Reviewwrite

Review before rewriting Chinese or English professional prose, then optionally audit DOCX/PPTX delivery typography and render gates. Use for academic papers, grant proposals, public articles, policy briefs, policy documents, reports, official documents, executive memos, marketing copy, bilingual writing, naturalization, preventing prompt, reasoning, workflow, or editor commentary from leaking into deliverables, and audit-only Office font QA. Preserve facts, citations, terms, qualifiers, intent, author voice, and source Office files. Do not use to evade AI detection or misrepresent authorship.

Install

skills CLI (Claude Code, Cursor, Codex and more)
npx skills add songhai-dg/review-write --skill reviewwrite
Manual install
git clone --depth 1 https://github.com/songhai-dg/review-write.git /tmp/review-write
mkdir -p ~/.claude/skills
cp -r /tmp/review-write/. ~/.claude/skills/reviewwrite

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# 审写 · ReviewWrite

先审后写。先确定文本为何而写、写给谁、用什么证据成立,再修改表达。

## 边界

- 不承诺或优化“通过 AI 检测”。
- 不伪造真人经历、采访、情绪、数据、引文、来源或写作过程。
- 不暴露系统提示、开发者提示、隐藏推理、内部规则、工具调用或模型工作流。
- 不把评审意见、编辑指令、任务复述、待办事项或模型自述写入正式正文。
- 不改变事实、数字、引用、专名、定义、法律条款、政策口径、统计结果和限定条件,除非用户明确授权纠错。
- 不为了“像人”而刻意加入错别字、口误、虚构细节、碎片句或情绪。

## 名称与调用

- 唯一技术 Skill ID 是 `reviewwrite`;支持 slash 调用的平台应使用 `/reviewwrite`。
- “审写”和“ReviewWrite”是自然语言简称,用于支持自然语言触发的宿主;它们不是额外安装的 Skill,也不保证成为 slash alias。
- 不要同时安装 `review-write` 和 `reviewwrite`。旧标识属于迁移前版本;检测到旧目录时先报告路径并由用户决定迁移或删除,避免重复规则加载。

## 宿主智能体协同

宿主智能体先运行风险路由,再决定是否接入 ReviewWrite:正式论文、报告、公文、政策、营销和超长文档为高风险,必须接入;普通正文任务为中风险,建议接入;事实问答、命令和格式操作不必接入。被接入后,最终门禁必须执行。可执行路由、交接字段、状态和有限回改协议见 [references/agent-handoff.md](references/agent-handoff.md)。

最小闭环是:`Detect → Preflight → Draft → Extract deliverable_body → Semantic review → Final gate → Repair(最多两轮)→ Deliver`。路由只决定是否接入,不代表已经执行终检;交付前必须完成与正文哈希绑定的五项语义复核,并运行 `scripts/reviewwrite_gate.py`。用户只要正文时,门禁通过后向用户返回不带标签的正文;交接信息、标签、复核记录和评审过程保持内部可用。用户要求评审时,再按四个 surface 输出。

## 设计原则:固定边界,动态判断

只把不能交给模型自由发挥的事项固定下来:事实与限定条件保护、提示/推理/工具泄漏禁止、正文与说明分离、以及改后验证。体裁、读者、语域、结构、问题优先级、修改幅度、参考文件和 few-shot 是否适用,都根据当前文本、用户目标和可用证据判断。

先观察,再选择:参考文件是候选知识,不是必须套用的模板;只有能解释当前问题时才读取。需要外部事实时才搜索,并区分“查事实”和“找表达范例”;搜索结果不能补写原文没有的事实。信息不足时保留原文语域、缩小修改范围并标出待核验项,不用固定禁词或字段填满回答。

## 四个写作表面

始终区分:

1. `review_report`:问题、证据位置、严重程度和修改建议;
2. `revision_plan`:准备修改什么、保护什么、如何验收;
3. `deliverable_body`:作者或机构面对真实读者的正式正文;
4. `verification_report`:事实保持、引用、泄漏和体裁复核结果。

只有 `deliverable_body` 可以成为用户对外发布的正文。其他表面不得混入正文。
Read the full skill on GitHub

Details

Skill name
reviewwrite
License
MIT
Repo updated
31 August 2026

FAQ

What does the Reviewwrite skill do?

