Fiction Polish Distill Expert

Problem

AI-generated fiction often shows measurable stylistic artifacts: repetitive connectives, unnaturally stable sentence lengths, formulaic endings, and term-heavy passages. Manual edits can also damage the author’s original intent, terminology, or voice. This skill turns those corrections into an inspectable workflow rather than a black-box rewrite.

How it works

• Diagnose first: users can scan a passage before choosing de-AI, polish, distillation, or imitation, so the process stays optional and targeted.
• De-AI and polishing: rules focus on phrase blacklists, sentence-length variation, repetition, connective density, dialogue ratio, and terminology lock, preserving meaning while reducing mechanical patterns.
• Style distillation and imitation: the skill extracts lexical, syntactic, tonal, and structural parameters from multiple samples, then applies them to rewrites; it includes a built-in writer-style library for Chinese, English, and Japanese authors.
• Multilingual support: Chinese, English, Japanese, and Korean modes are supported, with fallback to English for other languages.
• Checks and learning: quality_check.py can report anti-AI score, content-safety results, and statistical features; evolve_blacklist.py manages the blacklist used by the pipeline.

Boundaries

It does not promise to evade AI detection. Very short inputs, fewer than three style samples, uncertain content safety, and platform disclosure obligations still require human judgment.