Reverse engineered prompt

Build me a project called POML, a simple markup language for creating structured prompts for AI models. I want it to feel like HTML for prompts, so people can write things like roles, tasks, examples, output format, and also include outside content like documents, tables, and images. It should support templating too, with variables, basic conditionals, loops, and reusable definitions so prompts can be generated from data.

Please make it useful in real life, not just a parser. I want a VS Code extension that gives syntax highlighting, autocomplete, hover help, live preview, error checking, and a way to test prompts against a configured model provider. I also want SDKs so people can render and use these prompt files from both JavaScript or TypeScript and Python.

Include solid docs, examples, and a few sample prompts that show image input, structured examples, and different formatting styles. Keep the language maintainable and modular, with content separated from presentation. If you need details, look up the current docs and common editor extension patterns online.

Want more depth? Deep Reverse