Skills
skills/zenine/resume-intelligence-hub/resume-intelligence-hub

resume-intelligence-hub

Installation
SKILL.md

Resume Intelligence Hub

An opinionated methodology + scaffolding for a personal, AI-driven career management repository. Industry-agnostic, seniority-agnostic, and available in both Chinese and English. The research/grant track is optional — most users won't need it.

When to use this skill

Use when:

  • User wants to create a personal resume/CV management system from scratch
  • User has scattered resumes/CVs and wants to consolidate into a single source of truth
  • User needs JD-tailored resume generation
  • User wants interview coaching backed by their actual experience library
  • User is applying for academic grants / research funding and needs a research-profile track (optional)
  • User asks to adopt this methodology for someone else (friend, family member, colleague)

Don't use when:

  • User just wants one resume written once (no hub needed — write it directly)
  • User wants generic resume tips without infrastructure

Core philosophy (the non-obvious design decisions)

These are what make the hub work. Preserve them when bootstrapping.

  1. Single source of truth — profiles/ holds factual data; generated resumes are derivatives. Historical resumes are archive-only, never edited.
  2. Positioning lock in AGENTS.md — the target role level / industry is declared once at the top of AGENTS.md; every generated resume biases toward it. Change target = edit one file, not reprompt each time.
  3. Dual-track separation (if enabled) — recruitment and research/grant applications have entirely different formats, languages, and emphasis. Their data sources are split: profiles/master.md vs profiles/research.md. Do not merge.
  4. Verification system, path-references only — sensitive originals (IDs, contracts, certificates, appointment letters, diplomas) are NOT copied into Git. verification/references.md holds a path-reference index to local/cloud/public-URL locations. The repo stays shareable; originals stay out of version control.
  5. Public-source cross-check before high-stakes submissions — before senior-role interviews or grant applications, run a verification/{date}-web-check.md pass to cross-check every load-bearing claim.
  6. todo.md vs changelog.md — todo is only pending work; completed items migrate to changelog, dated, reverse-chronological. No stale [x] marks accumulating.
  7. Per-application folders — each attempt lives at jobs/applications/{company}-{role}-{YYYY-MM-DD}/. Same company, second attempt = second folder.
  8. STAR stories are a separate asset — profiles/stories.md holds interview-ready narratives, distinct from factual master.md.
  9. Monolingual output — the hub is either all Chinese or all English. Pick one at bootstrap; don't mix.

Referenced frameworks (cite these by name when coaching)

The hub's workflows are built on established frameworks. Use the framework name explicitly when guiding the user — it gives them vocabulary to research further and signals craft.

Bullet writing — Google's XYZ formula (from Laszlo Bock, former Google SVP People Ops):

"Accomplished [X] as measured by [Y], by doing [Z]." Each bullet answers: what did you do, how was the impact measured, and what actions produced it.

Quantification — the one-number rule: every bullet should contain at least one number (headcount, percentage, dollar, timeframe, scale). Bullets without numbers get cut in rewrite.

Experience narration — STAR (Situation, Task, Action, Result): the dominant format for behavioral interviews and case-study writing. Used in profiles/stories.md.

  • CAR / PAR (Challenge-Action-Result / Problem-Action-Result) — simpler alternatives when STAR feels over-structured for short bullets.

Behavioral interviews — BEI (Behavioral Event Interviewing, David McClelland): the methodology underlying "tell me about a time when..." questions. Past behavior predicts future behavior — prepare specific events, not generalities.

Positioning — T-shaped skills (McKinsey/IDEO): deep expertise in one vertical + broad fluency across adjacent areas. For senior roles, sometimes π-shaped or M-shaped (multiple depths). Use when helping a user explain their unique value.

Grant proposals — Heilmeier Catechism (George Heilmeier, former DARPA director): nine questions every research proposal should answer. Use when drafting the "research basis" and project-narrative sections:

  1. What are you trying to do? (Articulate objectives with no jargon.)
  2. How is it done today, and what are the limits?
  3. What's new in your approach, and why will it succeed?
  4. Who cares? If you succeed, what difference does it make?
  5. What are the risks and payoffs?
  6. How much will it cost?
  7. How long will it take?
  8. What are the midterm and final "exams" to check for success?
  9. What's your exit strategy?

Career planning — career capital (Cal Newport, So Good They Can't Ignore You): rare-and-valuable skills build leverage. When doing gap analysis, ask: "What capital do you have? What's worth investing in?"

Goal framing — SMART (Specific, Measurable, Achievable, Relevant, Time-bound): used in todo.md to make priorities concrete.

Due diligence — triangulation: every load-bearing claim needs ≥2 independent public sources before submission. Foundation of workflows/verification.md.

Stretch target heuristic — "1-2 levels up, 1.5-3x comp, 70% match":

  • Target roles 1-2 levels above current title. Same-level switches rarely earn the switching cost; 3+ levels is usually delusional.
  • External moves command a premium: floor 1.2-1.5x current total comp; stretch ceiling 2-3x when simultaneously leveling up AND shifting company type (e.g., enterprise → fast-growing startup with equity).
  • Apply at ~70% match, not 100%. Senior hires are expected to grow into the role.
  • Warm intro >> cold apply (~10x conversion). Invest in network activation before mass-applying.
  • Exceptions: career pivots (pay cut normal), downshift by choice, at market ceiling, early-career (smaller multipliers).
  • Foundation of workflows/jd-sourcing.md.

