zh-check
Skill3 starsChanged 6 days agoMITView on GitHub
中文成稿检查:脚本扫描禁用词库与模式规则,再按规则库逐条复查。用于:文档、文案、脚本、README 等成稿质检。写作转 zh-write,对话措辞转 zh-talk。
In this repository28 other files it gives its agents.
- AGENTS.mdAGENTS.md
- client-deliverableSkill
- dev-planSkill
- diagnosing-bugsSkill
- frontend-designSkill
- grillSkill
- handoffSkill
- ops-triageSkill
- ponytail-debtSkill
- ponytail-reviewSkill
- ponytailSkill
- prototypeSkill
- researchSkill
- retroSkill
- reviewSkill
- shipSkill
- showcase-pageSkill
- tddSkill
- ui-shotsSkill
- algorithmic-artSkill
- gsap-awwwards-websiteSkill
- gsap-coreSkill
- gsap-scrolltriggerSkill
- gsap-timelineSkill
- web-animSkill
- writing-for-agentsSkill
- zh-talkSkill
- zh-writeSkill
The file
Files
中文文案检查
上下游
- 前接:任何中文成稿;
writing-for-agents写的中文 skill 必过。 - 后接:按问题清单回到
zh-write改写。
两遍制:机械抓得住的交给脚本,抓不住的按规则逐条复查。两遍都干净才算通过。
第一遍:机械扫描
python3 scripts/zh_check.py <file...>脚本输出按规则分组的问题清单:禁用词库命中、模式规则命中、超长句、格式统计。
词库在 data/banned.json,按误伤风险分两级:
delete:报告腔式套话——技术、规范文档里不存在合法用法,命中即删context:人类特定场景会用的词(黑话之于职场、销售腔之于市场文案、自白之于 changelog、高频修饰之于技术写作),命中结合语境判断;每组的description写明合法情形
词条可带 allow 豁免复合词(「落地」豁免「落地页」「落地窗」):词的每次出现都落在豁免词内部时不上报。行内代码不参与词与句式检查。
发现新词就增补到对应分组,分组本身也可加;某词误伤集中于固定复合词时,给词加 allow 而不是删词。
脚本是召回器不是裁判:命中即"可能存在问题的点",由复查时判断是否真问题。 两个已知误报源——文档在谈论禁用词本身(词库列举、规则说明)属于 mention 而非 use;命中多的规范类文档复核时直接说明即可,不算违规。
第二遍:模型逐条复查
脚本查不了的规则,逐条对照原文:
- 包装词:脚本词库之外,是否有"能被普通词替换且不损失信息"的词;自造术语是否满足准入三条(无通用词、定义一次多处复用、缺它读不懂系统)
- 截短词:是否有被压成单字或半截、且并非比全称更常用的词
- 拟人:是否有让系统、规则、工具"知道、觉得、想"的表述
- 语义重复:是否两句话讲同一件事、同段重复同一词、对偶两边同义
- 口语残留:白话疑问(能不能/该不该)、白话翻译术语
- 辩解与自白:是否在解释"为什么没做",或暴露未完成状态
- 结构腔:对偶句、加粗论点列表、戏剧化分节、金句收尾、公允体、隐喻堆叠、文末教训清单是否密集复现(定义见 zh-write「结构腔」);机械扫描的加粗与破折号计数可作初筛
输出
按规则分组报告:规则名 → 行号 → 原文片段 → 建议改法。没有问题的规则不列。
边界
- 代码块、报错原文、API 名不算问题
- 对话脚本、语音播报、面向儿童或长辈的文本不适用本检查
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

