# No AI Slop: a skill that strips 20+ AI writing patterns from your drafts > Peter Yang's No AI Slop is a Python-backed editing skill for Claude Code, Codex and ChatGPT that flags binary contrasts, throat-clearing openers and faux-insight setups while trying to leave your own cadence intact. It is a linter for prose, not a rewriter, and that distinction decides who should install it. **petergyang/no-ai-slop** — Removes 20+ patterns of AI slop from any piece of writing. That's the whole thing." | It also enforces the fundamentals that make writing good: Lead with the point when it helps, use active voice, untangle hard-to-follow sentences, and prefer concrete numbers over abstractions. - Repository: https://github.com/petergyang/no-ai-slop - Website: https://creatoreconomy.so/p/use-my-no-ai-slop-skill-to-remove-20-ai-slop-patterns - Stars: 11,343 · Forks: 775 - Language: Python - License: MIT - Published: 2026-08-04 · Updated: 2026-08-18 · Language: en - Canonical page: https://hysenlabs.com/projects/petergyang-no-ai-slop ## The problem No AI Slop names: everyone's drafts sound the same The README states the problem plainly: AI makes it easy to generate clean writing that all sounds the same. It lists the offenders, and they are recognisable. "It's not X. It's Y." "What nobody tells you is..." "The future isn't coming. It's already here." These are not grammar errors. A grammar checker will pass every one of them, because each sentence is correct. The failure is one of pattern, not of syntax. The second half of the problem is the one that matters more for anyone with a byline. When you hand a draft to an AI editor, the README says it can smooth away the vocabulary, cadence, humor, and imperfections that make the writing sound like you. A tool that removes slop by rewriting everything into neutral prose has traded one uniform voice for another. No AI Slop is aimed at writers who want the tics gone and the fingerprints kept. That framing sets the audience. It is for people who already write, who use an agent as an editor rather than a generator, and who can tell the difference between a sentence that is bad and a sentence that is merely theirs. If you are generating first drafts from a blank page, the skill has less to bite on. ## How the skill works: rules in SKILL.md, checks in eval.md The repository is small and the layout is the architecture. According to the README, skills/no-ai-slop/SKILL.md contains the editing rules and workflow, and skills/no-ai-slop/eval.md contains the checks the skill runs on its own work. That two-file split is the interesting design choice: the rules say what to change, and the eval file is a separate pass the skill applies to its output before handing it back. There is no model of its own and no service to call. The skill is instructions plus a checklist, loaded by whichever agent you already use. .codex-plugin/plugin.json holds the ChatGPT and Codex plugin metadata, and scripts/build_plugin.py builds and validates the plugin package, which is how the same skill reaches more than one host. Python appears in the repository as the language of that build script, not as a runtime you invoke to edit text. The pattern list is the substance. The README enumerates ten of the 20+ categories: binary contrasts, throat-clearing openers, faux-insight setups, colon reveals, dramatic fragments, superficial analysis, importance puffery, weasel attribution, synonym cycling, and fake-profound endings. Alongside those it checks what the README calls the fundamentals: lead with the point when that helps, use active voice, untangle hard-to-follow sentences, and prefer concrete details over abstractions. The first list is about taste; the second is about clarity, and they fail differently. A binary contrast is a stylistic tell, while a tangled sentence is a comprehension problem. Bundling both into one pass means a single run can return both kinds of note, which is convenient and also makes the output longer to read. ## Installing No AI Slop and running your first edit The README gives two install paths. The easiest, in its words, is to paste an instruction into ChatGPT, Claude Code, Codex, or your favorite coding agent. The text to paste is exactly this: ```text Install the /no-ai-slop skill globally from https://github.com/petergyang/no-ai-slop ``` The agent handles the fetch. The second path is the npx command, which is what you want if you would rather see the install happen yourself: ```sh npx skills add petergyang/no-ai-slop --skill no-ai-slop --global --yes ``` The --global flag is what makes the skill available outside a single project, and --yes skips the confirmation prompt. Once installed, editing is a slash command followed by your text: ```text /no-ai-slop (your writing) ``` The README says the skill removes the slop patterns, preserves your personal voice, and lists what it changed. That last part is the one to check first. Paste a paragraph you are happy with and read the change list before you accept anything. If the list is long on a paragraph you liked, the skill is editing past your voice and you should narrow what you ask it to touch. There is a second mode worth knowing on day one. Prefix the same command with a question: ```text /no-ai-slop is this slop? (your writing) ``` In this mode the README says the skill quotes every slop pattern it found without guessing whether AI wrote the text. That restraint is deliberate and it is the honest behaviour: pattern matching can tell you a sentence looks like a tell, and it cannot tell you who typed it. Run detect mode on your own old writing before you run it on anyone else's. ## Where No AI Slop gets it wrong, and what it cannot do The skill cannot tell whether AI wrote a piece, and it does not claim to. Detect mode quotes patterns. A human who writes "Here's the thing" gets flagged the same as a model that does, because the check is on the words, not on their origin. If you were hoping for an authorship verdict, this is the wrong tool and the README is upfront about it. The deeper limitation is that the pattern list is a list. Twenty-plus categories cover the tells the author kept seeing, and the space of ways to write badly is larger than that. A draft can be dull, evasive, or padded without tripping a single named