AI IDE compatibility

The canonical instruction file is AGENTS.md — the cross-IDE standard (Claude Code, Cursor, Codex, Cline, Windsurf, and others read it). The hub works with any AI IDE that loads a markdown context file. The onboarding section of readme.md walks users through wiring it to their specific IDE (symlink for Claude Code legacy, copy/symlink into .cursor/rules/, .windsurfrules, .github/copilot-instructions.md, etc.).

Setup workflow: the bootstrap interview

When a user invokes this skill to create a new hub, run this interview before creating any files. Ask one at a time — don't dump all questions at once.

Question 1 — Language

"This hub will be in Chinese or English? Pick one — the templates don't mix languages."

  • cn → use templates/cn/
  • en → use templates/en/

Question 2 — Existing materials (ask this early, before positioning)

"Do you have existing resumes, CVs, portfolios, or bio documents? Point me at the files, or drop them into a folder. I'll read them first — it makes every other question easier to answer."

Most users already have 1-5 old resumes. Reading these before asking about positioning is critical because:

  • The agent can infer industry, seniority, and career arc from real content
  • The user rarely describes their own positioning well in the abstract
  • Old resumes seed profiles/master.md in Step 4, saving hours of typing

Accept any format (PDF, DOCX, Markdown, pasted text). If provided, convert to Markdown and hold them aside — they'll move into the hub's archive directory in Step 2.

If the user has nothing yet, that's fine — proceed. Just note it will take longer to fill in profiles.

Question 3 — Field / Industry

"What field or industry do you work in? (e.g., software engineering, finance, healthcare, design, academia, law, marketing, manufacturing, sales…)"

If Q2 materials were provided, propose an answer based on what you read and confirm ("Looks like you're in {{field}} — right?"). Do not assume tech. A designer's hub looks different from a lawyer's which looks different from a manufacturing engineer's.

Question 4 — Target role / seniority

"What roles are you targeting? (e.g., entry-level, mid-level IC, senior IC, first-line manager, director, VP/executive, founder, academic PI)"

This becomes the "positioning lock" in AGENTS.md. Also ask what should be emphasized vs de-emphasized for that level. Example prompts to offer:

  • For senior management: emphasize team size, budget, P&L, cross-org, external influence; de-emphasize individual execution detail
  • For senior IC: emphasize depth, craft quality, ownership; de-emphasize people-management if not relevant
  • For career pivot: emphasize transferable skills and motivation narrative

Question 5 — Research / grant track?

"Do you also apply for academic grants, research funding, or need a publication/patent track? (If you just do job applications, answer no — it's optional.)"

  • no (most users) → skip profiles/research.md, skip 科研/ or research-archive/, skip grant-application workflow, remove "dual-track" language from AGENTS.md
  • yes → include research.md, research archive, grant workflow. Ask follow-up: "Which grant types do you typically apply for?" (affects the template's header examples)

Question 6 — Output language for resumes

"What language will your resumes/CVs be in?"

  • Chinese only / English only / both (bilingual by target market)

If bilingual, the hub language from Q1 is the management language; resume output language is picked per application.

Question 7 — Repo location

"Where should the repo live? (Suggest ~/github/<username>-career or ~/Documents/career-hub. It must be a private location — never push to a public remote.)"

Step 2 — Create the directory structure

Based on answers, copy from templates/cn/ or templates/en/. Remove research-track files if Q5 = no. Apply answers to the positioning section of AGENTS.md — other template files are structured so the agent fills them through conversation, not mail-merge (see "Template conventions for filling content" below).

Directory skeleton (Chinese version):

.
├── profiles/
│   ├── master.md          # 招聘数据源
│   ├── research.md        # 仅科研轨启用时
│   ├── skills.md
│   └── stories.md
├── jobs/
│   ├── templates/
│   ├── market-watch/
│   └── applications/
├── verification/
│   ├── references.md
│   └── 证书资质/            # 或 credentials/（英文）
├── assessments/            # 性格测评、360 反馈等（可选，interview-prep 工作流引用）
├── 简历/                    # 或 resumes-archive/（英文）
├── 科研/                    # 仅科研轨启用时
├── AGENTS.md
├── readme.md
├── todo.md
└── changelog.md

English version: translate directory names (简历/ → resumes-archive/, 科研/ → research-archive/, 证书资质/ → credentials/). assessments/ stays the same.

Why no output/ or prompts/: the per-application folder (jobs/applications/{company}-{role}-{date}/) already stores generated resume/cover/prep — a top-level output/ is redundant. Add prompts/ on demand if the user starts curating reusable prompt templates.

Step 3 — Move imported materials into the archive

If Q2 provided existing resumes/CVs, copy them into 简历/ (or resumes-archive/) and, if relevant, 科研/ (or research-archive/). Keep originals + add a Markdown conversion next to each.

