What it does

Lathe is a Go CLI plus LLM “skills” that generate hands-on, multi-part technical tutorials from any prompt. You run /lathe build a 3D Slicer in Erlang inside Claude Code, Cursor, or Codex, then open a local web UI at localhost:4242 to read and work through the tutorial by hand. The CLI stores, serves, and manages tutorials; the LLM skills handle generation, extension, verification, and Q&A.

The interesting bit

The author explicitly built this to avoid LLMs doing the thinking for you. The UI is designed for pleasant reading, but the code is deliberately left for your fingers. Every tutorial documents its sources, model, and prompt “voice” — a rare transparency move. The author also admits he “vibecoded” the tool itself, which is either irony or honesty, possibly both.

Key highlights

• Single self-contained Go binary; installs via Homebrew (macOS), curl script, or go install
• Skills install into .claude/skills/, .cursor/commands/, or .agents/skills/ depending on your LLM tool
• Commands: /lathe, /lathe-extend, /lathe-verify, /lathe-ask, /lathe-tag
• CLI never calls an LLM directly; it hands you the exact skill command to paste, then records results via subcommands like lathe store and lathe verify-result
• Dark mode included (the README makes a point of this)

Caveats

• macOS-only for Homebrew; Linux needs the install script or go install
• Author only tests on Claude Code + macOS; other setups “should work” but are unverified
• Tutorials are explicitly not as good as human-written ones; hallucinations possible, though the author argues the risk is lower when you’re actively typing the code

Verdict

Worth a look if you learn by doing and want to explore domains with scarce human tutorials (the author’s examples: 3D slicers, embedded Zig). Skip it if you want LLMs to write the code for you, or if you need battle-tested, pedagogically reviewed content.