agent-skills-cn
中英双语 Agent Skills —— 把实战打磨过的领域工作流做成开放 SKILL.md,Claude Code / DSH / Hermes / Codex 通用。
Bilingual (EN/CN) agent skills: battle-tested domain workflows packaged as open SKILL.md files, portable across Claude Code, DSH, Hermes, and Codex.
Skills / 技能列表(19 个)
每个 skill 独立成库:单独迭代、单独发版、名字即用途。Each skill lives in its own repo — one skill, one name, one release cycle.
| Skill |
What it does / 干什么 |
| email-deliverability-audit |
四道顺序 DNS 闸门(格式/死域/MX/角色占位符),逐地址判定 sendable / risky / dead,绝不编造通过。Four DNS gates classify email addresses before your campaign finds out the hard way. |
| competitor-recon |
抱怨驱动的竞品调研:评论/论坛/issue 跟踪器取证 → 聚类未满足需求 → 对比表 + 值得抄/不值得做清单。Complaint-driven competitor research with cited evidence. |
| resume-localize-cn2en |
中文简历 → 英文技术简历:删人口统计学字段、职责改量化成果、统一拼写与日期、ATS 可解析、附完整变更清单。CN→EN resume localization for tech roles. |
| delivery-checklist |
交付前核对:实开文件验行数/去重/占位符,增量文件对基线 diff,台账记录批次,候选数≠可发送数。Verify deliverables by opening the real file. |
| project-handoff |
项目跨会话交接:handoff.md 记录现状/带理由的决策/症状-原因-对策的坑,新会话免考古接手。Make the next session as smart as this one. |
| verify-claims |
主张核验:文档里每条可核验主张分类分级、追溯到最强来源、当次实测(开链接/重算总数),核不了的如实标注——文档声明自己的证据等级。Every claim carries evidence or a label. |
| decision-records |
轻量决策记录:决策/备选/理由/可逆性五字段,追加式、替代不重写——理由比结论活得更久。The why outlives the what. |
| plain-business-english |
商务英文语域(面向非母语者):先说重点、按事实本来的尺寸陈述、清除热情/道歉/确定性通胀。Facts at their size. |
| deliverable-versioning |
交付版本纪律:已发出即不可变、日期命名替代 final、一个权威交付目录+台账。Sent means immutable. |
| backlog-triage |
待办分诊:六种分类各带理由、放弃项归档、承诺变更并入单次审批。A list that only grows is not a plan. |
| web-cliplibrary |
网页剪藏库:URL+日期+逐字引文+笔记分离,源页死了照样可引用。Research that survives the session. |
| onboarding-pack |
项目上手包:是什么/怎么跑/权威地图/潜规则/找谁,十分钟读完。The month of asking around, prepaid. |
| expense-capture |
报销捕获:照实转录、不清晰就标注、税额只读不推、批次双向对账。Transcribe, flag, reconcile. |
| weekly-review |
周回顾:承诺对结果逐条对账、成果以证据入账、顺延必须带新理由。Close loops, not vibes. |
| ask-batch |
批量提问:问题停放分组、每题带默认值、一次问完、超时默认显式生效。One sitting instead of all-day interruptions. |
| estimate-before-build |
先估后建:S/M/L/XL 档位+不确定性点名、范围vs预算交用户选、终点对账校准。Make the scope decision explicit. |
| changelog-capture |
变更日志捕获:改动当下写 who/what/upgrade 条目,发版时与实际 diff 双向核对。One minute while the diff is in hand. |
| template-instantiator |
模板实例化:盘点→替换→机械扫描→出示零残留证明。Proven clean, not assumed clean. |
| context-budget |
上下文预算:分层读文件、笔记落盘抗压缩、检查点复述——长任务靠笔记收尾而不是硬撑。The window is working memory — spend it deliberately. |
Why bilingual / 为什么中英双语
Skill 正文用英文(agent 消费、最大化受众),description 内嵌中文触发词(中文任务描述也能自动命中),每个 README 双语、附真实实测案例。英文与中文生态对同一工作流的写法差异,本身就是这个系列的一手经验。
Bodies are English (agent-facing, maximum reach); descriptions carry Chinese trigger words so Chinese task phrasings still auto-load the skill. Every README is bilingual and includes a worked example from live testing.
Install / 安装
# Claude Code — the clone IS the install
git clone https://github.com/ChenneyZhuang/<skill-name> ~/.claude/skills/<skill-name>
# Hermes
cp -r <skill-name> ~/.hermes/profiles/<profile>/skills/
每个 skill 独立可用,互不依赖。Each skill is standalone — no cross-dependencies.
Status / 状态
v0.2.0 — 前六个 skill 在真实场景实测通过并按实测修订;后四个(verify-claims / decision-records / plain-business-english / deliverable-versioning)已实测发布。License: MIT.
v0.2.0 — the first six skills are live-tested in real scenarios and revised from findings; the latest four shipped after their own live tests. License: MIT.