Graphic tutorial: Claude Code as a job search pipeline, from filing to interview preparation图文教程：把 Claude Code 搭成求职流水线，从建档到面试准备走完一整轮

This step is the easiest for veterans to skip and the easiest for novices to step in:/setup will write your name, contact information, resume, and expected salary into the file tracked by git in the warehouse.

And GitHub doesn't allow private forks to be created on public repositories - what you fork out is always visible to everyone.

Therefore, if you use it yourself, the correct way is to create a new private warehouse, add the warehouse address of this project as upstream remote, and then pull the update from upstream. Warehouse SETUP.md Section 8 gives the complete order and can be done in two minutes.

Unzip the ai-job-search-master.zip downloaded from the "AI Tools" section of this site and put the code into your newly created private repository directory.

Make sure you have all three: Claude Code CLI available (requires Claude Pro/Max/Team subscription or Anthropic API credits, no free tier), Python 3.10 +, Bun installed.

Installation job search CLI: execute a loop in the repository root directory, and load the dependencies of the six tools jobbank-search, jobdanmark-search, jobindex-search, jobnet-search, linkedin-search, and freehire-search (given in both bash and PowerShell README).

The installation of linkedin-search and freehire-search can actually be omitted. They depend on zero runtime and run directly. Bun install just pulls the TypeScript type definition.

Optional but recommended: pip install pypdf, which will be used by the ATS resolvability check in the back/apply, and will degenerate into a keyword visual check if not.

Resume compilation also requires LaTeX: install either TeX Live/MacTeX/TinyTeX/MiKTeX to ensure that both lualatex and xelatex commands are available.

In the repository directory, execute claude into Claude Code and enter/setup.

It will allow you to choose one of three options: read your documents/directory (you can put a PDF of CV, LinkedIn export, degree certificate, letter of recommendation, past application materials), paste a resume into the conversation, or ask for information in a question-and-answer manner.

It will automatically detect what materials you have on hand before deciding which path to recommend. If you choose to read the catalog mode, this operation is idempotent - it is enough to fill the catalog with materials in the future and run/setup once again, so as not to mess up your files.

After the run, you will get a set of archive files, and all subsequent commands will read your background information from here.

It is recommended to write down the experience (project, number, result) in detail at this step, the quality of the resume and cover letter generated by the latter/apply depends directly on the granularity of this file.

Execute/scrape. It searches multiple recruitment portals for jobs that match your profile, removes duplicate jobs, and sorts them by match.

If you get too many jobs at once and don't want to look at them one by one, execute/rank first: it will batch all the jobs it captures according to the matching framework and give you a short list in order.

Pick the one you want to vote for the most from the shortlist, and its link can be fed directly to the next step. Each match in the/scrape result can be triggered directly/apply.

Note that the built-in search skills are based on the Danish market (Jobindex, Jobnet, Akademikernes Jobbank, etc.). If your target market is not here, you can rewrite your local recruitment platform according to the same structure, which is also the opening left in the design of this warehouse.

Execute/apply plus the job title link, for example/apply https://example.com/job/1234567.

Some recruitment sites will block automated visits, and you can paste the entire job description directly into it when the link is not caught:/apply.

Then it will run the whole chain: first evaluate your match with this position, then draft your resume and cover letter (LaTeX template, resume compiled with lualatex, job application credit xelatex compilation), then submit it to the review agent for picky and revision, and finally output the final PDF.

Once finalized, an ATS parsability check will be performed: if your PDF cannot be parsed properly by the resume screening system, it will prompt you - this check uses either pypdf or pdftotext, neither of which is sometimes downgraded to a keyword visual check.

The most important discipline: be sure to read it in its entirety before handing it over. This material represents you personally, and the numbers and statements should match your real experience.

Execute/interview after getting the interview invitation, it will generate a list of questions to be prepared and your answer materials for this specific role, instead of throwing a generic interview question to you.

Feed the results of each round of interviews and the questions asked back into your profile and project description, and the next round of application materials will be more accurate - the value of this framework is that it is cumulative.

If you are not using Claude Code, but Codex, Gemini CLI and other tools, start with the repository's AGENTS.md: portal search skills out of the box, and the community fork adapts to the full workflow.

Finally, there is the old saying: this set of things can help you compress the time you spend on changing your resume and writing a cover letter every week, but you still have to make your own judgments about who to vote for, whether it is worth going to, and how to talk about the offer.