pattern, and the skill will pass it. The fundamentals checks (active voice, concrete details) catch some of that, but they are judgement calls the agent makes in the moment, not enumerated rules, so results will vary between runs and between hosts. There is also a voice problem the skill creates rather than solves. Removing a pattern is not the same as replacing it with something better. Strip "It's not X. It's Y." from a paragraph and you may be left with a claim that now has no structure at all. The skill lists what it changed; it does not promise the result reads well. On a draft where the slop was load-bearing, you are the one who has to rebuild the sentence. Finally, the repository's own framing is narrow. It is a writing skill for English prose. The README does not document behaviour on other languages, on code comments, or on documentation, and the pattern list is built from English tells. Treat it as an editor for essays and posts, not as a general text-quality gate for a codebase. ## No AI Slop against Vale and plain linters The obvious comparison is a prose linter such as Vale, which also flags style problems in writing. The difference is where the rules live and who applies them. Vale is a standalone binary with rule files you write yourself, and it runs in CI or from a shell against files on disk. No AI Slop is a skill loaded by an agent, and its rules ship as SKILL.md and eval.md inside the repository. You do not write the rules; you install the author's list and take it as given. That trade cuts both ways. Vale gives a team a shared, versioned, testable ruleset and runs without a model in the loop, so the same input always produces the same output. No AI Slop gives you a curated list of AI-specific tells that a general linter has no reason to know about, plus the judgement to untangle a sentence rather than just flag it. It also inherits the variability of the agent running it. A second comparison is the editing pass inside a general assistant. Ask Claude or ChatGPT to "make this better" and you get a rewrite with no accounting. No AI Slop's contribution is the change list and the detect mode, which turn an opaque edit into something you can audit. If you already trust your agent's rewrite and never read the diff, the skill adds little. If you have been burned by a rewrite that flattened your voice, the audit trail is the point. ## Maintenance, licence and what the repository actually commits to The last push to the default branch was on 2026-08-01, the same day as the v1.0.6 release. The two releases before it, v1.0.5 and v1.0.4, landed on 2026-07-31 and 2026-07-27, so the project saw three tagged releases in the week before that push. The repository is not archived. Three releases in six days is a burst rather than a cadence, and the README does not describe a support policy, a compatibility matrix, or a deprecation process for the pattern list. Anyone pinning this for a team should pin a version rather than track main. The upgrade cost is low by construction. The skill is text: rules in SKILL.md, checks in eval.md, metadata in .codex-plugin/plugin.json. When the pattern list changes, your install changes with it, and there is no schema to migrate. The one thing to watch is the host. Because the skill runs inside ChatGPT, Claude Code, Codex or another agent, a change in how a host loads skills affects you more than a change in this repository does, and that is outside the project's control. The licence is MIT, which permits commercial use and modification. The repository also carries PRIVACY.md and TERMS.md alongside the LICENSE, which is unusual for a writing skill and worth reading if you plan to use it on client work. Nothing here is legal advice; read the files yourself. The README points to Behind the Craft and the author's newsletter and YouTube channel, so the skill and the commercial material around it share an owner. That is not a problem, but it does mean the pattern list reflects one writer's taste, and the licence does not make it a standard. ## Conclusion Adopt No AI Slop if you write in English, edit your own drafts inside Claude Code, Codex or ChatGPT, and want the slop patterns named rather than silently smoothed away. Skip it if you need a grammar checker, a style guide for a team, or a tool that works offline without an agent; the repository ships a SKILL.md, an eval.md and a plugin manifest, not a standalone CLI. Before relying on it, open skills/no-ai-slop/SKILL.md and confirm the 20+ patterns match the ones your own drafts keep producing, then run the detect mode on a piece you already know is clean to see whether it stays quiet. ## FAQ ### What does "no AI slop" mean? In this project it refers to a fixed list of 20+ writing patterns that make AI-assisted prose sound uniform, such as binary contrasts ("It's not X. It's Y."), throat-clearing openers, and faux-insight setups. The README frames the goal as removing those patterns without flattening the writer's personal voice. ### Is all AI AI slop? The repository does not make that claim. Detect mode quotes every slop pattern it finds without guessing whether AI wrote the text, so the tool judges the wording rather than the origin. A human who writes "Here's the thing" is flagged the same as a model that does. ### Why is AI being called slop? The README's stated reason is that AI makes it easy to generate clean writing that all sounds the same, and it lists the recurring lines that result. It also notes that using AI to edit can smooth away the vocabulary, cadence, humor, and imperfections that make writing sound like a specific person. ### How can I avoid AI slop? Install the skill globally and run /no-ai-slop on your writing, then read the list of changes it reports before accepting them. For a second opinion, run the detect form, /no-ai-slop is this slop?, which quotes the patterns it found instead of rewriting the text. ## Sources - [Official documentation](https://creatoreconomy.so/p/use-my-no-ai-slop-skill-to-remove-20-ai-slop-patterns) - [Official README](https://github.com/petergyang/no-ai-slop#readme) - [Project repository](https://github.com/petergyang/no-ai-slop) - [Release notes](https://github.com/petergyang/no-ai-slop/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/petergyang-no-ai-slop