humanizer_skill

Overview

This skill detects characteristic AI-generated patterns in Korean text and rewrites the content to sound natural and human-authored while preserving meaning. It is grounded in empirical linguistics research (KatFishNet) and targets 20 validated markers such as comma overuse, rigid spacing, low POS diversity, AI buzzwords, and monotonous structure. The aim is readable, culturally appropriate Korean that retains the original tone and intent.

How this skill works

The skill ingests Korean text (direct input, conversation context, or file) and runs a systematic analysis across five categories: punctuation, spacing, part-of-speech diversity, vocabulary, and structure. For detected markers it cites the scientific rationale (e.g., KatFishNet AUC metrics) and applies corrections that preserve factual content and the original formality level. The output includes a rewritten, human-like version and a concise summary of major changes when relevant.

When to use it

• You want to make LLM-generated Korean read like native human writing before publishing.
• You need to remove detectable AI traces while preserving meaning and tone.
• You are editing Korean marketing, academic, or business copy produced by a model.
• Short social posts or long reports created by ChatGPT/Claude/Gemini require naturalization.
• You want diagnostic feedback on why text looks AI-generated and how to fix it.

Best practices

• Provide the full Korean text and indicate desired formality (formal vs. casual).
• Accept small, natural irregularities; avoid forcing uniform stylistic 'perfection'.
• For very short inputs, acknowledge limited detection power and focus on clear markers.
• Preserve technical terms and one-word English insertions; correct surrounding Korean patterns.
• Request a change log when you need justification for key edits or educational feedback.

Example use cases

• Convert a model-written product description into natural, persuasive Korean for a web page.
• Polish academic or policy text generated by an LLM while keeping formal register intact.
• Review and humanize social media captions that sound repetitive or overly generic.
• Prepare client-facing reports by removing comma-heavy, machine-like phrasing.
• Assess a draft to identify which AI-writing markers appear and why they matter.