How to use ai-job-search to tailor a CV and cover letter to each posting with Claude Code

ai-job-search is an open-source job application framework that runs on Claude Code. Fill in your profile once and it finds postings, scores your fit, writes a CV and cover letter tailored to each posting in LaTeX, then has a reviewer agent revise them. A geophysicist who lost their job built it for their own search and open-sourced it (they say it took 69 applications to land a role). The repo has passed 50,000 GitHub stars and topped GitHub's daily trending list on October 19.

You need Claude Code (a Pro, Max or Team plan, or API credits), Python 3.10+, Bun, and a LaTeX distribution with lualatex and xelatex.

1. Get the repository. The README walks you through forking, but forks of public repos are always public, and /setup writes your name, contact details, work history and salary expectations into tracked files. For your own job search, push to a private repository and keep the original as upstream.

git clone https://github.com/MadsLorentzen/ai-job-search.git
cd ai-job-search
git remote rename origin upstream
gh repo create ai-job-search --private --source=. --push

1. Install the job-portal search tools.

python3 tools/install_portal_clis.py

1. Open Claude Code and build your profile. If you put a CV PDF or LinkedIn export in the documents/ folder, it reads those; otherwise paste a CV or answer its interview questions.

/setup

1. Find postings and turn the one you pick into an application. If a site can't be fetched, paste the full job description instead of the URL.

/scrape
/rank
/apply https://example.com/jobs/1234

/scrape collects postings sorted by fit. When there are too many to skim, /rank batch-scores them into a shortlist. A single /apply runs fit evaluation, CV and cover letter drafts, review and revision. Once an interview is booked, /interview builds likely questions and a mock interview, and /outcome records what happened.

The bundled search skills target Danish job boards, so use /add-portal to generate a search skill for boards in your market. Postings are treated as untrusted input, but that defense is instruction-level, not a sandbox, so skim what was fetched and written before you send anything from an unfamiliar site.