SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsjob-description-skill

job-description-skill

General & other

View repository

Updated 10-16 09:56

About this skill

Job Description Skill

Translate a JD from "recruiting talk" into "job-seeking intelligence you can use」，produce a single-file HTML Offer Strategy Report in one go, automatically opened in the browser.

Downstream can connect with one click to Resume Skill(take the Gap / Tailor results and continue polishing the resume) and BQ Skill(take the predicted questions and practice your stories), forming a complete job-seeking chain.

The entire skill has only 3 steps

Step 1 · Paste JD

Opening says only one sentence:

"Paste the Job Description link or full text you want to analyze to me. If the link content cannot be fetched, just paste the JD text directly."

After receiving it:

  • Link → try to fetch; if it cannot be fetched → ask the user to paste the full text,do not guess the JD based on the company name。
  • At the same time confirm company name / role / Level(if the JD does not state it clearly, ask once).
  • First check jd-bank/_index.md — if the same company and same role has been analyzed within the past six months, tell the user "this was decoded last time, do you want to reuse it directly".

Step 2 · Provide resume or LinkedIn

Choose one of three, list them clearly:

"In order to calculate match level / produce Gap / predict interview questions, I need to understand you. Choose one of three:

  • 🅰️ Upload / paste full resume text(PDF / Word / text all work) — strongly recommended, most accurate results
  • 🅱️ Paste your full LinkedIn text(not just the URL, full text required)
  • 🅲️ Briefly introduce yourself(for when there is no resume yet)"

If the user chooses 🅲️,ask one question at a time, do not list 6 at once: current Title + years of experience → industry → last 2-3 experiences (company + what you were responsible for + 1 brightest achievement) → strongest 1-2 skills → which direction you want to move toward.

⚠️ Side effect of choosing the 🅲️ path: all "match level / Tailor / Gap" conclusions in the report must be marked "based on user self-report, not verified against resume facts", and it is strongly recommended that after finishing, go to Resume Skill to make a formal resume and then come back to refine.

Step 3 · Generate HTML and open automatically

After collecting JD + resume, do not separately ask the user to run Decode / Match / Predict / Should I Apply — run all 5 processes in the background in one go (prompts/ 5 files under it), and directly assemble the results into an HTML Offer Strategy Report:

  1. Run Decode → Match → Predict → Should I Apply in memory (Tailor (targeted resume adjustment) only takes the key diff; overall resume polishing is left to Resume Skill).
  2. Assemble according to frameworks/offer-strategy-report.md 's specification + examples/offer-strategy-template.html 's skeleton a 10-section report (TL;DR · key metrics · 1-10).
    • ⛔ The brand footer is mandatory, not decoration. The report ending mustverbatimcontain the <footer> 's JD SKILL. brand mark + Created by Dreameryanyan Subtitle + three social buttons for LinkedIn / X / Xiaohongshu (with corresponding CSS:.brand-block / .brand-mark / .socials / .foot-meta). This is the author attribution,no matter how long the report is or how much data it has, it must not be deleted or simplified. When generating, directly copy the entire section from the end of frameworks/offer-strategy-report.md 「📌 Mandatory Footer Block」.
  3. Write to ~/Desktop/Claude skills/offer-strategy-<company>-<role>-<YYYYMM>.html。
    • After writingrun two self-checks, if either one fails = the report is unqualified, and these blocks must be rewritten before continuing (do not skip validation and directly claim "generated"):
      • ① Footer self-check — the file must contain Dreameryanyan、brand-mark、yanliudreamer、xiaohongshu all four keywords; if any is missing = the footer was dropped.
      • ② Bilingual self-check — <html lang="en"> must default to en(not allowed zh / zh-CN), and grep -c 'data-lang="en"' and grep -c 'data-lang="zh"' must be equal in number and ≥ 80. Fewer than 80 = the model wrote the body as monolingual again, a known failure mode; see Iron Rule Four. For how to write bilingual span pairs and their scope of application, see frameworks/offer-strategy-report.md。
  4. Auto-open: run open "<完整路径>"（macOS）/ xdg-open (Linux) / start (Windows), so the HTML pops up directly in the browser.
  5. Sync jd-bank/: follow jd-bank/_jd-template.md to write a <slug>.md, update jd-bank/_index.md, add a line at the end of the file > 📊 HTML 报告：~/Desktop/Claude skills/offer-strategy-<slug>.html。

Final closing sentence:

"The report has been generated and opened in the browser ✅ · 📊 ~/Desktop/Claude skills/offer-strategy-<slug>.html · Built-in Export PDF and Export Markdown two buttons · This JD has been saved to jd-bank/, and will be automatically reused next time it is mentioned

Next step:

  • Decide to apply → hand off to Resume Skill for overall resume polishing
  • Prepare for Behavior questions → switch to BQ Skill to dig for stories
  • View the next JD → just paste it over"

Four Iron Rules (all must be followed when running the internal 5-step process)

Iron Rule One · Decode first, before doing anything else. Before running Match / Tailor / Predictyou must first run all 5 layers of the JD Decoder in memory(hiring manager's real needs / Must Have / Nice to Have / Hidden Signals / Level judgment). Directly using the raw JD to calculate match level is aligning with "recruiting talk", not with the "hiring manager's real needs" — this is the biggest blind spot of 90% of job search tools.

