im-not-ai is a Claude skill that detects and rewrites Korean text generated by AI to read naturally, without altering content.
The tool addresses the problem that English-language humanizers perform poorly on Korean because AI-generated Korean text exhibits distinctive patterns rooted in translationese and mechanical phrasing rather than the errors typical of English AI writing. The approach identifies patterns across ten major categories spanning over seventy subcategories, classifying them by severity level, then rewrites detected spans to restore natural Korean rhythm, style, and expression. The tool preserves every word of the original content while modifying only phrasing, tone, and structure.
Developers working with Korean-language content generated by large language models should consider this tool if they need to post-process AI output for publication or presentation. The project is specifically designed for Korean and would not suit projects in other languages. The tool offers three processing paths of varying depth—light, standard, and heavy—allowing users to balance speed and cost against thoroughness; well-written input requires fewer API calls. Installation is available through plugin marketplaces for Claude Code and GitHub Copilot CLI, or via direct repository cloning with provided scripts.
The project maintains active development with regular updates distributed through automated detection and installation scripts. The codebase is written in Python and includes comprehensive installation documentation covering multiple integration paths. The tool has been refined through iterative pattern identification, with the README documenting specific examples of translationese and mechanical phrasing that the system targets, indicating ongoing refinement based on observed AI writing characteristics in Korean.
