Stop Slop: the Claude skill that removes AI writing tells.

Introduction

"In today's ever-evolving digital landscape, design isn't just aesthetics, it's strategy." You have read that sentence a hundred times. It says nothing, and you spot it at a glance: an AI wrote it, or someone who copied from one.

This kind of text has a name: slop. Sentences that sound right and carry no information, the same structures over and over, emphasis where none is needed. For designers it is a practical problem. Interface microcopy, a spec for developers, a case study, a cover letter: once the reader smells AI, they stop trusting what comes next.

Stop Slop is the Claude skill built to remove that smell. This article covers what it does, the rules it applies, how it scores a text, how to install it, and how to adapt it to a language other than English.

What Stop Slop is

Stop Slop is a skill written by Hardik Pandya and published on GitHub under the MIT license. As of 2 November 2026 the repository has about 18,500 stars and 1,400 forks, which makes it one of the most popular writing skills around.

Like every skill, it is a folder. Inside there is a SKILL.md file with the core rules and a references/ folder with three files Claude reads only when it needs them:

• phrases.md: phrases to remove, grouped by family (throat-clearing openers, emphasis crutches, business jargon, adverbs, meta-commentary);
• structures.md: structures to avoid, such as formulaic contrasts, dramatic fragmentation and rhetorical questions answered on the spot;
• examples.md: five before-and-after pairs.

The skill's description says when to trigger it: while drafting, while editing, or when reviewing a text for AI tells.

Since October, skills work outside Claude too. On 22 October 2026 Anthropic donated the Agent Skills standard to the Agentic AI Foundation under the Linux Foundation, and according to Anthropic more than 70 platforms have adopted it, including ChatGPT, Cursor, GitHub Copilot and VS Code. A skill like Stop Slop is written once and travels with you.

The eight rules

The core of the skill is eight rules.

• Cut filler phrases. Throat-clearing openers, emphasis crutches and every adverb go.
• Break formulaic structures. Binary contrasts, negative listings, dramatic fragments, rhetorical setups, objects acting like people.
• Use active voice. Every sentence needs a human subject. No passives, no "the decision emerges".
• Be specific. No vague declaratives and no lazy extremes like "always", "every", "never".
• Put the reader in the room. "You" beats "people", the specific beats the abstract.
• Vary rhythm. Mix sentence lengths. Two items beat three. End paragraphs in different ways. No em dashes.
• Trust readers. State facts directly, without softening or hand-holding.
• Cut quotables. If a sentence sounds like a pull-quote, rewrite it.

On top of these come twelve quick checks for the end of a draft: any adverbs? Any passive voice? A sentence starting with a Wh- word? Three sentences in a row of the same length? A paragraph ending on a punchy one-liner? An em dash? Each one has its fix.

Slop passes every grammar check: the grammar is right, the information is missing.

The score out of 50

Stop Slop also grades. Each text gets 1 to 10 points on five dimensions.

The total runs from 5 to 50. Below 35, the text gets rewritten. The threshold turns a hunch ("this sounds fake") into a number you can compare between the first and the second draft.

Before and after

The skill's examples file holds five pairs. These two say more than any explanation.

Examples from hardikpandya/stop-slop, MIT license.

In the fourth example of the same file, the "after" version contains an em dash: the exact thing rule six bans. Even the author slipped. Slop is a habit deep enough to escape the person who wrote down the rules, which tells you how to use the skill: as a final review, with your judgment switched on.

Four texts designers write every week, before and after. The examples are made up, the tells are not.

Using it in other languages

Stop Slop's rules are written for English, and some of them do not translate word for word. Adverbs ending in "-ly", sentences opening with "What" or "Why", the "It's not X, it's Y" contrast: every language has its own versions, and an AI writing in that language uses all of them.

I write in Italian as much as in English, so I built a mapping for Italian. It is not part of the original skill, and it shows the method for any language.

For Italian, I turned this mapping into stop-slop-it, a complete Italian version of the skill, free and open source. For any other language, add a file to the original skill's references/ folder and mention it in SKILL.md; the MIT license allows it. A starting point:

# references/phrases-it.md

## Openers to remove
Ecco il punto • Diciamocelo • La verità è che • Partiamo da un presupposto
In un mondo sempre più ... • Nell'era di ... • Oggi più che mai

## Adverbs to remove
davvero • semplicemente • assolutamente • fondamentalmente
sostanzialmente • letteralmente • profondamente • decisamente

## Structures to rewrite
"Non è X, è Y" / "Non si tratta di X, ma di Y" -> state Y
"non solo X, ma anche Y" -> pick the one that matters
"Perché è importante? Semplice: ..." -> just give the answer
"Si è deciso di ..." -> name who decided

## Punctuation
No em dashes: use a comma, a colon or parentheses.

How to install it

The skill's README lists four routes. Pick the one that matches how you use Claude.

• Claude Code. Add the folder as a skill. The quickest way is to clone it into your personal skills:

git clone https://github.com/hardikpandya/stop-slop.git ~/.claude/skills/stop-slop

• Claude.ai Projects. Upload SKILL.md and the references/ files to the project knowledge.
• Custom instructions. Copy the core rules from SKILL.md.
• API. Put SKILL.md in the system prompt; the reference files load on demand.

On the other platforms that adopted the Agent Skills standard the folder is the same: follow their docs to import it. Once installed, ask Claude to review a text "with stop-slop", or let it trigger on its own when you ask it to write or edit.

When not to use it

The repository does not say when to leave it aside. I switch it off in four cases.

• When the product's tone is deliberate. A playful brand uses exclamations and repetition on purpose. Stop Slop would read them as flaws.
• When repetition helps understanding. In microcopy, using the same word on a button and in its confirmation message helps the user. Varying it to avoid repetition confuses them.
• In legal text and safety instructions. There, impersonal phrasing and caution are how you stay precise.
• On your own voice, once you have one. Some rules are absolute ("no adverbs"). Treat them as a check, then decide.

I have one rule that holds in every project of mine, from this portfolio to Frigolo: no em dashes, ever. Stop Slop applies it along with everything else. The rest of its rules I use as a checklist before publishing.

If you already use other skills for copy, like the ones I collected in The 13 Claude Skills to Use if You Are an Experience Designer, Stop Slop goes last in the chain: write first, fix the substance, then remove the slop.

Frequently asked questions

Is Stop Slop free?

Yes. It is released under the MIT license: you can use it, change it and share it, for work too.

Does it work in languages other than English?

Claude applies it to other languages as well, but the rules are built on English examples. For better results, add a reference file with your language's tells, like the Italian one above.

Does it work with ChatGPT or Cursor?

Skills follow the open Agent Skills standard, which according to Anthropic is integrated in more than 70 platforms, including ChatGPT, Cursor, GitHub Copilot and VS Code. The rules in SKILL.md also work pasted into any assistant's instructions.

Does it make text undetectable by AI detectors?

That is not its purpose, and no skill can guarantee it. Stop Slop is for clearer, more direct writing. Whoever wrote the text, the reader gains.

How is it different from a copywriting skill?

A copywriting skill helps you write text that persuades. Stop Slop removes what makes a text sound fake. Use them together, in that order.

Conclusion

Slop takes half a second to spot and costs a lot: the reader stops trusting whatever follows. Stop Slop writes down what a good editor does by instinct, and makes it repeatable.

In another language you need one extra step, a list of that language's tells. The rest is the usual job: choosing what to say, and to whom.

Updated 2 November 2026. Repository figures (stars and forks) refer to that date.