humanizer-kill-gptlisted

# Humanizer: Kill GPT Tells When starting, let the user know in one line that this skill is being used. Remove recurring AI-writing fingerprints from English and Korean text. Despite the skill name, these patterns are not unique to GPT: large frontier language models commonly converge on the same over-complete, over-explained, overly balanced, and translation-shaped prose. The catalog keeps “GPT tell” as a familiar shorthand. Some users call this style `자폐 말투`. Treat that phrase only as colloquial request vocabulary for rigid or socially miscalibrated model prose. MUST NOT equate the writing patterns with autism or use the phrase to characterize autistic people. Most translated Korean tells start upstream: an English model-writing construct survives literal translation. This skill therefore uses two coupled passes. | Text situation | Run | |---|---| | English draft | Pass A only | | Korean draft written natively | Pass B only | | Korean translated from English, or English that will be translated to Korean | Pass A first, then Pass B | This skill targets cross-language patterns shared by large frontier models. A broader Korean-language cleanup workflow may cover additional punctuation, spacing, and lexical-distribution patterns that are outside this skill's scope. When the user asks for generation-time speaking guidance instead of post-editing, read [references/generative-speaking-directive.md](references/generative-speaking-directive.md). Do not invert the catalog into