Back to search
job-description-skill
SkillGitHub
Claude skill that reads a job description and hands back an HTML report — match score, resume tailor, interview predictor, and a should-I-apply verdict.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/yanliudesign--job-description-skill)HTML
<a href="/listings/yanliudesign--job-description-skill"><img src="/api/listings/yanliudesign--job-description-skill/badge.svg" alt="findsafeskills safety rating" /></a>Install
i
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
SKILL.md · Manifest · 160 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- ---
- name: job-description-skill
- description: "Job Description 解码器 + Offer 策略系统。把任何 JD 翻译成「这家公司到底在招什么人 / 我的匹配度 / Gap 在哪 / 为什么投/不投 / 面试会问什么 / 下一步怎么走」。只有 3 步:贴 JD → 给简历 → 自动生成并打开一份 HTML Offer Strategy Report。是整个求职链路的入口,向下衔接 Resume Skill 和 BQ Skill。关键词:job description, JD, 职位描述, 匹配度, ATS, resume tailor, 简历定制, 招聘经理, 投不投, should I apply, 面试预测, offer strategy。"
- ---
- # Job Description Skill
- 把一份 JD 从「招聘话术」翻译成「**你能用的求职情报**」,**一次性产出一份单文件 HTML Offer Strategy Report**,自动在浏览器打开。
- 下游可一键接 **Resume Skill**(拿着 Gap / Tailor 结果继续打磨简历)和 **BQ Skill**(拿着预测题目去练故事),形成完整求职链路。
- ---
- ## 整个 skill 只有 3 步
- ### Step 1 · 贴 JD
- 开场只说一句话:
- > "把你想分析的 **Job Description 链接或全文** 贴给我。链接抓不到内容就直接粘贴 JD 文本。"
- 收到后:
- - 链接 → 尝试抓取;抓不到 → 让用户粘全文,**别根据公司名瞎猜 JD**。
- - 同时确认 **公司名 / 岗位 / Level**(JD 没明写就追问一次)。
- - **先查 [jd-bank/_index.md](jd-bank/_index.md)** — 同公司同岗位半年内分析过,告诉用户"上次解码过,要不要直接复用"。
- ### Step 2 · 给简历或 LinkedIn
- 三选一,明牌列出来:
- > "为了算匹配度 / 出 Gap / 预测面试题,需要了解你。三选一:
- > - 🅰️ **上传 / 粘贴简历全文**(PDF / Word / 文本都行)— 强烈推荐,结果最准
- > - 🅱️ **粘贴你的 LinkedIn 全文**(不是只给 URL,要全文)
- > - 🅲️ **简略介绍一下自己**(还没有简历的情况)"
- 如果用户选 🅲️,**一次问一个问题**,别一次列 6 个:当前 Title + 年限 → 行业 → 近 2-3 段经历(公司 + 负责什么 + 1 个最亮成果)→ 最强 1-2 个技能 → 想往哪个方向走。
- ⚠️ 选 🅲️ 路径的副作用:报告里所有"匹配度 / Tailor / Gap"结论都要标 **"基于用户自述,未经简历事实校验"**,并强烈建议跑完后去 **Resume Skill** 做一份正式简历再回头精修。
- ### Step 3 · 生成 HTML 并自动打开
- **收齐 JD + 简历后,不要再分别请示用户跑 Decode / Match / Predict / Should I Apply** — 一口气在后台跑完 5 条流程([prompts/](prompts/) 下的 5 份文件),把结果直接组装成一份 HTML Offer Strategy Report:
- 1. 在内存里跑 Decode → Match → Predict → Should I Apply(Tailor(简历定向调整) 只取关键 diff,简历整体打磨交给 Resume Skill)。
- 2. 按 [frameworks/offer-strategy-report.md](frameworks/offer-strategy-report.md) 的规格 + [examples/offer-strategy-template.html](examples/offer-strategy-template.html) 的骨架组装一份 10 节报告(TL;DR · 关键指标 · 1-10)。
- - ⛔ **品牌 footer 是强制项,不是装饰。** 报告结尾必须**原样**包含 `<footer>` 里的 `JD SKILL.` brand mark + `Created by Dreameryanyan` 副标题 + LinkedIn / X / 小红书三个社交按钮(含对应 CSS:`.brand-block` / `.brand-mark` / `.socials` / `.foot-meta`)。这是作者署名,**无论报告多长、数据多少都不许删或简化**。生成时直接从 [frameworks/offer-strategy-report.md](frameworks/offer-strategy-report.md) 末尾「📌 强制 Footer 区块」整段抄过去。
