Microsoft's .prompty: a markdown file format for LLM prompts

What it does

Prompty is a .prompty file format that combines YAML frontmatter (model config, inputs, tools) with a markdown body for the actual prompt. The same file runs via VS Code extension, Python package, or TypeScript package. It handles templating (Jinja2 or Mustache), role markers (system:, user:, assistant:), and provider-specific execution for OpenAI, Microsoft Foundry, and Anthropic.

The interesting bit

The VS Code extension treats prompts as a first-class asset: live preview with template interpolation, a connections sidebar for managing endpoints, chat mode with tool calling, and .tracy trace files that expose the full pipeline timing. The runtime splits execution into explicit stages—load, render, parse, execute, process—so you can step through or invoke the whole thing in one shot.

Key highlights

• Single .prompty file travels across Python, TypeScript, and VS Code without conversion
• Variable resolution supports ${env:VAR}, ${env:VAR:default}, and ${file:path.json}
• VS Code extension includes trace viewer, live preview, and interactive chat panel
• Runtimes offer both one-shot invoke() and step-by-step load → prepare → run
• v1 files auto-migrate with deprecation warnings

Caveats

• v2 is alpha; API, format, and tooling are actively shifting
• Provider support requires separate package installs (@prompty/openai, prompty[jinja2,openai], etc.)

Verdict

Worth a look if you’re managing more than a handful of prompts across languages and want observability without building your own toolchain. Probably overkill if you’re just calling OpenAI directly from one codebase and never looking back.