Stripping AI Clichés from LLM Outputs with Stop-Slop Skill Files
Stop Slop is an open-source system prompt skill file designed to eradicate typical AI-sounding phrases from generated text. Keep your documentation, UI copy, and changelogs completely natural.
Why it matters
It instantly purges cheesy AI jargon from your generated code comments, pull request descriptions, and user documentation.
TL;DR
- 01Copy the Stop Slop rules list into your .cursorrules or Claude system instructions
- 02Add your own custom domain-specific jargon words to the negative prompt blacklist
- 03Use few-shot rewrite templates to show the AI exactly how you prefer sentences to be structured
Patterns of AI Prose
AI-generated writing is riddled with predictable patterns like 'in today's digital landscape' or excessive use of adverbs. The Stop Slop project provides a structured set of rules to identify and strip these clichés. It covers structural issues, such as dramatic fragmentation, as well as sentence-level rules like banning specific filler phrases.
Integration Strategies
- Claude Code: Add the skill folder directly.
- Claude Projects: Upload the
SKILL.mdfile to your project knowledge. - API Calls: Include the rule definitions directly within your system prompt.
- Custom Instructions: Copy the core rules to your environment configuration.
By enforcing active voice and banning specific tropes, developers can ensure that generated documentation and commit messages maintain a professional, human-like tone.