone can install it and use it quickly. Look up current docs online if you need to.