Peter Yang Open-Sources /no-ai-slop, a Claude Skill for De-Sloppifying Writing

TL;DR: AI educator Peter Yang open-sourced /no-ai-slop on July 22, 2026 — a free Claude skill that strips over 20 repetitive AI writing patterns (fake-profound closers, binary contrasts, throat-clearing openers) while explaining every edit and preserving the writer's voice. The GitHub repo hit nearly 1,000 stars within a day. It lands the same week Substack launched AI detection via Pangram, sharpening a live debate: does removing AI "tells" from writing matter if the writing was still AI-assisted?

What Peter Yang released

@petergyang announced the skill on X:

"I'm sick of reading AI slop, so today I'm open-sourcing my /no-ai-slop skill that removes 20+ slop patterns from any piece of writing."

He listed three representative patterns the skill targets:

• Binary contrasts — "It's not X. It's Y."
• Throat-clearing openers — "Here's what nobody tells you."
• Fake-profound endings — "The future isn't coming. It's already here."

These are the kind of stylistic tics that have become recognizable enough to be their own meme — patterns that show up disproportionately often in AI-generated prose, regardless of topic, because they're statistically common completions the underlying models gravitate toward. Yang's skill is a targeted editing pass aimed specifically at these tells, not a general grammar or style checker.

The GitHub repo is at github.com/petergyang/no-ai-slop. Yang wrote a companion explainer post at creatoreconomy.so covering the reasoning and responsible-use guidance in more depth.

TL;DR — what the skill does

How Yang says he uses it responsibly

Yang was explicit that the skill is meant to assist editing, not replace drafting:

"I use AI to edit my writing because it helps me fix spelling, grammar, and clarity. But even the best models keep producing the same slop that this skill removes... Use this skill responsibly. I always write a first draft manually before iterating with AI on edits and I make sure to do another manual pass at the end as well. That's in contrast to using AI to automate pumping out slop end-to-end."

That workflow — manual first draft → AI-assisted edit pass → manual final pass — is a specific, checkable claim about process, not just tone. It matters because the value proposition of /no-ai-slop depends heavily on when in the writing process you use it. Applied to a genuinely human-drafted piece, it's a targeted cleanup tool. Applied to a fully AI-generated first draft with no human authorship at any stage, it's closer to a disguise — the same underlying critique that fuels skepticism about AI writing generally.

This is analogous to the difference explainx.ai has covered between using agent skills as scoped, auditable tools for a specific task versus using them to fully automate a workflow end-to-end without human review — the tool itself doesn't determine which pattern you're following; the workflow around it does.

Landing the same week as Substack's AI detector

The timing here is not coincidental context — it's the same news cycle. Substack launched an AI detection feature on July 21, 2026, integrating with Pangram. The feature scans posts, notes, replies, and comments over 100 words published since July 21, showing an estimated split between human-written, AI-assisted, and fully AI-generated content, live on web and iOS with Android coming.

Substack CEO Chris Best (@chrisbest) has framed the goal as preventing the platform from becoming "like LinkedIn" — flooded with low-quality AI-generated posts — while writers frame the detector as validation for genuine effort.

Putting the two stories side by side surfaces the real tension:

Removing an em-dash-heavy, binary-contrast-laden AI voice makes writing more pleasant to read, but it doesn't necessarily change what a detector like Pangram measures, since detectors work off statistical text patterns beyond just surface stylistic tics — token-level distributions, sentence-length variance, and other signals that a targeted style edit may not touch. Whether /no-ai-slop-cleaned text still gets flagged by Pangram-style detection is an open, testable question nobody has published data on yet.

Reactions and the deeper debate

Shreyas Doshi, a prominent product commentator, weighed in with a contrarian framing:

"It really is this simple: [flowchart: Is it useful? → Yes → Then why does it matter that AI wrote it? Take what's useful, ignore the rest. → No → same box]"

Doshi's flowchart argues that usefulness should be the filter, not origin — if a piece of writing serves the reader, its AI/human provenance is beside the point. He followed up clarifying the nuance:

"The vast majority of people simultaneously read much more than what is written and much less of what is written... 'Shreyas loves AI writing' would be the wrong conclusion. 'Shreyas hates AI writing' would also be the wrong conclusion."

Other reactions pointed at a harder problem underneath the slop-removal conversation — detector false positives on human writing:

"i'm 1000% for AI slop to be removed from the world. also, when i run my old tweets from 2021/2022 into these programs they come back ~80% AI. The world has people who wrote a certain way and the LLMs learned how to write from those people. And now their way of writing can be [flagged]" — @AdamRy_n

This is the core detector-reliability problem: because LLMs were trained on human writing, some genuinely human stylistic patterns overlap with what detectors flag as AI-typical. A writer whose natural voice already leans toward certain rhythms or phrasings can get misclassified — which is exactly why Substack's rollout of visible AI-percentage scores is contentious among writers who never used AI at all.

And the arms-race dynamic showed up immediately:

"i'm launching a new startup. we're training a custom model whose sole purpose is to defeat ai writing detectors... every detector will confidently certify it as 100% human." — @signulll

Whether that's a genuine venture or a joke, it captures the structural problem with detection-vs-evasion: any classifier trained to catch AI patterns creates an incentive to build a generator or editor that specifically defeats that classifier — the same cat-and-mouse dynamic seen in spam filtering, plagiarism detection, and adversarial ML generally.

What this means for writers and builders

If you write with AI assistance:

• /no-ai-slop is worth trying as a targeted style pass — it's free, open source, and addresses a real, widely-recognized problem (readers actively dislike the recognizable AI voice).
• It is not a substitute for drafting your own ideas. Yang's own stated workflow (manual draft → AI edit → manual pass) is the credible version of "using AI responsibly" here; skipping the manual drafting step changes what the tool is actually doing for you.
• Don't assume style-cleanup guarantees you'll pass a detector like Pangram — those are different signals.

If you're evaluating agent skills generally:

• This is a good example of a narrow, single-purpose Claude skill — it does one job (pattern removal in text) rather than trying to be a general writing assistant. Explainx.ai's agent skills guide covers this pattern in more depth: scoped tools tend to be more trustworthy and auditable than broad, do-everything agents.
• Watch the GitHub repo's issue tracker as it grows past 1,000 stars — community-reported false positives/negatives on the pattern list will be a good signal of how well it generalizes beyond Yang's own writing style.

Related on explainx.ai

• Substack launches AI detection with Pangram — what it flags and how
• What are agent skills? Complete guide
• AI token black market — Claude resellers and distillation
• AI skills developers need in 2026 — roadmap

Primary sources: @petergyang on X · github.com/petergyang/no-ai-slop · creatoreconomy.so companion post · @Substack AI detection announcement

Details reflect Peter Yang's July 22, 2026 release and public X reaction as of publication. GitHub star counts and repo contents change continuously — check the repo directly for the current pattern list and installation instructions.