Tuipo — Development skill for Claude Code

tuipo

**Grammarly for your terminal.**

The red squiggly line from your editor — now for the prompts you type into Claude Code, Codex, Aider, and any other terminal app.

[](#license) [](#install) [](https://crates.io/crates/tuipo)

tuipo catching typos in a prompt to Claude Code. It never touches the app it's wrapping — the underlines are drawn on top.

Why I made this

I type a lot of prose into terminal apps — long prompts for **Claude Code**, **Codex** and other AI agents, commit messages, SQL, quick notes. None of them spell-check, so my typos just sail straight through into prompts and history. My editor has had the red squiggly line for years; my terminal never got it.

So I built it. You wrap a terminal program once, and you get that squiggly line back — in Claude Code, Aider, Codex, vim, `psql`, your shell, whatever. It only shows up when you actually misspell something, and it never changes how the app you're running behaves.

To be fair: modern LLMs shrug off everyday typos — mistype "recieve" and Claude knows what you meant. But that robustness comes from the model having seen a word (and its common misspellings) countless times, so it fades exactly where the stakes are highest: rare words, names, and jargon, where the model quietly guesses — and an agent doesn't ask, it acts on its guess. It shrinks with model size, too: what a frontier model shrugs off, the local Qwen or Gemma behind your coding agent has to guess at. Spelling also outlives the prompt: agents echo your words into commit messages, PR text, c