finds.dev

ai-job-search-dev/ai-job-search

The job search that runs on your machine. AI job application framework built on Claude Code: evaluate postings, tailor CVs, write cover letters, prep interviews. Fork it and own it.

View on GitHub

★ 67,534PythonMITupdated Oct 2026

// the take

A fork-and-own framework that turns Claude Code into an end-to-end job search assistant: it builds a profile from your CV/LinkedIn export, searches job boards, writes a per-posting CV and cover letter in LaTeX, and preps you for interviews. Built by one person who ran it on their own job hunt before open-sourcing it; it's aimed at developers willing to run Claude Code locally and maintain a LaTeX toolchain, not non-technical job seekers.

The drafter-reviewer split (a second agent with fresh context critiques the draft) plus a mandatory compile-and-inspect loop - lualatex/xelatex, page-count checks, and pdftotext-based ATS text-layer verification - actually addresses the broken-PDF problem most LaTeX resume generators ignore. Job postings are treated as untrusted input: a deterministic integrity gate screens for hidden text and injection-shaped lines before any agent reads a posting, and the fetch/review agents are read-only. The extension model is real rather than aspirational - /scrape auto-discovers any portal skill folder that matches the CLI contract, no registration step, and /add-template or free-form evaluation criteria plug in the same way. CI runs LaTeX smoke compiles, skill linting, and per-portal CLI unit tests, which is more test coverage than most prompt-driven frameworks bother with.

The core workflow has a hard dependency on a full LaTeX distribution with lualatex and xelatex - a real install burden, and font/package failures on something like MiKTeX are exactly the kind of toolchain debugging a tool pitched as 'fork and run' shouldn't require. There's no free tier: it needs a Claude Pro/Max/Team subscription or metered API credits, so cost scales with how many jobs you apply to. Maintained job-portal coverage is Denmark-only (Jobindex, Jobnet, Jobbank, Jobdanmark); everywhere else you're on LinkedIn's unauthenticated endpoints (against LinkedIn's terms, with an explicit keep-volume-low warning) or a community fork you have to find, read, and manually copy in yourself. Handing an agent your actual job search - what claims go in a cover letter, what counts as 'verified' - is meaningful agentic surface area, and the project's own security docs concede the defenses against a hostile posting are instruction-level, not a sandbox.

We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →

// at a glance

Stars
67,534
Forks
21,428
Language
Python
License
MIT
Last commit
27 Oct 2026

// nearby

If ai-job-search isn't quite it.

All repos →