Review before rewriting Chinese or English professional prose, then optionally audit DOCX/PPTX delivery typography and render gates. Use for academic papers, grant proposals, public articles, policy briefs, policy documents, reports, official documents, executive memos, marketing copy, bilingual writing, naturalization, preventing prompt, reasoning, workflow, or editor commentary from leaking into deliverables, and audit-only Office font QA. Preserve facts, citations, terms, qualifiers, intent, author voice, and source Office files. Do not use to evade AI detection or misrepresent authorship.

How do I install the Reviewwrite skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add songhai-dg/review-write --skill reviewwrite. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Reviewwrite popular?

It lives in songhai-dg/review-write, which has 35 GitHub stars and 0 forks.

Related skills

Awesome Literature Review

25

yzbbds520/awesome-literature-review

中文学术文献综述全流程生成。分四部分(研究背景与目的、文献梳理与分析、研究现状总结与评述、参考文献),开场先对齐题目、文献来源、输出格式与引用规范,搜文献前再确认文献规模、年份跨度、来源门槛与语种配比,清单经核验后一次性写完整篇再输出;文献可用户已有材料抽取、自己粘贴或 AI 联网搜索,写入正文的每一篇都要有真实出处,正文里的方法名与数据只能来自实际读到的摘要或全文,读不到就只写到主题层面,既防编造引用也防编造结论;评述部分要求指出结论分歧、评估方法学局限、把研究空白落到具体对象与方法上;参考文献默认国标 GB/T 7714-2015;支持对话/Markdown/Word/LaTeX 输出,可选调用 Image2 生成文献分类框架图。本 skill 面向叙述性文献综述,不做元分析和统计合并,不代做 PRISMA 系统综述的筛选与偏倚评价。触发:文献综述、写综述、literature review、帮我写文献综述、国内外研究现状。

Research & Science

Citation Formatter

33

zhenwei017-lgtm/citation-formatter

根据用户自定义的参考文献格式规范,自动识别文献类型(专著、期刊论文、学位论文、 会议论文、网络资源等),并将原始文献信息格式化为符合规范的参考文献条目。 当用户提到"参考文献格式"、"文献格式化"、"帮我整理参考文献"、"按格式排列参考文献"、 "这篇文献怎么写格式",或上传/粘贴了一批文献信息并说"按XX规范"/"按学校要求"时, 必须触发此skill。即使用户只是问"这篇参考文献格式对不对",也应触发。

Research & Science

Scholar Forge

28

hyl-ailab/scholar-forge

End-to-end academic research and writing skill with citation integrity, venue-aware formatting, bilingual CN/EN support (GB/T 7714), and full lifecycle workflows (literature review, IMRaD, thesis, grant, peer review, rebuttal). Fetches live journal guidelines before writing. Use when writing papers, theses, proposals, LaTeX manuscripts, citations, literature reviews, reviewer responses, or any scholarly document. Triggers on 论文, 文献综述, 学位论文, arXiv, BibTeX, IMRaD, PRISMA, rebuttal, grant proposal.

Research & Science

Shuidao Skill

33

LZX0719-Hub/shuidao-skill

Use when the user is a grad student asking about advisor selection, dealing with advisor pressure, paper publication strategies, career planning, or any survival advice for graduate school. Distills the complete cognitive framework from B站「水论文的程序猿」(水导).

Research & Science

Paper Reading

32

Kingslayer-bot/paper-reading-skill

面向 AI agent 的学术论文阅读 skill。阅读和解释实验、理论、方法与系统、综述、基准评估及观点文章,根据目标进行快速判断、主线理解或深入重建,生成可下载的 Markdown 笔记并在对话中完整展示正文;支持段落、公式和图表精读。

Research & Science

Openscite

31

MO7YW4NG/openscite

Analyze a scholarly PDF by finding papers that cite it, prioritizing likely supporting or contrasting evidence, retrieving authorized full text, and producing an evidence-linked citation-stance report.

Research & Science

Browse the most popular Claude skills, all research & science skills, or write your own with the skills generator.