---
name: editorial-style-matcher
description: "Rewrite a draft to match a target publication's house style (sentence length, voice, diction) without changing meaning."
title: Editorial Style Matcher
category: writing-editing
difficulty: beginner
license: MIT
author: admin
source_url: "https://github.com/martiansideofthemoon/style-transfer-paraphrase"
icon: ✒️
input: text
output: text
phase: transform
domain: content
tags: style-transfer,editorial-rewriting,copy-editing,house-style-matching,tone-adaptation,corpus-analysis,semantic-preservation,statistical-validation,publication-alignment,prose-decomposition,readability-metrics
best_for:
  - multi-article publication workflow standardization
  - freelance writer submission prep for target journals
  - brand voice consistency across contributor content
  - academic paper adaptation for journal submission
  - blog rewrite for magazine house style
---

## Description

A style-transfer skill: takes a draft and a target style profile (derived from a sample corpus of the target publication) and rewrites the draft to match. Preserves factual claims, adjusts tone, sentence rhythm, and characteristic vocabulary.

## Why it works

Writers can usually describe a publication's style ('terse, punchy, no adverbs'), but rewriting their own prose to match is exhausting and inconsistent. The skill separates 'what to say' (unchanged) from 'how to say it' (rewritten), which is the decomposition human copy editors use. Extracting the style profile from a real corpus of 20+ articles — rather than a vibes-based prompt — grounds the target in measurable stats (avg sentence length, Flesch score, named-entity density).

## How it works

1) Corpus phase: compute style statistics from 20+ target articles — sentence length distribution, syllables/word, passive-voice rate, comma density, characteristic n-grams. 2) Draft phase: decompose the input into atomic claims (a list of what is asserted, stripped of style). 3) Rewrite phase: instruct the LLM to re-express each claim matching the measured style targets. 4) Validation: re-measure stats on the rewrite; if any metric is > 1σ off the target, request a targeted re-run. 5) Meaning-preservation check: embed-similarity each rewritten claim vs original; anything below threshold gets flagged for human review.