What does this skill do, and when should you use it?
No AI Slop is a writing-editing skill that sharpens drafts into clearer, more human prose without flattening the writer's voice. It ships a detailed rulebook covering 20+ named AI patterns (binary contrasts, throat-clearing openers, fake-profound endings, and more) plus an explicit banned-word list (delve, leverage, robust, etc.). It runs in two modes: the default edit mode returns the revised draft plus a "What changed" section, while detect mode names each pattern with a quoted line and never guesses whether AI wrote the text. The skill is pure SKILL.md instructions with no code dependencies, so it works in any Agent Skills-compatible client.
Reads a pasted draft and performs one of two jobs. Edit mode: makes the minimum effective changes — cutting banned words, empty qualifiers, and filler phrases; rewriting binary contrasts, colon reveals, superficial -ing analysis clauses, and fake-profound kickers; converting passive and weak verb constructions to direct verbs; replacing abstractions with concrete facts — then outputs the full edited draft with a short change summary. Detect mode: quotes each slop pattern it finds with the offending line and a few-word fix, without rewriting, scoring, or judging authorship. After editing, the skill self-checks the draft against the checks in eval.md. If the draft, audience, or goal is unclear, it asks the user first.
- A blogger or newsletter writer who drafts with AI and wants obvious AI-sounding lines removed before publishing
- An editor reviewing a submission who wants specific slop patterns identified without the text being rewritten
- A technical writer who wants drafts more direct and opinionated but resists a generic "polished" uniform style
- A marketer using ChatGPT for copy who needs template sentences like "It's not X, it's Y" replaced with substantive statements
- Anyone who wants quotable evidence of AI patterns in their own writing instead of a probabilistic AI-detector verdict
How do you install this skill?
- Edit quality depends entirely on subjective model execution with no automated guarantee; review important text manually.
- No Chinese-writing rules; effectiveness on Chinese text is undeclared.
- The self-check workflow assumes eval.md is present in the workspace; installing SKILL.md alone may break that step.
- No version numbers or changelog; updates are not trackable.
- Local filesystem
eval.md (self-check reference file in skills/no-ai-slop/)
Option 1: paste into ChatGPT, Claude Code, Codex, or your coding agent: "Install the /no-ai-slop skill globally from https://github.com/petergyang/no-ai-slop". Option 2: install with npx: npx skills add petergyang/no-ai-slop --skill no-ai-slop --global --yes. The core file lives at skills/no-ai-slop/SKILL.md in the repo. It is also available as a ChatGPT plugin (metadata in .codex-plugin/plugin.).
tmp="$(mktemp -d)"
git clone --depth 1 https://github.com/petergyang/no-ai-slop.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/skills/no-ai-slop" ~/.claude/skills/
rm -rf "$tmp"Generated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- /no-ai-slop (your writing)
- /no-ai-slop is this slop? (your writing)
- Draft an AI slop post about (topic)
Edit mode: send "/no-ai-slop (your writing)" and get back the revised draft plus a What changed list. Detect mode: send "/no-ai-slop is this slop? (your writing)" to get each found pattern quoted. For fun: ask it to "Draft an AI slop post about (topic)" to generate satirical slop. If no draft is supplied, or the audience or goal is unclear, the skill asks one clarifying question before editing.
What are this skill's strengths and limitations?
- Rules are concrete and verifiable: 20+ named patterns plus a banned-word list, each quoted from the text rather than vague "make it better" advice
- Explicitly prioritizes preserving the writer's voice and makes the minimum effective edit
- Detect mode honestly refuses to guess authorship and only offers checkable pattern evidence
- Pure-instruction implementation, MIT licensed, no runtime dependencies, highly portable
- Includes an eval.md self-check the skill runs on its own edits
- No GitHub topics and no automated test suite in the repo — eval.md is a checklist the model self-applies, not independent test code
- Rules and examples are English-only; no guidance on whether it works for non-English writing
- The long rule list can drive large edits on long drafts; you may want to verify "minimum edit" claims manually
- Detect mode names patterns but provides no quantified score, so it won't suit anyone needing a probability-style AI-detection verdict
- Results depend entirely on the model executing the text rules; quality may vary across models/platforms, and the repo offers no comparative benchmarks
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| No AI Slop this page | 59 · Recommended | ★ 13k | 2mo ago | MIT |
| Unslop | 49 · Use with care | ★ 12k | 6d ago | Apache-2.0 |
| Humanizer Writing Skill | 53 · Use with care | ★ 66k | 9d ago | MIT |
| Beautiful Prose Style Contract | 50 · Use with care | ★ 61 | 12d ago | — |
| e2e Docs Authoring Skill | 54 · Use with care | ★ 12k | 6d ago | Apache-2.0 |
How did FollowSkills review this skill?
Pure prompt-based editing instructions: no scripts, no network calls, no external side effects, no sensitive-data handling; MIT license clearly stated. Deducted for absence of versioned release verification and unverified publisher (unknown, not suspicious).
SKILL.md is self-consistent with a clear workflow (read, identify voice, edit, self-check via eval.md, output), and eval.md defines a failure-feedback loop. Deducted because eval checks are subjective ('preserve voice'), no automated tests cover key paths, and static review cannot verify execution — capped at 10 by calibration.
Trigger conditions are explicit: two well-separated modes (edit default, detect), boundaries declared (will not guess AI authorship), asks user for missing drafts. Deducted for no Chinese-language support indication; rules center on English idioms and word lists, and boundaries rely on subjective judgment.
Well-organized docs: SKILL.md rules, eval.md checks, README with install/use, MIT license, CI validating the plugin package. Deducted for no versioning/changelog, no stated maintenance ownership or update path, and a hidden dependency on eval.md being present in the workspace.
Instructions are concrete and actionable with a detailed pattern catalogue and defined output format (edited draft plus What changed), likely better than generic de-slop prompts. Deducted because static review cannot confirm edit quality, outcomes depend on subjective model execution, and no representative outputs or third-party usage evidence exist — capped at 7.
Repo includes eval.md self-checks and CI, but CI only validates plugin packaging, not editing quality; README links a demo video whose content cannot be verified statically. Deducted for no third-party execution evidence and no independently reproducible tests — capped at 5.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →