Skip to content
TopicTracker
From HackerNewsView original
TranslationTranslation

Show HN: Margarita - Programming Agents with Markdown

Margarita is a programming language that combines markdown with logical operators, enabling deterministic code structures alongside LLM dynamic code. It also supports composable prompts inspired by React, addressing composability issues in AI workflow skill libraries. The project is open source on GitHub.

Background

Margarita is a new programming language (creator's "build it from scratch" project) that lets developers write AI agent workflows using Markdown — the familiar plain-text formatting syntax used in apps like Notion, GitHub READMEs, and chat platforms. Its core pitch: Markdown is already used to describe step-by-step AI workflows, but those files become brittle and repetitive at scale. Margarita adds logical operators (if/else, loops, variables) directly into Markdown, so you get deterministic control where you want it and fall back to LLM-powered "dynamic code" where flexibility helps. It also treats prompts like React components — composable and reusable, avoiding copy-paste across hundreds of files. The author built it primarily as a learning exercise and to address frustrations with existing "skills libraries" for AI workflows, but notes lukewarm reception.