A developer has released NoBuzz, a Claude Code skill that tackles one of the most relatable problems in AI-assisted coding: Claude's habit of writing like it is delivering a TED talk about its own pull request. The skill, nicknamed Claudette, takes Claude's last response and pipes it through Google's Gemini CLI, which rewrites it in plain, human language. The project shot to the top of Hacker News within a day of launch.
The setup is refreshingly simple. Clone the repo, copy the debuzz skill folder into ~/.claude/skills, and install the Gemini CLI. When you run /debuzz, Gemini translates the response and prints it verbatim, because letting Claude tidy up the translation would just reintroduce the exact voice being removed. The README's before-and-after examples show the difference: a rambling explanation about load-bearing assumptions becomes three concrete bug fixes with file names and line numbers.
NoBuzz is a small project with a big point: no amount of prompting fully cures an AI's house style, and routing output through a second model is a pragmatic fix. It also highlights how developers are increasingly stacking models, using each one where it is strongest. With 71 stars in its first day, it clearly struck a nerve.