AI Job Search: Transform Claude Code into a job search pipeline to evaluate jobs, customize resumes, write cover letters, prepare for interviewsAI Job Search：把 Claude Code 改造成一条求职流水线，评估岗位、定制简历、写求职信、备面试

In a nutshell: Turn Claude Code from a “coding tool” into a “resume partner,” and the whole process is structured and repeatable, rather than just asking and generating a cover letter.

The core is a pipeline of five slash commands:/setup builds a personal profile,/scrape searches and weighs jobs from multiple recruitment sites,/rank scores and sorts the captured jobs according to the matching degree,/apply produces customized materials for a certain job, and/interview prepares interviews for you.

The README specifically states one thing: the core workflows (self-portrait, fit assessment, draft-review application pipeline) are language- and country-independent, but the built-in job search skills are done by the Danish market (Jobindex, Jobnet, Akademikernes Jobbank, etc.).

However, this model is itself a product: those search skills are replaceable, just follow the same structure and connect with your local mainstream recruitment platform.

It is embedded with a set of career coaching methodologies: structured assessment criteria, a forward-looking cover letter narrative framework, and optional salary benchmarking.

Step 1: Fill in the file:/setup provides three paths - read your documents/directory (put the CV PDF, LinkedIn export, degree certificate, recommendation letter, past application materials), paste a resume into the conversation to import, or ask the information in a question-and-answer manner. It will automatically recognize what materials you have and then decide which one to go, and the mode of reading the catalog is idempotent, and you can add things to the catalog repeatedly.

Step 2 Grab Jobs:/scrape searches multiple hiring sites for jobs that match your profile, and sorts them by match after deduplication. If there are too many positions you don't want to look at one at a time, you can run/rank to get a short list.

Step 3 Application:/apply will go through the entire process - assessing matches, drafting resumes and cover letters, reviewing the agent for faults, revising, outputting final drafts. Some recruitment sites block automated access, and you can glue the entire job description in when you can't catch the link.

Resumes and cover letters are generated in LaTeX and compiled into PDFs: resumes are compiled in lualatex (pdflatex on modern MiKTeX is often wrong due to fontawesome5 fonts), and cover letters are compiled in xelatex (because the template relies on fontspec).

Another practical detail:/apply comes with an ATS parsability check. Use pypdf or Poppler's pdftotext to determine whether your PDF can be parsed properly by the resume screening system. When both are missing, it will be relegated to a keyword visual check.

The overwhelming majority of people use AI to write cover letters when the question is “once in a while,” and you can use whatever the model writes. This frame breaks it into two characters.

The drafting role is responsible for generating the first version based on the job description and the real experience in your profile; the review role then intervenes and selects the questions according to a clear set of criteria before they are revised and finalized.

This is the same idea as code review in software engineering: generating and reviewing are separated to steadily push quality up.

The same idea applies to interview preparation -/interview generates the questions and responses you need to prepare for a specific role, rather than giving a generic interview eight strands.

If you only take one trick from this article, take this one: let the AI write it again, the cost is very low, and the effect is much worse.

Claude Code (CLI) - Note that it does not have a free file, you need a Claude Pro/Max/Team subscription or Anthropic's API credits (pay as you go, it's usually cheaper to use occasionally).

If you are using other agent tools (Codex, Gemini CLI, Antigravity, etc.), you can start with the repository's agents.md: the portal search class skill is ready to use out of the box, and the community fork is adapted to the full workflow.

Python 3.10 + and Bun (job search CLI relies on it).

A set of LaTeX distributions with lualatex and xelatex: TeX Live, MacTeX, TinyTeX or MiKTeX. With a lite distribution like TinyTeX/BasicTeX, you need to refill the extra macro packages with SETUP.md.

Optional: pip install pypdf for the ATS parsing check mentioned above.

/setup writes your personal data - name, contact details, resume, expected salary - to a file in the warehouse that is tracked by git.

GitHub does not allow private forks to be created for public repositories: the repository you fork out is always public.

So if you are using it for your own use and not to contribute to the upstream, the right thing to do is to create a new private warehouse, add this project as an upstream remote, and then write your file to the private warehouse. Section 8 of SETUP.md gives steps that take two minutes to complete.

The fork path should only be taken if you intend to contribute changes upstream.

This site organizes resources from the GitHub open source repository MadsLorentzen/ai-job-search (mit license), for learning exchanges only. The project is an independent open source project and is not affiliated with Anthropic. Please note when using: Do not push the warehouse containing your resume to a public location; please make sure to read and submit the generated materials yourself, and the resume and cover letter represent you. The process in this article is organized from the project public README and SETUP.md. The specific commands may evolve with the version. Please refer to the latest documents of the warehouse.