Claudette: Make Claude stop talking like a BuzzFeed article

Anthropic's Claude has a well-known verbal tic: it explains technical issues like a millennial clickbait article, full of 'load-bearing assumptions,' numbered revelations, and dramatic kickers. A developer (repo: adnanakil/nobuzz) built a Claude Code skill called /debuzz that hands Claude's last response to a different model — Gemini, via Google's Antigravity CLI (agy) — and prints the translation verbatim, because letting Claude 'tidy up' its own rewrite reintroduces exactly the voice being removed. The author jokingly calls it 'Claudette' while disclaiming the name, blaming a litigious Cat Wu.

The before/after example shows the value: Claude's retry-logic explanation is three paragraphs of buildup with a 'kicker,' while /debuzz produces three concise bullet points with exact file paths (syncQueue.ts:142) and concrete fixes (strip timestamp from dedupe key, raise backoff cap to 30s, re-throw ETIMEDOUT).

Installation is simple: git clone https://github.com/adnanakil/nobuzz and copy the debuzz folder into ~/.claude/skills. It requires Claude Code and the Antigravity CLI (installable via curl or irm). Usage is /debuzz [mode] [text]; with no text it translates Claude's most recent reply, and it also triggers on natural phrases. Three audience modes are available: colleague (default, engineer-focused, keeps code and file paths), manager (about a third the length, no code), and director (3-5 sentences with outcome, impact, ask).