- 3. 写到 `~/Desktop/Claude skills/offer-strategy-<company>-<role>-<YYYYMM>.html`。
- - 写完后**跑两道自检**,任一道失败 = 报告不合格,必须重写这几块再继续(不要跳过校验直接宣称“已生成”):
- - ① **Footer 自检** — 文件里必须能搜到 `Dreameryanyan`、`brand-mark`、`yanliudreamer`、`xiaohongshu` 四个关键词,缺任一 = footer 被丢了。
- - ② **双语自检** — `<html lang="en">` 必须默认为 `en`(不允许 `zh` / `zh-CN`),且 `grep -c 'data-lang="en"'` 与 `grep -c 'data-lang="zh"'` **数量相等且 ≥ 80**。少于 80 = 模型又把内文写成了单语,已知失败模式,参见铁律四。双语 span 对的写法与适用范围见 [frameworks/offer-strategy-report.md](frameworks/offer-strategy-report.md#双语输出--bilingual-output必须)。
- 4. **自动打开**:跑 `open "<完整路径>"`(macOS)/ `xdg-open` (Linux) / `start` (Windows),让 HTML 直接在浏览器弹出来。
- 5. **同步 [jd-bank/](jd-bank/)**:按 [jd-bank/_jd-template.md](jd-bank/_jd-template.md) 写一份 `<slug>.md`,更新 [jd-bank/_index.md](jd-bank/_index.md),文件末尾加一行 `> 📊 HTML 报告:~/Desktop/Claude skills/offer-strategy-<slug>.html`。
- 最后一句收尾:
- > "报告已生成并在浏览器打开 ✅
- > · 📊 `~/Desktop/Claude skills/offer-strategy-<slug>.html`
- > · 内置 **Export PDF** 和 **Export Markdown** 两个按钮
- > · 这份 JD 已存进 [jd-bank/](jd-bank/),下次提到自动复用
- >
- > 下一步:
- > - 决定投 → 交给 **Resume Skill** 做整体简历打磨
- > - 准备 Behavior 题 → 转 **BQ Skill** 挖故事
- > - 看下一份 JD → 直接贴过来"
- ---
- ## 四条铁律(内部 5 条流程跑的时候都要守)
- **铁律一 · 先解码,再做任何事。**
- 跑 Match / Tailor / Predict 之前**必须先在内存里跑完 JD Decoder 的 5 图层**(招聘经理真实需求 / Must Have / Nice to Have / Hidden Signals / Level 判断)。直接拿原始 JD 去算匹配度,是在跟「招聘话术」对齐,不是跟「招聘经理真实需求」对齐 — 这是 90% 求职工具最大的盲区。
- **铁律二 · 不替用户编简历内容、不编公司情报。**
- - Resume Tailor 只能**重新组织 / 突出 / 翻译**用户简历里已有的事实,不能凭空捣造成就、数字、职责。
- - 公司文化、团队信号只能从 JD 文本 + 用户提供的信息推断;不要装作知道某公司内部情况。所有推断都标“我从 ___ 推断”。
- - 报告 §1 公司背景 / §2 本职位产品分析 用到的公司画像、产品节点等若做了 web 调研,标明来源;查不到就标 `[需用户补充]`,**别编**。
- **铁律三 · 数据没齐就不要硬出报告。**
- JD 缺关键内容 / 用户没给简历或自述 → 先把缺的要齐,**别硬跑**。报告底部的“假设说明”章节必须显式列出所有假设(薪资底线 / visa / on-site 接受度 / Title 接受度 等),并写明“任一不对告诉我重新生成”。
- **铁律四 · 报告必须真双语,不是代码已插了切换按钮就算数。**
- 模板 `.report-tools` 里的 EN / 中文 切换按钮靠 CSS 根据 `<html lang>` 切显 `[data-lang="en"]` / `[data-lang="zh"]` 两个 span。若你把正文写成单语(全中文或混排),按钮看起来能点但切什么都不变 — **这比报告内容自身错更严重**,因为它伪造了交付完整度。
- - 默认 `<html lang="en">`。EN 在前、ZH 在后:`<span data-lang="en">English</span><span data-lang="zh">中文</span>`。
- - **每一句用户可见拷贝**都要双胞胎:Verdict / Assumptions / Must Have 卡 / Matrix 三列 / Gap / 不投理由 + HM probe / Top 10 题 / 6 周时间线 / DM 模板…具体范围及例子见 [frameworks/offer-strategy-report.md](frameworks/offer-strategy-report.md#双语输出--bilingual-output必须)。
- - 注释 / 占位符 / 纯数字 / 专有名词不用双胞胎。
- - 自检 ② 会卡住单语报告(`data-lang="en"` 与 `data-lang="zh"` 数量不相等或总数 < 80)。
- ---
- ## 内部流程(用户看不到,由 Step 3 调用)
- | 流程 | 干什么 | Prompt 文件 |
- |---|---|---|
- | 1 · Decode | 5 图层拆解 JD 真实意图 | [prompts/jd-decoder.md](prompts/jd-decoder.md) |
- | 2 · Match | 匹配度(0.6/0.2/0.2 加权)+ Must Have 逐条命中 + Gap 三档 | [prompts/match-score.md](prompts/match-score.md) |
