Claudette review (2026): a free, clever Claude Code skill that de-buzzes AI-tell via Gemini, but it's a terminal workaround, not a content engine.
Claudette (really the /debuzz skill in the open-source "nobuzz" project) is a small, clever, free fix for a real annoyance: Claude narrating simple work like a keynote. Judged as a terminal utility it does its one job well, and the choice to translate through a second model (Gemini) rather than ask Claude to fix its own prose is smart. Judged as anything bigger, keep expectations honest — it de-buzzes one reply, after the fact, and depends on the Gemini CLI. It is a quality-of-life hack, not a content or brand-voice system.
Claudette showed up on Hacker News and GitHub in August 2026 with a premise every heavy Claude user recognizes: ask why a test is flaky and you get a "load-bearing assumption," three numbered revelations, and a note that the third is the most instructive. Claudette — the /debuzz command in Adnan Akil's "nobuzz" project — exists to strip that off. This review is about whether it works and where it fits, not just whether the idea is funny.
The short version. As a terminal utility it is genuinely good at its one job. It captures Claude's previous reply, pipes it through the Google Gemini CLI with plain-English instructions, and prints the rewrite, with colleague, manager, and director modes for different audiences. The standout design decision is that it does not trust Claude to de-buzz itself — it uses a different model, on the premise that prompting Claude alone rarely holds the style down. For making a single message readable, that works.
The honest catch is scope and dependency. It only touches one reply at a time, only in your terminal, and only if you have the Gemini CLI installed and authenticated — so the tool that removes Claude's AI-tell leans on a second model to do it. It outputs plain text and cannot brand, format, schedule, or publish anything. The author frames it openly as a workaround, and that framing is correct.
This review scores Claudette on its own terms as a free developer utility, then is candid about the ceiling: de-buzzing a terminal message is a quality-of-life win, not a content workflow. Where it genuinely earns its keep, this page says so plainly.
Claudette is the informal name — the README jokes it is "definitely absolutely not called 'Claudette'" — for "nobuzz," an MIT-licensed Claude Code skill that installs as the /debuzz slash command. When Claude answers in its theatrical register, you run /debuzz and it writes that reply to a temp file, pipes it through the Gemini CLI with instructions to rewrite flat and direct, and prints Gemini's output verbatim; if Gemini errors (usually auth) you see the real error instead of Claude re-buzzing its own translation. It ships three modes: colleague (default) keeps every file path and code block with zero theatrics, manager compresses to roughly a third the length with no code, and director cuts to three to five executive sentences of outcome, impact, and ask. You invoke it as /debuzz [mode] [text] — with no text it de-buzzes the last reply — and it also triggers on natural phrases like "say that in normal english." Setup is cloning the repo, copying the debuzz folder into ~/.claude/skills/, and having the Gemini CLI installed and authenticated.
Claudette fits developers and technically comfortable Claude Code users who are tired of Claude's over-dramatized explanations and want a fast, free way to flatten them — for readable standups, changelogs, and manager or exec summaries pulled straight from a long technical answer. It is a poor fit for anyone who wants AI-tell kept out of content they publish, a consistent brand voice enforced automatically, or any output beyond plain text, because it does none of those and is not trying to. Non-technical creators who just want on-brand posts made and shipped are working at the wrong layer with a terminal skill, however neat.
| Dimension | Score | Why |
|---|---|---|
| Core idea | 4.5 / 5 | Targets a real, specific pain — Claude's theatrical prose — and the fix is intuitive and immediately useful. |
| De-buzzing effectiveness | 4.0 / 5 | Works because it translates through a second model (Gemini) rather than asking Claude to flatten its own writing. |
| Mode design | 4.2 / 5 | Colleague, manager, and director modes map cleanly to real audiences and are a genuinely thoughtful touch. |
| Ease of setup | 3.6 / 5 | Clone-and-copy install is simple, but it also requires the Gemini CLI installed and authenticated. |
| Dependencies & friction | 3.0 / 5 | The tool that removes Claude's AI-tell depends on a second model and account; every rewrite is an extra round-trip. |
| Scope | 2.8 / 5 | Fixes one terminal reply at a time; no brand memory, no formats, no publishing — narrow by design. |
| Value | 4.5 / 5 | Free and MIT-licensed; the only cost is your own Gemini usage. |
| Maturity | 3.2 / 5 | A tiny, new hobby project the author openly frames as a workaround rather than a durable fix. |
Claudette has no price tag — it is MIT-licensed and free to clone. That is the strongest part of its value story: for a developer who already lives in Claude Code, it costs nothing to add and immediately makes verbose answers readable. As a free quality-of-life tool it is easy to recommend.