Iron Rule Two · Do not fabricate resume content or company intelligence for the user.

  • Resume Tailor can onlyreorganize / highlight / translatethe facts already in the user's resume; it must not fabricate achievements, numbers, or responsibilities out of thin air.
  • Company culture and team signals can only be inferred from the JD text + information provided by the user; do not pretend to know the internal situation of a company. Mark all inferences with "I infer from ___".
  • If web research was done for the company profile, product milestones, etc. used in Report §1 Company Background / §2 Product Analysis for This Role, indicate the sources; if it cannot be found, mark [需用户补充]，Do not make it up。

Iron Rule Three · If the data is not complete, do not force out a report. If the JD lacks key content / the user did not provide a resume or self-description → first get the missing items,do not force-run itThe "Assumptions" section at the bottom of the report must explicitly list all assumptions (salary floor / visa / on-site acceptance / Title acceptance, etc.), and state "if any of these is wrong, tell me and I'll regenerate."

Iron Rule Four · The report must be truly bilingual; it doesn't count just because the code has already inserted a toggle button. In the template .report-tools , the EN / 中文 toggle button relies on CSS to show/hide based on <html lang> the two spans [data-lang="en"] / [data-lang="zh"] . If you write the body as monolingual (all Chinese or mixed), the button looks clickable but nothing changes when toggled — this is worse than the report content itself being wrong, because it fakes delivery completeness.

  • Default <html lang="en">. EN first, ZH after:<span data-lang="en">English</span><span data-lang="zh">中文</span>。
  • Every sentence of user-visible copymust have a twin: Verdict / Assumptions / Must Have cards / Matrix three columns / Gap / reasons not to apply + HM probe / Top 10 questions / 6-week timeline / DM template… for the specific scope and examples see frameworks/offer-strategy-report.md。
  • Comments / placeholders / pure numbers / proper nouns don't need twins.
  • Self-check ② will block monolingual reports (data-lang="en" and data-lang="zh" counts unequal or total < 80).

Internal process (invisible to the user, invoked by Step 3)

Process What it does Prompt file
1 · Decode 5-layer breakdown of the JD's true intent prompts/jd-decoder.md
2 · Match Match score (0.6/0.2/0.2 weighted) + Must Have hit one by one + Gap three tiers prompts/match-score.md
3 · Tailor Resume diff (only key rewritten snippets go into the report; overall polishing goes to Resume Skill) prompts/resume-tailor.md
4 · Predict Top 20 interview questions (only Top 10 go in the report, funneling to BQ Skill) prompts/interview-predictor.md
5 · Should I Apply Five-piece decision (⭐ / worth applying / not worth applying / probability of getting an interview / next step) prompts/should-i-apply.md

These steps are invoked once in Step 3, and all results are put into the same HTML report. The user cannot see internal states like "now running the matching step."

If the user explicitly wants only one process("just decode, nothing else" / "just calculate the match score") → skip Step 3's report generation and directly output the plain-text result of that one process. This is anexception path, not the default.

JD Bank

  • Location: under this skill directory jd-bank/, one .md。
  • Filename:<公司缩写>-<岗位slug>-<YYYYMM>.md, e.g. openai-product-designer-202606.md。
  • frontmatter required: company / role / level / source_url / decoded_at / status (decoded / matched / tailored / applied / interview / closed).
  • jd-bank/_index.md is a reverse lookup table: grouped by company, by role type, by status.Every time Step 3 generates a report, it must sync _index.md。
  • When the user comes back and says "that ___ from last time," check the index first; don't make the user re-paste the JD.

Connection with Resume Skill / BQ Skill

This skill does not replace the other two:

  • Resume Skill is responsible for making a resume "excellent overall + consistent personal story." The Tailor diff in the JD Skill report only doestargeted adjustments for a single JD — tell the user "for overall optimization, go to Resume Skill."
  • BQ Skill is responsible for mining / structuring / maintaining the story library. JD Skill report §9 only gives "what will be asked" — tell the user "use BQ Skill to mine these questions into stories."

The user path formed by the three:

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

Reference files (read on demand, don't load them all at once)

Prompts (execution scripts for 5 internal processes)

Frameworks (knowledge dictionary + report specifications)

Examples

JD Bank (intelligence repository)