- | 3 · Tailor | 简历 diff(仅取关键改写片段进报告,整体打磨交 Resume Skill) | [prompts/resume-tailor.md](prompts/resume-tailor.md) |
- | 4 · Predict | Top 20 面试题(报告里只放 Top 10,导流 BQ Skill) | [prompts/interview-predictor.md](prompts/interview-predictor.md) |
- | 5 · Should I Apply | 五件套决策(⭐ / 值得投 / 不值得投 / 拿面概率 / 下一步) | [prompts/should-i-apply.md](prompts/should-i-apply.md) |
- **这几个步骤在 Step 3 里被调用一次,结果全部塞进同一份 HTML 报告。** 用户看不到"现在在跑匹配步"这种内部状态。
- **如果用户明确只想要某一条流程**("只解码不要别的" / "只算匹配度")→ 跳过 Step 3 的报告生成,直接吐这一条流程的纯文本结果。这是**例外路径**,不是默认。
- ---
- ## JD Bank
- - 位置:本 skill 目录下 `jd-bank/`,每份分析过的 JD 一个 `.md`。
- - 文件名:`<公司缩写>-<岗位slug>-<YYYYMM>.md`,如 `openai-product-designer-202606.md`。
- - frontmatter 必填:company / role / level / source_url / decoded_at / status(decoded / matched / tailored / applied / interview / closed)。
- - [jd-bank/_index.md](jd-bank/_index.md) 是反查表:按公司、按岗位类型、按状态分组。**每次 Step 3 生成报告都要同步 `_index.md`**。
- - 用户回头说"上次那个 ___"时,先查 index,别让用户重新粘 JD。
- ---
- ## 与 Resume Skill / BQ Skill 的衔接
- 这个 skill 不替代另外两个:
- - **Resume Skill** 负责把一份简历做到「整体优秀 + 个人故事一致」。JD Skill 报告里的 Tailor diff 只做**针对单个 JD 的定向调整** — 告诉用户"要做整体优化去 Resume Skill"。
- - **BQ Skill** 负责挖掘 / 结构化 / 维护故事库。JD Skill 报告 §9 只给"会被问什么" — 告诉用户"用 BQ Skill 把这些题挖成故事"。
- 三者形成的用户路径:
- ```
- 看到 Dream Job
- ↓
- JD Skill:贴 JD → 给简历 → HTML 报告自动弹出
- ↓(决定投)
- Resume Skill:整体简历打磨
- ↓(拿到面试)
- BQ Skill:挖故事 / 模拟面试
- ↓
- 拿 Offer
- ```
- ---
- ## 参考文件(按需读取,别一次全加载)
- **Prompts(5 条内部流程的执行脚本)**
- - [prompts/jd-decoder.md](prompts/jd-decoder.md)
- - [prompts/match-score.md](prompts/match-score.md)
- - [prompts/resume-tailor.md](prompts/resume-tailor.md)
- - [prompts/interview-predictor.md](prompts/interview-predictor.md)
- - [prompts/should-i-apply.md](prompts/should-i-apply.md)
- **Frameworks(知识词典 + 报告规格)**
- - [frameworks/decode-patterns.md](frameworks/decode-patterns.md) — JD 话术翻译表 + Hidden Signal 词典
- - [frameworks/match-rubric.md](frameworks/match-rubric.md) — 匹配度评分规则 + Gap 三档分类
- - [frameworks/resume-tailoring.md](frameworks/resume-tailoring.md) — ATS / Recruiter / HM 三版改写法
- - [frameworks/go-no-go.md](frameworks/go-no-go.md) — Should I Apply 打分法 + 拿面概率估算
- - [frameworks/offer-strategy-report.md](frameworks/offer-strategy-report.md) — **最终 HTML 报告的 10 节骨架 + 视觉规范 + 生成说明(Step 3 的核心规范)**
- **Examples**
- - [examples/offer-strategy-template.html](examples/offer-strategy-template.html) — HTML 报告骨架(不含个人数据)
- **JD Bank(情报库)**
- - [jd-bank/_jd-template.md](jd-bank/_jd-template.md)
- - [jd-bank/_index.md](jd-bank/_index.md)
README.md · README · 151 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- <div align="center">
- [中文](./README.zh.md) · **English**
- # 🔍 Job Description Skill
- ---
- **Decode any JD → get a one-page Offer Strategy report.** Should I apply · how well I match · top interview questions.