Step 4 — Seed AGENTS.md

This is the most important file. Walk through templates/<lang>/AGENTS.md filling in the positioning-lock section from answers to Q3 + Q4. If Q5 = no, remove the research-track section.

Step 5 — Seed starter profiles

Open profiles/master.md. Two scenarios:

  • If Q2 provided materials: propose a first draft by consolidating the archived resumes into master.md, then ask the user to fill gaps (contacts, quantified outcomes, recent roles not on old resumes).
  • If no materials: walk through section by section, starting with the most recent role.

Step 6 — Initialize git

git init
echo '.DS_Store' > .gitignore
echo '*.tmp' >> .gitignore
# do NOT gitignore resume content
git add . && git commit -m "Initial hub scaffold"

Remind user: private remote only. Never push to a public GitHub repo.

Step 7 — Present the next-steps checklist

Don't leave the user staring at a half-empty hub. Scan the seeded profiles and write a prioritized punch list into todo.md under "本周优先" / "This week's priorities". Typical items right after bootstrap:

  • Add P&L / budget / headcount numbers for the most recent 3 years
  • Fill in LinkedIn / personal site link
  • Drop certificate/credential PDFs into verification/credentials/ (or 证书资质/) and update references.md
  • Write the first 2-3 STAR stories in profiles/stories.md (examples already provided as format reference)
  • If research track enabled: populate at least one row in each of the publications/patents/projects tables

Tell the user: "You have a working hub. Next session, say '继续完善档案' / 'continue filling in profiles' and I'll pick up the todo list."

Template conventions for filling content

All templates use plain-prose instructions instead of {{placeholder}} tokens. Reason: the user will always have an AI agent fill profiles, not hand-type them — {{}} adds visual noise and parsing ambiguity without helping. The conventions:

  • Blank field after a colon (- **Name**:) — agent writes the value after the colon
  • Italicized _待补充_ / _To be filled_ — agent replaces this line with generated content
  • Blockquote > ... — guidance that stays as context even after content is filled; don't delete it
  • Parenthetical (example or format hint) — inline hint the agent replaces with the real value
  • Table with only a header row — agent appends data rows through conversation
  • Example sections (like the two demo stories in stories.md) — stay as format reference; don't remove when filling the user's real stories

Workflows after setup

Once the hub exists and AGENTS.md is loaded, the agent handles these. See workflow files next to this SKILL.md:

  • workflows/jd-sourcing.md — find good JDs (before you can tailor one) — channel selection by region/seniority/field, stretch-target calibration, warm-intro protocol
  • workflows/jd-tailored-resume.md — generate a custom resume for a specific JD
  • workflows/interview-prep.md — predict questions, prepare STAR answers, tech review
  • workflows/verification.md — pre-submission public-source cross-check
  • workflows/grant-application.md — only if research track enabled

Complementary skills from the ecosystem

This hub is a framework — a methodology + repo structure for long-term career management. For specific one-shot tasks, several community skills compose well on top of it. Suggest them to the user when the corresponding use case comes up:

Install any of the above with npx skills add <owner/repo@skill> -g -y. They coexist — the hub remains the source of truth; specialized skills consume or augment it.

Adaptation notes

Non-tech fields: the framework works identically. Adjust vocabulary — a chef's skills.md lists cuisines and techniques; a lawyer's lists practice areas and jurisdictions; a designer's lists tools, mediums, and portfolio pieces. The structure (single source of truth, positioning lock, verification, todo/changelog split) is invariant.

Early-career users: the "positioning lock" still applies — it just targets different levels. Scale the profile depth to the user's experience; don't force placeholders they can't fill.

Users applying internationally: if they need both Chinese and English resumes, keep the hub in whichever language they manage better, and pick resume output language per application.

Syncing across devices: private GitHub repo or self-hosted Gitea. Do NOT use only cloud sync (Dropbox/OneDrive/iCloud) — you lose the Git diff history that makes profile evolution trackable.

Anti-patterns to avoid

  • Don't assume tech/engineering context. Ask in Q2. A marketing manager and a mechanical engineer need very different keyword vocabularies.
  • Don't force the research track on users who don't need it. Ask Q4 first.
  • Don't mix Chinese and English in the hub. The framework supports both, but pick one per hub.
  • Don't copy sensitive originals (IDs, contracts, salary slips) into the repo. Use verification/references.md path-references.
  • Don't let todo.md accumulate [x] completed items. Move to changelog.md each session.
  • Don't generate a resume by rewriting an old resume. Always regenerate from profiles/master.md against the specific JD.
  • Don't merge recruitment and research profiles into one. The separation is load-bearing when both tracks exist.
  • Don't skip verification before high-stakes submissions. One unverifiable claim can sink a whole application.

Template index

  • templates/cn/ — Chinese template set (all files in Chinese)
  • templates/en/ — English template set (all files in English)
  • workflows/ — workflow guides (English, agent-facing; apply to both languages)
Weekly Installs
2
Repository
zenine/resume-i…ence-hub
First Seen
8 days ago
Security Audits
Gen Agent Trust HubPass
SocketWarn
SnykWarn