POML (Prompt Orchestration Markup Language) (microsoft.github.io)

POML (Prompt Orchestration Markup Language) is a newly documented, HTML-like markup for building modular, data-aware prompts for large language models. It standardizes prompt components (e.g., <role>, <task>, <example>) and adds specialized data tags (<document>, <table>, <img>) so prompts can embed or reference external sources like text files, spreadsheets, and images. By separating presentation from content with a CSS-like <stylesheet> system, POML explicitly targets LLM format sensitivity—letting developers tune verbosity, syntax, or output format without changing core logic—and supports an internal templating engine ({{variables}}, loops, conditionals, <let>) to generate data-driven prompts programmatically. The release is notable because it pairs expressive syntax with practical tooling: a VS Code extension (syntax highlighting, autocomplete, previews, diagnostics, interactive testing) plus Node.js and Python SDKs for easy integration into model pipelines. The project also links to research on joint content/format optimization (arXiv papers) and has nascent community implementations (Rust renderer, Ruby gem) and a Discord. For practitioners, POML promises more maintainable, testable, and reproducible prompt engineering workflows, smoother data integration, and a pathway to systematically optimize both prompt content and formatting across LLM applications.