- [](./LICENSE)
- []()
- []()
- [](https://github.com/yanliudesign/job-description-skill/stargazers)
- [](https://claude.ai/code)
- []()
- []()
- []()
- []()
- </div>
- > 📦 Part of the **[offer-toolkit-skill](https://github.com/yanliudesign/offer-toolkit-skill)** — the full job-hunt bundle (JD · Resume · BQ). Install the bundle to get all three in one shot.
- A Claude skill for reading job descriptions. Give it a JD and your resume, and it hands back an HTML report telling you: whether this job is worth applying to, how well you match it, where the gaps are, what interviews will probably ask, whether the salary is reasonable, and what to do over the next six weeks. A JD is rarely just a recruiting blurb — it's usually the hiring manager's real expectations wrapped in HR boilerplate, and this skill unwraps that.
- ---
- ## How it works — just 3 steps
- Invoke the skill with anything like "help me with this job" / "走一下这个 JD" and you go through exactly:
- 1. **Paste the JD** — link or full text.
- 2. **Give it your resume** — three options:
- - 🅰️ Upload / paste your full resume (recommended)
- - 🅱️ Paste your full LinkedIn
- - 🅲️ No resume yet → short self-introduction. Results will be flagged as "unverified against a real resume" and you'll be nudged toward **[Resume Skill](https://github.com/yanliudesign/resume-builder-skill)** to build one.
- 3. **HTML report auto-opens** — a single-file Offer Strategy Report is generated to `~/Desktop/Claude skills/offer-strategy-<slug>.html` and pops open in your browser. Inside: TL;DR verdict, two gauges, the 10 sections below, and in-page Export PDF / Export Markdown buttons.
- That's the whole thing. No multi-step wizard, no separate "只解码" / "只算匹配度" outputs to wade through — the 5 internal flows (Decode / Match / Tailor / Predict / Should I Apply) all run silently and assemble into the same HTML report.
- If you explicitly want only one flow ("only decode this" / "just match score") the skill skips the report and gives you the raw output for that one flow.
- ---
- ## Offer Strategy Report · the 10-section framework
- Every wizard run ends with a single-file HTML report at `~/Desktop/Claude skills/offer-strategy-<slug>.html`. Fixed 10-section spine, plus a TL;DR preamble and a key-metrics gauge block:
- 
- *Hero of an actual run — verdict + assumptions + TL;DR + two gauges, plus the in-page Export PDF / Export Markdown buttons in the bottom right.*
- <sub>More screenshots: [English gallery](docs/en/) · [中文相册](docs/zh/)</sub>
- ### ▶ Video walkthrough (~2 min)
- <a href="docs/en/demo.mp4" title="Click to play the full 2-minute walkthrough">
- <img src="docs/en/preview.gif" alt="▶ Play 2-min walkthrough — Anthropic × Product Designer, Claude Code" width="100%">
- </a>
- <sub>⬆ First 8 seconds auto-play as a preview. <a href="docs/en/demo.mp4"><strong>Click to watch the full 2-minute walkthrough</strong></a> — live run against a real Anthropic JD (paste JD → attach resume → HTML report opens in your browser, ~15 MB, opens in a new tab). 中文版本 · <a href="README.zh.md#▶-视频演示-2-分钟">Chinese walkthrough</a>.</sub>
- | # | Section | What it answers |
- |---|---------|-----------------|
- | — | **TL;DR · one-minute verdict** | Should I apply? ⭐ 1–5. |
- | — | **Key metrics** | Match Score gauge + Interview Probability gauge. |
- | **1** | **Company background** | Who runs this place, what are the cultural signals, what shipped in the last 12–24 months. |
- | **2** | **Product analysis for this role** | What product / surface this role owns + the 4 core design challenges + the *real* scope behind the JD. |
- | **3** | **JD deep decode** | Why this JD reads the way it does (JD 解码精华 + Hidden Signal dictionary hits). |
- | **4** | **JD ↔ resume line-by-line match** | Must Have cards + full Match Matrix (3 columns: JD ask / resume evidence / score+gap) + Nice + 10-dim Hidden Signal radar. |