The cost that does not show up in the repo is the dependency. Claudette de-buzzes Claude by piping the text through the Google Gemini CLI, so you must install and authenticate Gemini, and at any real volume you are paying Google for those calls. It is a curious shape — a free tool whose whole function is spending a second model's tokens — and worth pricing honestly before you assume "free" means free.
More to the point for content work, the model layer is the cheap part of making content in 2026. De-buzzing a terminal reply is close to zero; the expensive, time-consuming layer is generating on-brand video, images, and long-form and distributing them. Claudette does not touch that layer, so budget it as a free developer convenience, not a content solution.
| Use case | Fit | Why |
|---|---|---|
| Making Claude Code explanations readable | Strong | This is precisely what it was built for, and it does it well. |
| Manager or exec summaries from a long technical answer | Strong | The manager and director modes compress a verbose reply into audience-appropriate length. |
| A free, hackable terminal tool | Strong | MIT-licensed and only a few files, so it is trivial to install and easy to fork. |
| One-off rewrite of arbitrary pasted text | OK | /debuzz [mode] [text] handles any text, though it is optimized for Claude's output. |
| Avoiding a second model or API entirely | Weak | It depends on the Gemini CLI, so the de-buzzing is not self-contained. |
| Removing AI-tell from content you publish | Weak | It fixes one terminal reply; it has no brand-voice profile and cannot govern a batch of content. |
| On-brand content published across platforms | Weak | No formats beyond text, no captions, no scheduling, and no publishing — that layer is absent by design. |
Kompozy is not a developer skill and does not compete with Claudette head-on — it sits one layer up. Claudette answers "how do I make this one Claude reply read normally?" Kompozy answers "how do I keep AI-tell out of the content I publish, and get it onto every platform on a schedule?" The most telling detail from Claudette's own launch discussion is that people keep re-teaching Claude their style because a growing context window makes it forget the rules. A Persona Brief is the durable version of that: a style contract Kompozy applies to every generation without needing a second model to translate after the fact.
So the honest way to use them is together. Keep Claudette for your Claude Code sessions. When the output is content — a caption, a script, a carousel, a blog, a newsletter — let Kompozy generate it under a Persona Brief and banned-word filter so the voice is right at the source, run each asset through a per-post review gate, and schedule and publish across 9 platforms plus Mailchimp and blog. Where Claudette leans on the Gemini CLI to fix Claude's prose, Kompozy runs managed Claude and OpenAI directly, with a bring-your-own-key option on the Founding tier. Kompozy pricing runs from Starter at $99/mo (5,500 credits) to Pro at $299/mo (18,000 credits), with a custom, sales-led Enterprise plan.
Claudette is the informal name for "nobuzz," an open-source Claude Code skill by Adnan Akil that installs as the /debuzz slash command. It takes Claude's verbose, clickbait-style reply and pipes it through the Google Gemini CLI to rewrite it in plain English, with colleague, manager, and director modes. It surfaced on GitHub and Hacker News in August 2026 and is MIT-licensed and free.
If you use Claude Code and dislike its theatrical prose, yes — it is free, quick to install, and does its one job well. Just set expectations: it fixes one reply at a time, depends on the Gemini CLI being installed and authenticated, and does nothing beyond rewriting text. As a terminal convenience it earns a spot; as a content or brand-voice tool it is out of scope.
Not officially. The real project is nobuzz and the command is /debuzz; the README jokes it is "definitely absolutely not called 'Claudette'." The nickname stuck anyway, including in the Hacker News thread where it circulated.
The skill is free, but it de-buzzes Claude by running the text through the Google Gemini CLI, which you install and authenticate. Light use may fit free tiers, but at volume you pay Google for Gemini calls — so "free tool" does not always mean zero cost.
On purpose. The project's premise is that prompting Claude to stop writing in its clickbait register rarely works for long, so it delegates the rewrite to a different model rather than asking Claude to translate itself. Whether that is elegant or a smell is a fair debate, but it is a deliberate design choice, not a bug.
They work at different layers. Claudette removes clickbait style from one Claude reply after it is written, in your terminal. Kompozy prevents it for content you publish — a Persona Brief and banned-word filter govern every generation across 18 formats, so nothing needs a second model to de-buzz, and Kompozy then schedules and publishes across 9 platforms. Use Claudette for your terminal and Kompozy for your feed.
See Claudette (nobuzz / the /debuzz skill) vs Kompozy comparison → · Get Started →