The current workflow for AI-assisted scaffolding is broken. LLMs are excellent at planning architectures (outputting ASCII trees) but terrible at executing them (outputting shell scripts that often fail).

There was no standard "receiver" for that visual output, so I built one.

tree-fs parses text-based trees directly into filesystem structures. It handles the messy reality of LLM outputs-inconsistent indentation, inline comments, and mixed file/folder notations.

It’s designed to be a deterministic, zero-dependency primitive that you can pipe text into.