AI/TLDR

NoBuzz

A Claude Code skill that hands Claude's last reply to a second model and prints back a plain-English version

Agent Skills & PluginsOpen source
Updated
21 Aug 2026
License
MIT
Coverage
1 story
$git clone https://github.com/adnanakil/nobuzz

What's new

21 Aug 2026

NoBuzz published: the /debuzz Claude Code skill with colleague, manager and director modes, routing Claude's last reply through the Antigravity CLI.

Latest news

Overview

NoBuzz is a small Claude Code skill, invoked as `/debuzz`, that rewrites the assistant's previous reply in plain English. Its target is a specific failure mode of coding assistants: the answer is technically right but arrives as a keynote — load-bearing assumptions, numbered revelations, and a kicker — when what you asked was why a test is flaky. The skill's premise is that no amount of prompting fully removes that voice from the model producing it.

So it does not ask Claude to rewrite itself. The skill writes the previous reply to a temporary file and shells out to Google's Antigravity CLI (`agy`), whose only job is to restate the content like a person would, and then prints that output verbatim. Printing it unedited is the load-bearing rule: letting the original model tidy up the translation reintroduces exactly the register being removed. When `agy` fails — usually an auth problem — you see the real error, and any fallback rewrite is clearly labelled as one.

Three modes set how much survives. `colleague`, the default, is aimed at another engineer and keeps every file path and code block intact while dropping the theatrics. `manager` compresses to what happened, why it matters and what is next, at roughly a third of the length and without code. `director` is three to five sentences — outcome, impact, ask. With no text argument it translates the last reply; paste text after the mode to translate that instead, and it also triggers on natural phrasings like "say that in normal english". It is MIT-licensed and installs as a directory copied into your skills folder.

What it does

  • `/debuzz` slash command that rewrites Claude's previous reply, or any text you paste after the mode
  • Cross-model by design: the rewrite is done by Google's Antigravity CLI, not by the model that wrote the original
  • Output is printed verbatim — the originating model never gets to edit the translation back into its own voice
  • Three audience modes: colleague (keeps code and paths), manager (about a third the length, no code), director (three to five sentences)
  • Errors surface honestly — an `agy` failure is shown, and a fallback rewrite is labelled as a fallback
  • Installs as a plain skill directory; no service, no configuration file

Getting started

You need Claude Code and the Antigravity CLI (`agy`). The skill is a directory you copy into your personal skills folder.

Install the skill

Clone the repository and copy the `debuzz` directory into your Claude Code skills folder.

bashbash
git clone https://github.com/adnanakil/nobuzz
mkdir -p ~/.claude/skills
cp -r nobuzz/debuzz ~/.claude/skills/

Install and authenticate the Antigravity CLI

The rewrite runs through `agy`. Install it, then run it once to complete the Google sign-in flow.

bashbash
curl -fsSL https://antigravity.google/cli/install.sh | bash
agy

Use it

With no arguments it translates Claude's previous reply. Add a mode to change the audience, or paste text after the mode to translate that text instead.

texttext
/debuzz
/debuzz manager
/debuzz director

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

When to use it

  • Get a straight answer out of a long agent reply without re-prompting for it
  • Turn a coding session's output into a manager- or executive-readable update
  • Read agent output in a second language more easily, by stripping idiom-heavy phrasing
  • See a worked example of a cross-model skill: one model's output post-processed by another, printed unedited

How NoBuzz compares

NoBuzz alongside other open-source agent skills & plugins tools AI/TLDR tracks, ranked by GitHub stars.

ToolStarsWhat it does
Superpowers★ 283kA composable skills plugin that installs a spec-first, TDD, subagent-driven development methodology into Claude Code, Codex, Cursor, Gemini CLI and other coding harnesses.
ECC★ 252kAn installable plugin that adds 68 agents, 286 skills, hooks, rules, memory and an agent-config security scanner to Claude Code, Codex and other coding harnesses.
NoBuzz★ 254A Claude Code skill that hands Claude's last reply to a second model and prints back a plain-English version
AREX-Skill★ 219A library of 5,000+ executable Agent Skills distilled from 1,000+ ML repositories, plus the DisCo CLI that distils a repository into skills and a router that narrows them to one branch.
Portal AI Plugins★ 216Spotify's plugin marketplace that brings Portal's software catalog, docs search and actions into Claude Code, Codex and Cursor, plus shunt, which routes bulk file reads to a cheaper model.
Vomit★ 191A Go CLI that hooks into Claude Code and pipes each reply through a local LLM, so the agent's output is rewritten in plainer language before it reaches the terminal.
HumanizerAn agent skill that rewrites AI-sounding prose against a list of 25 named writing patterns, grouped into five sections and applied without changing the claims in the text.