- | **5** | **Gap — what to upgrade** | 2–4 gap cards with severity (high/middle) + 4-week mitigation. |
- | **6** | **Why apply / why not** | 3 reasons each side. Every "why not" comes with an HM probe response. |
- | **7** | **Salary reality check** | JD band + market data + negotiation talking points. |
- | **8** | **Next step** | Career arc fit + referral playbook (who, DM template) + portfolio audit (4-week action list). |
- | **9** | **Top 10 interview questions** | 面试预测预览. Full Top 20 + behavior stories → hand off to BQ Skill. |
- | **10** | **6-week action plan** | Week-by-week to-dos: before applying / after applying / before interview. |
- The report supports **Export to PDF** (CJK-safe Noto font embedding) and **Export to Markdown** (single-file MD download). Spec: [`frameworks/offer-strategy-report.md`](frameworks/offer-strategy-report.md) · skeleton: [`examples/offer-strategy-template.html`](examples/offer-strategy-template.html).
- ---
- ## Three rules
- 1. **Decode before you act** — Decoding the JD first is mandatory before doing match or tailor. Matching against the wrong target means the match is wrong.
- 2. **Never fabricate** — Never invent experience, skills, or numbers the candidate didn't claim. If real evidence is missing, mark it ❌.
- 3. **One thing at a time** — No mixing. If the decode step left ambiguity, clarify before moving on.
- ---
- ## File Structure
- ```
- job-description-skill/
- ├── SKILL.md # Entry · routing · three rules
- ├── prompts/ # Execution prompts for each flow
- │ ├── jd-decoder.md
- │ ├── match-score.md
- │ ├── resume-tailor.md
- │ ├── interview-predictor.md
- │ └── should-i-apply.md
- ├── frameworks/ # Reusable frameworks / rubrics
- │ ├── decode-patterns.md # JD keyword → real intent dictionary
- │ ├── match-rubric.md # Scoring rubric (Must / Nice / Hidden)
- │ ├── resume-tailoring.md # Three-version resume strategy
- │ ├── go-no-go.md # ⭐ rating + interview probability formula
- │ └── offer-strategy-report.md # Final HTML report — sections, visual system, content rules
- ├── examples/ # Reference skeletons (no personal data)
- │ └── offer-strategy-template.html
- └── jd-bank/ # Local cache of analyzed JDs (gitignored)
- ├── _index.md # Index
- └── _jd-template.md # Starter template for new JDs
- ```
- ---
- ## How it thinks
- - **A JD reflects three layers** — company signals, the hiring manager's real expectations, and the product / role in question. The skill's job is to see through the HR packaging to those three.
- - **Decide before you optimize** — most candidates keep polishing resumes for the wrong roles. Decide whether to apply at all first.
- - **Every score has a formula** — not magic. Every Match Score and Interview Probability comes from an explicit formula in `frameworks/`. Disagree with the premise? You can argue back.
- ---
- ## Related skills
- Pairs with these two — decode the JD here, then hand off:
- - [offer-toolkit-skill](https://github.com/yanliudesign/offer-toolkit-skill) — the all-in-one bundle (JD · Resume · BQ)
- - [resume-builder-skill](https://github.com/yanliudesign/resume-builder-skill) — Resume Builder & Beautifier (11 print-ready templates)
- - [Behavior-question-skill](https://github.com/yanliudesign/Behavior-question-skill) — Behavioral interview / story bank
- ```
- See a dream job → JD Skill (decode · match · should-I-apply)
- ↓ decide to apply
- Resume Skill (tailor + polish)
- ↓ get the interview
- BQ Skill (mine stories · mock interview)
- ```
- ---
- ## License
- MIT — fork it, remix it, ship your own version.
- Created by [Dreameryanyan](https://www.linkedin.com/in/yanliudesign/) ·
- [LinkedIn](https://www.linkedin.com/in/yanliudesign/) ·
- [X](https://x.com/yanliudreamer) ·
- [Xiaohongshu](https://www.xiaohongshu.com/notification)
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.