Paper Refinery — Development skill for Claude Code

Paper Refinery 🛢️✨

**From raw drafts to submission-ready manuscripts — one skill, every standard.**

论文精炼厂：把粗稿炼成可投稿的成品。命名规范、主张校准红线、六遍写作引擎、图表标准、LaTeX 工程、强制收尾扫描——全部蒸馏进一个 skill。

[](https://claude.ai) [](https://zcode.ai) [](#) [](#license)

---

Why

Most polishing skills do one thing: fix structure, or strip AI flavor, or reduce defensive writing. Used alone they fight each other — the de-AI pass deletes the very hedging the calibration pass requires; the anti-defensive pass flattens calibrated claims into overclaims.

**Paper Refinery** distills five complementary skills into one engine and — the key part — adds explicit **conflict-resolution rules** for every place they disagree. It also enforces the things no writing skill covers: naming discipline (no `C1`/`C2` codenames, no code-style dataset names), claim-vs-evidence verb calibration, figure design standards, LaTeX engineering pitfalls, and a mandatory seven-step final sweep with zero-tolerance gates.

What's inside

paper-refinery/
├── SKILL.md                 Dual-mode framework (Mode A: edit / Mode B: final sweep)
└── references/              progressive disclosure — read only what the task needs
    ├── writing.md           ★ integrated six-pass writing engine + conflict-resolution table
    ├── section-rules.md     per-section shapes, sentence discipline, title/abstract rules
    ├── claim-red-lines.md   claim-calibration red lines, verb–evidence table, audit format
    ├── naming.md            dataset/pipeline naming conventions + residue-scan commands
    ├── figures.md           design tokens, verified glyph library, render pipeline, acceptance
    ├── latex-engineering.md compile chains, 9 diagnosed pitfalls, zero-tolerance gates
    └── fi