$ show iKora128/stop-ai-slop-jp
This repository provides a Claude Skill for detecting and removing 'AI slop' from Japanese prose. It consists of a set of guidelines (SKILL.md) and reference files that enumerate problematic phrases, structural patterns, and examples of AI-generated text versus human-revised versions. The skill focuses on restoring human agency, stance, specificity, rhythm, and reduction in LLM-generated Japanese text. It is designed to be used during drafting, editing, or review stages to improve the authenticity of AI-assisted writing.
The LLM identified 3 files worth highlighting for reuse:
The core skill document containing principles, prioritized rules (agency, stance, structure, vocabulary, symbols), quick checklists, and a 5-axis scoring system. It is self-contained and can be extracted as a standalone guide for any text-editing workflow focused on removing AI-like artifacts from Japanese prose.
A comprehensive list of phrases and lexical patterns to avoid, categorized into eight groups (e.g., AI-favored nouns, translationese, hedges, IT metaphors). The list is independent, easily portable, and can be reused in any text analysis or rewriting tool targeting Japanese language quality.
Details six systemic structural patterns that generate AI-like prose (false agency, fabricated drama, imported English blog structures, uniform rhythm, evasion of stance, harmless tuning). Each pattern includes AI vs. human examples and concrete rewriting advice, making it a reusable reference for detecting non-human writing tendencies.