A prose linter that catches AI writing patterns. Based on Wikipedia's Signs of AI Writing.
AI slop:
React Query is a groundbreaking library that serves as a pivotal tool in the modern frontend landscape. It seamlessly handles data fetching, caching, and synchronization, showcasing a meticulous approach to state management. The library boasts a vibrant community and has garnered significant adoption, underscoring its enduring value in the ecosystem.
After no-slop:
React Query handles data fetching, caching, and background sync for React apps. You describe what data you need, and it handles refetching, deduplication, and cache invalidation. The community is large — over 40k GitHub stars — and most major React codebases have adopted it.
Same information. No slop.
| # | Rule | What it stops |
|---|---|---|
| 1 | Banned vocabulary | 40+ words like "delve," "tapestry," "pivotal," "vibrant," "leverage" |
| 2 | Simple copulas | Replacing "is/has" with "serves as," "stands as," "boasts" |
| 3 | Promotional tone | "Groundbreaking," "revolutionary," marketing-speak in technical writing |
| 4 | Vague attributions | "Experts say," "industry reports suggest" with no named source |
| 5 | Structural formulas | Rule of three, "not just X but Y," challenges-and-prospects endings |
| 6 | Participle chains | "-ing" filler: "highlighting," "showcasing," "emphasizing" |
| 7 | Elegant variation | Swapping synonyms for the same thing across sentences |
| 8 | Overstating significance | "Marks a turning point," "leaves an indelible mark" |
| 9 | Em dash overuse | More than one per paragraph |
| 10 | Collaborative language | "Let's explore," "as we can see," "we will examine" |
| 11 | Knowledge-cutoff disclaimers | "As of my last update," apologizing for gaps |
| 12 | Formatting excess | Over-bolding, emoji, title-case headings, "key takeaways" |
| 13 | Human writing habits | Contractions, varied sentence length, specifics over adjectives |
Every rule is sourced from Wikipedia:Signs of AI writing, maintained by WikiProject AI Cleanup from observations of thousands of AI-generated texts.
# Clone into your skills directory
git clone https://github.com/Byk3y/no-slop.git ~/.claude/skills/no-slopThat's it. Claude Code auto-discovers skills in ~/.claude/skills/. It will activate whenever you ask it to write prose.
You can also trigger it manually:
/no-slop
Copy the contents of agents/cursor.md into your .cursorrules file at your project root.
See agents/codex.md for setup instructions.
Automatic: Just ask your AI to write something — a blog post, email, PR description, documentation. The skill activates on its own for writing tasks.
Manual: Use /no-slop (Claude Code) to force-activate it when auto-detection doesn't kick in.
What it does NOT activate for: Code generation, short chat replies, commit messages. It only applies to prose.
no-slop is a constraint skill, not a rewriter. It doesn't take your text and fix it after the fact. It teaches the AI what patterns to avoid while writing, so the first draft comes out clean.
The rules come from Wikipedia's article on signs of AI writing — a living document maintained by editors who review thousands of AI-generated texts. This isn't one person's opinion. It's a collaboratively maintained, evidence-based list of the patterns that give AI writing away.
no-slop/
├── SKILL.md # Core rules (13 constraints)
├── banned-vocabulary.md # 40+ banned words with plain alternatives
├── examples/
│ ├── bad-examples.md # Annotated AI-sounding text
│ └── good-examples.md # Human-sounding rewrites
├── agents/
│ ├── claude-code.md # Claude Code install guide
│ ├── cursor.md # Cursor rules format
│ └── codex.md # Codex CLI format
├── CONTRIBUTING.md
└── LICENSE
| no-slop | humanizer | anti-ai-slop-writing | avoid-ai-writing | |
|---|---|---|---|---|
| Approach | Constraint (prevents slop) | Rewriter (fixes after) | Banned words | Tiered word replacement |
| Source | Wikipedia (evidence-based) | Wikipedia | Wikipedia + CMU | Mixed |
| Annotated examples | Yes (before/after per rule) | One example | No | One example |
| Conditional bans | Yes (context-aware) | No | No | No |
| Multi-agent | Claude Code, Cursor, Codex | Claude Code, OpenCode | 11+ agents | Claude Code, OpenClaw |
All rules are derived from Wikipedia:Signs of AI writing, a collaboratively edited reference maintained by WikiProject AI Cleanup. The article documents patterns observed across thousands of AI-generated texts by Wikipedia editors.
See CONTRIBUTING.md. The short version:
- New banned words need evidence (link to Wikipedia article or published research)
- New examples need annotations explaining what's wrong and why the rewrite is better
- PRs should be focused — one rule or one example set per PR
If no-slop made your writing better, give it a star. It helps others find it.