Don't Paste the AI vs Vomit vs NoBuzz: AI Slop Tools Compared
TL;DR
dontpastetheai.com, Vomit, and NoBuzz hit Hacker News in the same week. A social contract, a local rewrite, and a second-model Claude Code skill - compared with a table and a when-to-stay guide.
Three projects hit Hacker News this week for the same irritation: LLM prose nobody read before it landed in Slack, a review, or a Claude Code session. dontpastetheai.com is a page you send as a link. Vomit rewrites Claude 5's "token vomit" through a local LLM. NoBuzz is a Claude Code skill that pipes the last reply through Gemini CLI.
They are not three products fighting for one install. They are three jobs: send a link, rewrite locally, or hand the last answer to a second model. Here is which job is which, and when to skip them and write the reply yourself.
Last updated: August 21, 2026
Official Sources#
| Tool | Job | Official resource | License |
|---|---|---|---|
| dontpastetheai | Social contract (send a link) | dontpastetheai.com / GitHub | See LICENSE in the repo |
| Vomit | Local rewrite of Claude output | GitHub / author note | GNU GPLv3 |
NoBuzz (/debuzz) | Second-model Claude Code skill | GitHub | MIT |
Commands below are copied from those READMEs (August 21, 2026). NoBuzz needs the Gemini CLI. Vomit talks to Llama.app, Ollama, or an OpenAI-compatible endpoint.
Three jobs, not three competitors#
Text slop is the cousin of visual slop: a generator default, shipped without a human pass. Teams now encode that pass as skills. These three sit earlier. They change the words after the model speaks. They do not make it smarter.
| Axis | dontpastetheai | Vomit | NoBuzz |
|---|---|---|---|
| Job | Social: send a link | Local rewrite of Claude's display | On-demand rewrite via a second model |
| Where it runs | A static site you share | Your machine, hooked into Claude Code | Claude Code skill + Gemini CLI |
| Who it is for | Someone pasting unread LLM text into chat or review | You, reading Claude 5 all day | You, when the last Claude reply needs a human register |
| Install | None. Copy the URL. | go install, then vomit init, then vomit scrub -claude | Clone the repo, copy debuzz into ~/.claude/skills/ |
| Extra model | None | Local LLM (Llama.app, Ollama, or OpenAI-compatible) | Gemini CLI (gemini) |
| Privacy | Public page | Fully local, no telemetry (per the README) | Source text leaves your machine for Gemini |
| Author stance | Satire. Use the tools, read them first. | Art project. Slow, Mac-tested, hallucinates a bit. | Joke about BuzzFeed voice. MIT skill, not a product. |
| Production? | Yes, as a link. That is the whole product. | Useful hobby tool. README is honest about the gaps. | Useful skill. Gemini auth is a real dependency. |
dontpastetheai: send a link#
dontpastetheai.com (mirror: dontquotetheai.com) is a spiritual cousin of nohello.net and dontasktoask.com: static HTML, translations, and a work-safe "smooth" page. The Hacker News thread passed a thousand points. The README is explicit: this is satire, and the point is not "never use AI."
Use AI for drafts. Read them. Rewrite. Do not be a middleman between a chatbot and a human who already has the same chatbot.
The site's four "Do this instead" steps, quoted from the smooth English page:
- You can use AI. Seriously! It's a great tool for drafting. Just read what it gave to you, then write your own version, or polish the text. Don't be a middleman between it and the answer.
- Take the part that actually answers the question and ignore the rest. Three sentences is all it takes, and even if they're copied, at least you read them.
- If some part of the model's answer is genuinely useful, quote it and explain why. "I asked Claude and this bit here makes sense:".
- If you have nothing to add, just say so. "No strong opinion here". Silence is also an option.
That is the contract. They asked you for context, taste, and judgment. Pasting eight hundred unread words shifts comprehension onto everyone else, the same tax as unreviewed AI code.
The site has a louder variant. Do not send it at work. The smooth page is the one for a coworker, a manager, or a review thread.
Vomit: rewrite locally#
Vomit is Zach Ahn's Go CLI for Claude 5's display text. It buffers what Claude tried to say, forwards that to a local LLM, and shows the rewrite. The author's post calls it an art project. The HN thread treated it as a real itch: AGENTS.md style rules fade as the session grows.
Install, copied from the README:
go install github.com/zachahn/vomit@latest
vomit init
vomit scrub -claude
vomit init stores local-model connection details. vomit scrub -claude prints the Claude Code hook instructions. Non-invasive mode from the same README:
vomit list- list Claude session identifiersvomit tail [<session_identifier>]- translate a session, or follow the latestvomit help- the rest of the commands
Named backends: Llama.app, Ollama, and anything that speaks the OpenAI API. If you do not have a local model, the README suggests Llama.app plus GPT-OSS 20B, then vomit init.
The README is the spec sheet. The local LLM only sees what Claude tried to communicate, not files or tool results, so it hallucinates a bit. It is pretty slow, vibe-coded, and only tested on Mac. You can miss the original message; it writes under TMPDIR and does not rewrite the session on disk. AgentsView is the suggested recovery path. License: GNU GPLv3. Use it when the pain is your Claude Code reading, not someone else's Slack paste. Hooks in general: Claude Code tips.
NoBuzz: a second model as editor#
NoBuzz is Adnan Akil's Claude Code skill. The command is /debuzz. The README jokes that Claude was trained on old BuzzFeed articles, and that they considered calling it "Claudette" and then definitely absolutely did not. The HN thread is mostly CLAUDE.md one-liners. A second model is the honest workaround: do not ask the buzzer to debuzz itself.
Install, from the README:
git clone https://github.com/adnanakil/nobuzz
mkdir -p ~/.claude/skills
cp -r nobuzz/debuzz ~/.claude/skills/
Requirements listed there: Claude Code, and the Gemini CLI (npm install -g @google/gemini-cli), authenticated. Run gemini once and use /auth, or set GEMINI_API_KEY.
/debuzz [mode] [text]
| Mode | Audience | What you get |
|---|---|---|
colleague (default) | An engineer | Same content, file paths and code blocks intact, zero theatrics |
manager | A technical-adjacent manager | What happened, why it matters, what's next - about a third the length, no code |
director | An executive | Three to five sentences: outcome, impact, ask. Assumes thirty seconds of attention |
With no text it translates Claude's previous reply. Paste text after the mode to translate that instead. It also triggers on phrases like "say that in normal english."
Mechanically: write the previous reply to a temp file, pipe it through gemini -p with plain-English instructions, print Gemini verbatim. If Gemini errors (usually auth), you see the error. Claude only offers its own rewrite as a labeled fallback. MIT. That last rule is the point: a debuzzer that asks the original model to edit itself is how you get another load-bearing translation.
Decision guide#
- Unread LLM paste in Slack, email, or review. Send dontpastetheai.com. Do not install anything.
- You live in Claude Code and the display register is the tax. Vomit if you already run a local LLM and want a rewrite on every turn. NoBuzz if you want on-demand
/debuzzand will install Gemini CLI. - You need altitude, not just tone. Only NoBuzz has colleague / manager / director modes.
- You cannot send source text to Gemini. Vomit stays local.
- Not on a Mac, or you need a fleet standard. Vomit is Mac-tested only. Weekend tool, not a rollout.
- You want a different model in the coding loop. Harness change, not a rewriter. Vomit and NoBuzz hook Claude Code. If you run OpenCode instead, those two do not apply; dontpastetheai still does. See Ox Alpha on OpenCode.
When to stay: write the reply yourself#
The default is still: draft with a model if you want, then write the message in your own words. That is step one of dontpastetheai, and it costs nothing.
Skip these tools when the reply is three sentences you can type faster than you can paste; when you have no strong opinion ("No strong opinion here" is already the answer); when you can already skim Claude for the file path and the bug; or when you would only install a rewriter so you can paste that unread. The second model is not a hall pass.
Satire and art projects can be funny and incomplete. dontpastetheai is done: it is a page. Vomit and NoBuzz are small, honest tools for a voice problem the labs have not fixed. Love them for that. The culture is still: read it, cut it, sign it.
FAQ#
What is dontpastetheai.com?#
A satirical static site you send to someone who answers with unread chatbot text. Cousin of nohello.net and dontasktoask.com. Smooth page: use AI for drafts, then read and rewrite. GitHub.
What is Vomit for Claude Code?#
A Go CLI that pipes Claude's displayed tokens through a local LLM. vomit scrub -claude prints hook setup. GPLv3, local, slow, Mac-tested, a bit hallucination-prone. GitHub.
What is NoBuzz, and is it called Claudette?#
NoBuzz is the repo. /debuzz is the skill. "Claudette" is the README joke. It sends Claude's last reply through Gemini CLI in colleague, manager, or director mode. MIT. GitHub.
Should I send the angry version of dontpastetheai?#
No, not at work. The site marks the smooth page as work-safe. This post quotes only that version.
When should I skip these tools and write the reply myself?#
Whenever you can. Use them when unread paste in a shared channel, or Claude Code's register on every turn, is recurring. A one-off is still cheaper as three sentences you wrote.
Continue Reading#
- AI Design Slop and How to Spot It
- Taste Skills Are Turning Agent Review Into Infrastructure
- Write Code Like a Human Will Maintain It
- 60 Claude Code Tips and Tricks
- Ox Alpha on OpenCode
- Thinking in Python: Bruce Eckel Revives His 2008 Book With Claude in 2026 - the strongest counterexample to unedited slop, a Claude-completed book with build-verified examples
Sources#
- dontpastetheai.com (smooth English page, fetched August 21, 2026)
- dontquotetheai README (fetched August 21, 2026)
- Hacker News: Don't paste the AI, please (fetched August 21, 2026)
- Vomit README (fetched August 21, 2026)
- Zach Ahn: How to fix Claude 5's token vomit (fetched August 21, 2026)
- Hacker News: Vomit (fetched August 21, 2026)
- NoBuzz README (fetched August 21, 2026)
- NoBuzz
/debuzzSKILL.md (fetched August 21, 2026) - Hacker News: Claudette / NoBuzz (fetched August 21, 2026)
- Gemini CLI (from the NoBuzz README, checked August 21, 2026)
- OpenCode Go - referral signup if you want a non-Claude-Code harness (code
M6HEHM4JM5)
Some links to tools above are referral links - see our affiliate disclosure.
Get the next comparison like this in your inbox
One email a week on Comparison and the rest of the AI dev stack. Free.
Read next on Claude Code
AI Design Slop: 16 Patterns That Out Your App as Vibe-Coded
AI design slop is the set of 16 landing page patterns that mark a site as vibe-coded. Adrian Krebs scored 1,590 Show HN pages: 22% heavy, 32% mild, 46% clean.
7 min readTaste Skills Are Turning Agent Review Into Infrastructure
GitHub trending is full of anti-slop, taste, and compound-engineering skills. The real signal is not that agents need more prompts. It is that teams are trying to make subjective review criteria executable.
8 min readWrite Code Like a Human Will Maintain It - The AI Era Debate
A new essay argues that letting AI generate sloppy code creates a downward spiral where future AI absorbs those bad patterns. HN's 250+ comment thread is split between believers and pure vibe-coders.
6 min readTechnical content at the intersection of AI and development. Building with AI agents, Claude Code, and modern dev tools - then showing you exactly how it works.
