What problem does it solve? Japanese text generated by AI often carries detectable habits—stock phrases, translationese, monotonous rhythm, uniform paragraph structure—that make documents read as machine-written. This Skill detects those patterns deterministically with lint scripts and guides a structured design-write-inspect-converge workflow to produce natural, readable Japanese business documents. ## Core Features & Use Cases - Deterministic AI-smell detection: Run lint.py to flag forbidden phrases, translationese, antithesis repetition, sentence-length uniformity, and lexical diversity issues, with corpus-calibrated thresholds and genre profiles (essay/tech/business). - Write and rewrite workflows: Draft or revise minutes, reports, guides, memos, slide outlines, and blog essays through a design → writing → inspection → convergence loop, with quick and full execution modes. - Diagnosis-only scoring: Score a document's naturalness from 0-100 without rewriting it, with band labels and top reasons, optionally including semantic embedding analysis. - Use Case: Paste a meeting-minutes draft and ask for a naturalness check; the Skill runs the linter, reviews structure and readability, and returns a revised version with a judgment ledger of what was changed or kept. ## Quick Start Ask the AI to rewrite the attached Japanese report so it reads naturally and does not sound AI-generated.
