realrossmanngroup/no_ai_slop_writing_rules

This repository is a portable set of writing rules designed to stop AI-assisted text from sounding like it was written by an AI. It is packaged as an Agent Skills bundle, which means you drop it into a coding agent like Claude Code, Cursor, or Windsurf and the agent reads the rules before generating any prose. The rules come in two parts. The first is an anti-slop checklist: a long reference list of phrases, patterns, and habits that mark machine-generated writing. This includes things like intensifier words, hollow one-sentence paragraphs used for fake drama, dramatic section headings that tease rather than describe, AI transition phrases, and specific constructions like "It's not X. It's Y." that the README singles out as a dead giveaway. The second part is a voice profile modeled on Louis Rossmann, a well-known YouTube commentator and repair-shop owner. The profile was built from corpus analysis of over 500,000 words of his writing and captures measurable traits: high variance in sentence length, a specific density of concrete numbers, a claim-then-proof paragraph structure, and using precision to show skepticism rather than adjectives. The repository includes a CLAUDE.md file (the entrypoint for Claude Code), a cross-agent AGENTS.md, two skill files with wrong-versus-right worked examples, and a full banned-words reference organized by category. Every file in the repo is written to follow its own rules, so the files themselves serve as style examples. You can install it with a single npm command, through the Claude Code plugin marketplace, or by cloning and copying the files manually. The voice examples in the skill use repair and consumer-rights topics because that is what the source corpus covers, but the rules are structural and the authors say they apply to any subject.