Humanize Korean Ai Text

If you've ever pasted ChatGPT or Claude output into a Korean document and cringed at how obviously machine-generated it reads, this is built for that. It runs pattern detection across 40+ tells organized into 10 categories: translation artifacts like "~를 통해", hedging stacks like "~할 수 있을 것으로 보인다", mechanical bullet structures, and AI clichés like "시사하는 바가 크다". Fast mode processes under 5,000 characters in 2-3 minutes with a single agent pass. Strict mode runs a five-agent pipeline with separate detection, rewriting, fidelity audit, and naturalness review stages, saving each step to disk. It caps modification rate at 30% to prevent over-editing and won't touch numbers, proper nouns, or quoted text. Results land in timestamped workspace folders with before/after diffs and severity scores.