Skip to content

Repository files navigation

Polish Chinese Fiction — an editorial brush untangling prose above a cloud sea

Polish Chinese Fiction

A full-book editorial system for Chinese fiction.
From plot bones to native Chinese prose—edit the whole novel without breaking the novel.

中文小说精修工作台
从故事骨架改到一字一句,把文字改顺,也把剧情、人物和伏笔守住。

English · 简体中文

Package validation Codex Skill Node.js 18+ Zero runtime dependencies Chinese


Most writing tools polish whatever sentence happens to be in front of them. A novel needs more discipline: protect the plot, keep every character's voice, trace every change, and make sure the twelfth revision has not quietly duplicated a paragraph or displaced an illustration.

Polish Chinese Fiction is a production-tested Codex skill for that job. It separates editorial work into six gates, runs only the gates a manuscript actually needs, and ships deterministic scripts for the parts that should never depend on taste or memory.

Not a one-click “rewrite everything” prompt. Scanners find candidates; the editor still judges context.

What it can do

  • Build the book before polishing the sentences — premise, world rules, character arcs, romance beats, clues, payoffs, and chapter turns.
  • Repair continuity from the earliest causal break — ages, timelines, knowledge boundaries, objects, identities, institutions, and consequences.
  • Rewrite scenes without flattening voices — objective, resistance, turn, aftermath, subtext, and relationship movement.
  • Remove translationese at the mechanism level — English-style pronoun mapping, object agency, self-observation, nominalized conclusions, unnecessary passive voice, and report-like explanation.
  • Restore emotional temperature — colloquial adverbs, modal force, punctuation, direct emotion, pauses, and character-specific speech rhythms.
  • Keep revisions auditable — unique baseline matches, automatic chapter location, change ledgers, deterministic replay, hashes, and structural checks.
  • Protect publication outputs — headings, paragraph counts, illustration anchors, DOCX/PDF/site text, public copy, and clean reading editions.

Six editorial gates

Six editorial gates: architecture, continuity, scene and voice, native Chinese, public copy, publication QA

Use only what the current draft needs:

Gate Question Typical output
A · Architecture Does the story work? Story bible, act and chapter architecture
B · Continuity Does every cause still lead to the right effect? Continuity ledger, targeted repairs
C · Scene & Voice Does each scene turn, and does each person sound like themselves? Scene rewrite, voice bible
D · Native Chinese Would a native Chinese writer naturally phrase it this way? Revised manuscript, complete change table
E · Public Copy Is the reader seeing the book rather than the production process? Blurb, metadata, captions, clean edition
F · Publication QA Did every derivative preserve the same book? Hashes, structure, anchor and browser checks

Translationese, seen under an editorial X-ray

Rigid machine-like syntax becoming fluid Chinese prose

The skill treats a bad sentence as a defect-family sample, not a one-line exception. If a reader flags an English-style it/that mapping, the workflow searches the whole manuscript for the same mechanism—then revises each occurrence in paragraph context.

Before After What changed
“它认识什么?” “信号是什么?” Names the concept instead of assigning cognition to an object.
她察觉自己后退了一步。 她已经退了一步,鞋跟抵住门槛才停下。 Replaces self-observation with action and physical consequence.
三十日从现在算。 三十日之期,从此刻算起。 Restores the missing time frame and natural register.

Install

macOS / Linux

git clone https://github.com/ding7015869-alt/polish-chinese-fiction.git
mkdir -p "${CODEX_HOME:-$HOME/.codex}/skills"
cp -R polish-chinese-fiction/skill/polish-chinese-fiction \
  "${CODEX_HOME:-$HOME/.codex}/skills/"

Windows PowerShell

git clone https://github.com/ding7015869-alt/polish-chinese-fiction.git
$skills = if ($env:CODEX_HOME) {
  Join-Path $env:CODEX_HOME "skills"
} else {
  Join-Path $HOME ".codex\skills"
}
New-Item -ItemType Directory -Force -Path $skills | Out-Null
Copy-Item -Recurse -Force \
  "polish-chinese-fiction\skill\polish-chinese-fiction" $skills

Restart Codex after installation. The skill triggers on Chinese fiction editing requests, or invoke it explicitly:

Use $polish-chinese-fiction to line-edit this manuscript for native Chinese,
remove translationese globally, preserve plot and character voice,
and deliver a traceable revision ledger.

Quick start

1. Ask for the editorial layer you need

Use $polish-chinese-fiction to audit continuity only.
Freeze the prose, list the earliest causal break in each thread,
and propose the smallest downstream repairs.
Use $polish-chinese-fiction to perform a native-Chinese final pass.
Treat my examples as defect families, search the full manuscript,
and do not change plot, paragraph breaks, or illustration anchors.

2. Scan for candidates

node skill/polish-chinese-fiction/scripts/scan_candidates.mjs \
  manuscript.md candidates.json

Candidate output is evidence for review—not permission for global replacement.

3. Apply a traceable revision ledger

node skill/polish-chinese-fiction/scripts/apply_revision.mjs \
  --source baseline.md \
  --changes changes.json \
  --output revised.md \
  --report-md revision-report.md \
  --report-csv revision-report.csv

Every old string must occur exactly once in the locked baseline. Overlapping entries are rejected instead of being silently applied in sequence.

4. Audit the final manuscript

node skill/polish-chinese-fiction/scripts/audit_revision_integrity.mjs \
  --source baseline.md \
  --revised revised.md \
  --changes changes.json \
  --strict-duplicates true \
  --output integrity.json

This checks deterministic replay, heading order, per-section paragraph counts, empty sections, punctuation balance, exact duplicate paragraphs, repeated cross-paragraph sentences, and near-duplicate paragraphs.

Included tools

Tool Purpose
scan_candidates.mjs Find likely native-Chinese, reference, collocation, and translationese problems.
scan_emotion_modality.mjs Compare colloquial adverbs, modal force, emotion words, and punctuation before/after.
apply_revision.mjs Apply a uniquely matched ledger and generate Markdown/CSV reports with hashes.
audit_revision_integrity.mjs Detect structural drift, duplicate content, ledger conflicts, and replay mismatches.
version_inventory.mjs Inventory multiple manuscript versions without counting the same editorial work twice.

All runtime scripts use Node.js built-ins only.

Repository layout

.
├── skill/polish-chinese-fiction/   # The installable skill
│   ├── SKILL.md
│   ├── agents/openai.yaml
│   ├── references/
│   └── scripts/
├── examples/                       # Small, synthetic test manuscript
├── assets/                         # README artwork and workflow diagram
└── tools/                          # Package validation

Battle-tested, not benchmark-decorated

This workflow grew out of a real Chinese novel edited across six manuscript versions and hundreds of traceable changes. The public package contains the generalized editorial system—not the private manuscript, prompts, or project logs.

Run the package tests locally:

npm test

Contributing

Native-language edge cases are especially welcome: strange pronoun chains, translated syntax that remains grammatically “correct,” dialogue that loses social register, and revision ledgers that break under overlap. See CONTRIBUTING.md.

A small ask

If this skill saves even one chapter from sounding translated—or one late revision from breaking the book—leave a ⭐. It helps the next Chinese-fiction writer find it.


Created by ilegend.

About

✍️ Codex skill for Chinese fiction | 从故事骨架改到一字一句:清掉翻译腔,守住剧情和人物口吻,防住漏章、重复、错位

Topics

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages