AI Style Cleaning Master for Serialized Web Novels

Problem

Long-form web novels often suffer from visible AI-generated patterns: repeated formulaic phrases, over-explained emotions, templated action descriptions, and characters whose dialogue sounds indistinguishable. Serialized platforms also reward readable pacing and chapter-end hooks, so the issue is not just “polish” but removing narrative habits that make text feel mechanical.

How It Works

The skill processes chapters through a structured cleaning pipeline: scan, layered revision, validation, and scoring.
- Layer 1 scanning: detects high-frequency clichés by genre, such as exaggerated combat phrases or generic physical reactions, and estimates AI density.
- Layer 2 sentence restructuring: fixes “not X, but Y” constructions, adverb stacking, long modifiers, passive voice, and overused connectors.
- Layer 3 narrative rewriting: for high-scoring chapters, removes metaphor, commentary, and explicit emotional instruction, keeping facts and actions.
- Layer 4 character voice separation: adjusts sentence length and tone by character type so different roles do not sound alike.
- Validation and scoring: checks tone, pacing, emotional expression, and readability, then outputs a replacement-ready full chapter revision.

Boundaries

It is useful for batch cleanup in serialized novels, especially when chapters need de-AI flavor, tighter pacing, or stronger character voice. It is not meant for single-sentence grammar fixes, outline drafting, or word-count checks. Avoid cleaning too many chapters at once, do not over-polish dialogue-heavy sections, and sample edits afterward to confirm distinctive expressions have not been flattened.