A 500-line rulebook for making AI text sound less like AI
A SKILL.md that bans the words, phrases, and structural tics that make AI writing instantly recognizable.
What it does
It’s a writing skill file — essentially a long list of prohibitions — that gets injected into any AI agent’s output pipeline. When the model writes a tweet, email, or article, the skill filters out the vocabulary, sentence openers, structural habits, and formatting quirks that readers (and detectors) associate with machine-generated prose. It ships as a Claude Code skill but the SKILL.md format works across Cursor, Codex, Gemini CLI, and other agents, or can just be pasted into any chat.
The interesting bit
The rules aren’t vibes — they’re compiled from named sources: Carnegie Mellon research (2025), Wikipedia’s “Signs of AI Writing” page, Buffer’s analysis of 52 million posts, and community detection patterns from X and Reddit. It’s essentially a blacklist-as-a-service for statistical tells, from “delve” to the rule-of-three habit to em dash overuse.
Key highlights
- 50+ banned words and 35+ banned phrases drawn from multiple detection studies
- 16 banned sentence openers (“Certainly,”, “Moreover,”, etc.) and 10 structural patterns like hedging seesaws and uniform sentence length
- Catches formatting leaks too: markdown in plain text, emoji bullets, hashtag stacks
- Also targets accuracy failures — invented statistics, fabricated quotes, fake anecdotes
- Core rules stay under 500 lines, with the full banned vocabulary loaded on demand
Caveats
- The README doesn’t quantify how effective the constraints actually are — no benchmarks on detection rates before/after
- “Works with 11+ agents” is asserted but the compatibility list beyond the big names isn’t shown
Verdict
Worth a look if you produce AI-assisted writing and cringe at the output; it’s a thoughtful, source-backed blacklist. If you need proof it beats detectors, that evidence isn’t in the README.
Frequently asked
- What is jalaalrd/anti-ai-slop-writing?
- A SKILL.md that bans the words, phrases, and structural tics that make AI writing instantly recognizable.
- Is anti-ai-slop-writing open source?
- Yes — jalaalrd/anti-ai-slop-writing is an open-source project tracked on heatdrop.
- How popular is anti-ai-slop-writing?
- jalaalrd/anti-ai-slop-writing has 620 stars on GitHub.
- Where can I find anti-ai-slop-writing?
- jalaalrd/anti-ai-slop-writing is on GitHub at https://github.com/jalaalrd/anti-ai-slop-writing.