Zero Slop Open-Sources an AI Writing Editor for Removing Clichés

Summary

Zero Slop is a free, open-source agent skill for finding and removing formulaic “AI slop” from drafts while checking that the original message survives the edit. It can be used through a browser editor, compatible assistants such as Claude Code and Codex, hosted MCP, a CLI, or a REST API. Its local scorer uses 294 weighted patterns and a 96-term lexicon to flag canned openings, binary contrasts, vague attribution, inflated significance, promotional wording, repeated sentence structures, crowded statistics, and overworked formatting. The workflow has eight stages covering scoring, interpretation, rewriting, fact checks, copy editing, read-aloud review, source comparison, and final verification; these stages combine Python tools with the user’s AI application rather than eight separate models. In a saved test on an 18-item corpus using GPT-5.4, Zero Slop reached a mean writing score of 12.8, passed all 18 local and source-detail gates, and reduced length by 8.9%. A separate audit of 7,627 anonymous transcripts found mean scores from 14.5 for DeepSeek V3 to 25.5 for Llama 3.3 70B, but the project emphasizes that the score cannot identify an author and may misclassify non-native English. Local scoring can run without a model call, while hosted editing sends text to the service. The CLI requires Node.js 22+, and the project is released under the MIT license.