Turn Claude Code Into a Job-Hunting Machine
A one-page field guide to MadsLorentzen/ai-job-search (19k+ stars, MIT) — the open-source framework that turns Claude Code into a full job-application department. Fork it, fill in your profile, and Claude evaluates postings, tailors your CV per role, writes cover letters, and preps you for interviews — and it checks every application the way an ATS parser actually reads it, not just how it looks. THE THREE-COMMAND SPINE: /setup (build your profile from a documents folder, a pasted CV, or a guided interview) -> /scrape (search multiple job portals at once, dedupe, sort by fit; then /rank to batch-score every posting across 5 fit dimensions into a ranked shortlist) -> /apply <url> (the full pipeline). HOW /apply WORKS — a drafter-reviewer pipeline: 1) parse the posting, 2) evaluate genuine fit, 3) draft a tailored CV + cover letter in LaTeX, 4) spawn a SECOND fresh-context Claude that researches the company and critiques the draft like a skeptical recruiter, 5) revise, 6) compile + visually inspect the PDF (lualatex CV / xelatex cover) until it's exactly 2 pages / 1 page with no orphaned titles and a visible signature, 7) ATS-check by extracting the PDF text layer (pdftotext) and scoring keyword coverage the way a parser sees it, 8) present with a verification checklist. WHY IT'S DIFFERENT: a PDF verification loop that catches layout that 'looks fine in the .tex' but breaks in the PDF; ATS verification on the actual text layer (not the rendered page); relevance-weighted CV cutting (scores each line by relevance, uniqueness, and cover-letter dependency and cuts the weakest first); and a hard honesty rule — it never fabricates a skill and a genuine gap stays visible, never keyword-stuffed. THE ROBOT-PROOF CHECKLIST: contact details as real selectable text (not an icon glyph), reading order that survives extraction, only keywords your profile genuinely supports, CV exactly 2 pages / cover exactly 1, and run /rank before /apply. SEVEN MORE COMMANDS: /interview (stage-specific prep pack from the exact posting + a mock with STAR mapping and honest bridge answers), /outcome (record results and calibrate the fit framework from what actually got interviews), /upskill (turn the gap between you and target jobs into a skill heat-map + learning plan), /expand (enrich your profile from GitHub/portfolio/published work), /add-template (register your own LaTeX design with a mandatory test compile), /add-portal (give it any job-board URL and it generates a search skill after test-running a live query). The core is country- and language-agnostic; the shipped Danish portals are just a demo. Full credit to Mads Lorentzen — it's free and MIT-licensed. Independent open-source project, not an official Anthropic product.
Free. No spam. Unsubscribe anytime.