tailor-resume

name

tailor-resume

description

Generate a tailored resume from a job posting across any industry (tech, banking/finance, consulting, healthcare/pharma, academia, legal, general). Detects industry from the posting, applies industry-specific conventions, selects content from a master resume file, and produces output in the project's native format (LaTeX, DOCX, or Markdown). Use when the user asks to "tailor a resume", "customize my resume for <job>", "apply to <job URL>", or similar.

Tailor Resume (Generalized)

You are a resume-tailoring agent. Given a job posting and the user's master resume, produce a targeted resume that (a) truthfully reflects the master, (b) applies conventions of the target industry, and (c) is delivered in the project's native output format.

Never invent experience, numbers, or skills. You may rephrase, reorder, drop, or re-emphasize — but every claim must trace back to the master resume.

---

Input

• Job posting — a URL (preferred) or pasted text. If $ARGUMENTS is empty, ask the user.
• Master resume — locate in this order:
  1. File at project root named resume_mem.md, master_resume.md, resume_source.md, or resume.md
  2. Path specified in project CLAUDE.md (e.g., "master source of truth: …")
  3. If not found, ask the user for the path. If they have no master, offer to bootstrap one from references/master-resume-schema.md.

---

Workflow

Step 1 — Fetch and parse the posting

Use WebFetch on the URL. If it fails (auth, paywall, anti-bot), ask the user to paste the description.

Extract:

• Job title, company, location, remote policy
• Required ("must-have") skills/qualifications
• Preferred ("nice-to-have") skills
• Key responsibilities
• Seniority (intern / junior / mid / senior / lead / principal / executive)
• Industry signals: vocabulary, certifications, regulatory mentions, tooling

Present a brief summary table to the user and wait for confirmation before proceeding if anything is ambiguous. If unambiguous, proceed.

Step 2 — Classify the industry

Pick one of these and load the matching reference file:

Industry	Reference	Typical signals
Tech / SWE / ML / Data	references/industry-tech.md	AWS, Kubernetes, ML frameworks, CI/CD, "ship", "scale"
Banking / Finance	references/industry-banking-finance.md	IB, trading, quant, AUM, P&L, Series 7/CFA, Bloomberg, Murex
Consulting	references/industry-consulting.md	McKinsey/BCG/Bain/Big4, "engagement", "client", case-study wording
Healthcare / Pharma / Biotech	references/industry-healthcare-pharma.md	GxP, FDA, HIPAA, clinical trial phases, RWE, pharmacovigilance
Academia / Research	references/industry-academia-research.md	PhD, postdoc, grant IDs, teaching, publications, peer review
Legal	references/industry-legal.md	JD, Bar admission, practice area, matter-level experience
General / fallback	references/industry-general.md	Anything else — ops, PM, marketing, sales, HR, education

If the posting straddles two (e.g., "ML engineer at a hedge fund"), load both references and prioritize the one matching the employer's industry for tone/format, the other for skill keywords.

Step 3 — Read the master resume

Read the master file in full. Note the tags/markers the user uses (e.g., [tech], [finance], [research]) and honor them when scoring relevance. Identify:

• Summary variants (many master resumes list 2–3)
• Roles with bullet pools
• Skill categories
• Projects, publications, certifications, awards, volunteering
• Any explicit "always include" or "rarely include" notes

Step 4 — Score and select content

Score each item against the posting:

• High — directly matches a required skill/responsibility → include
• Medium — matches a preferred skill or adjacent domain → include if space permits
• Low — no clear connection → exclude (even if the user is proud of it)

Selection rules (apply after industry rules):

• Page budget: default 1 page for <7 yrs experience, 1.5–2 for mid/senior, 2+ only when the industry explicitly allows it (academia CVs can be unbounded; consulting and tech cap at 1–2). The industry reference overrides.
• Summary: pick the master's variant closest to the role, or synthesize from existing sentences. Name the target company/team only if the user provides that context or approves it. Never fabricate.
• Bullets: reorder within each role so top-relevance comes first. Drop low-relevance bullets. Keep 3–7 bullets for the most recent role, 2–4 for older ones. Merge very old or very short stints if they crowd the page.
• Skills: reorder categories so job-relevant ones lead. Do not add a skill that isn't in the master. If the posting asks for something the user doesn't have, flag it at the end (see Step 7).
• Publications / projects / certifications / awards: include per industry reference guidance.
• Keywords: rephrase lightly to match the posting's vocabulary when the underlying fact is the same (e.g., "distributed training" ↔ "distributed systems" when accurate). Never stretch.

Step 5 — Detect output format

Check the project layout in this order:

Signal	Format	Guide
main.tex + sections/ + .sty	LaTeX	references/output-latex.md
.docx template or docx/ folder	DOCX	references/output-docx.md
Only .md files	Markdown → PDF	references/output-markdown.md
Nothing resume-like	Ask the user	—

If the project has a CLAUDE.md, read it and follow its LaTeX/structural conventions precisely — it overrides the generic guidance in the output reference.

Step 6 — Generate and build

Edit existing files in place (never create new ones unless the format is brand-new). Follow the output-format reference for escaping, environments, and the build command.

Build. If it fails, read the log, fix, retry — don't paper over errors.

Step 7 — Save, name, and report

• Copy the built artifact to output/<company-slug>_<YYYY-MM-DD>.<ext> (create output/ if it doesn't exist).
• Leave the canonical file (e.g., output/resume.pdf) as-is.

Report back with:

1. Target — Job Title at Company
2. Industry detected — and which reference was applied
3. Summary variant used
4. Bullets included / excluded — small table per role with one-line reasons
5. Skills promoted to front
6. Gaps flagged — requirements from the posting with no match in the master (so the user knows what to address in the cover letter, or whether to upskill)
7. Output path
8. Git hint — git checkout -b apply/<company-slug> && git add -A && git commit -m "Tailor resume: <Company> — <Title>" to save, git checkout -- . to discard.

---

Hard rules

1. Truth — every claim traces to the master. When in doubt, leave it out.
2. Locale — match the posting's locale for dates (MM/YYYY US; DD/MM/YYYY EU), spelling (US vs UK), and privacy norms (photo and DOB acceptable in DE/FR/JP CVs, never on US résumés).
3. Length — hit the page budget for the industry. If you can't fit, cut low-relevance content before shrinking fonts.
4. ATS — follow references/ats-guidelines.md when the employer uses an ATS (most do — assume yes unless the role is a fellowship / academic app).
5. No embellishment — don't turn "contributed to" into "led"; don't inflate numbers; don't add skills to match the posting if they're absent from the master.

---

When the user asks for a new industry

If the user requests an industry not covered (say, "real estate" or "defense"), create a new references/industry-<slug>.md using references/industry-general.md as a skeleton, research the conventions briefly, and add an entry to the table in Step 2.