Audited 2026-10-18 · Sources: GitHub repo + README, tool structure, install scripts. Tabletop audit — nothing installed.
The card
| Repo | MadsLorentzen/ai-job-search |
| Stars | 53.9k (GitHub API, audit date) |
| License | MIT |
| Born | March 2026 |
| What it is | An automated job-hunting pipeline built on Claude Code — portal CLIs search listings, rank matches, generate tailored LaTeX CVs and cover letters, track applications in a Notion dashboard |
| Requires | Claude Code with a paid plan or API credits — no free tier |
What problem it solves
Job hunting is a numbers game played by hand — this makes it a pipeline. Search portals, rank fit, generate a custom CV per posting, track everything. For someone facing months of applications, the appeal is obvious.
Our gate findings — and an honest correction
This deserves a fairer read than "skip it." The engineering is better than the use case:
| Gate | Finding |
|---|---|
| Upstream identity | Verified — named solo maintainer, MIT, active discussions, honest scam disclaimer ("no crypto, no sponsorship program — anything claiming otherwise is a scam") |
| Security engineering | Genuinely careful. A security_guards.py CI check enforces permission allowlists; the portal-CLI installer skips packages with install-time lifecycle scripts (the classic malware vector) instead of blindly bun install-ing them; Notion sync is one-way, read-only, OAuth-scoped |
| Data handling | Local-first files are the system of record — the repo owns truth, dashboards are views |
So why the NO-GO? Not the code — the workflow. The pipeline's function is broadcasting your identity: name, resume, work history, phone, sometimes address — at platform scale. Bulk-applying collides with every major job board's terms of service, and the entity that gets rate-limited, filtered, or flagged is you, by your real name. A beautifully-built tool can still automate a risky strategy.
Verdict: NO-GO (workflow risk, not code risk). If you must automate parts of a search, keep human review on every outbound application — the tool's own architecture supports that; the "spray at scale" use doesn't.
Best for
Honestly, as a reference: its security guards and one-way-sync design are worth reading if you build agent pipelines. As a spray-and-pray job cannon, the verdict stands.