Two open source AI job application frameworks built on Claude Code just crossed a combined 77,000 GitHub stars, and one of them added nearly 10,000 in a single week. Career-Ops scans job portals, grades every listing from A to F across ten weighted criteria, writes ATS optimized CVs per posting, flags ghost jobs, and even finds the hiring manager you should message. Its creator used it to evaluate 740 listings and landed a Head of Applied AI role. The smaller AI Job Search framework takes a fork it and go approach: fill in your profile, run three commands, and Claude drafts your applications while a second agent critiques them. Neither tool submits anything for you. Both run on a standard Claude subscription. Best for: job seekers drowning in applications who already pay for Claude. Not ideal for: anyone expecting a one click spray and pray bot.
You apply to 40 jobs. You hear back from two. One is a rejection sent at 3:47am, which tells you exactly who read your resume.
Nobody did. Software did.
Companies have been screening candidates with AI for years, and everyone just quietly accepted that a robot gets first look at your career. This week, the other side of that transaction showed up on GitHub trending. Job seekers built their own robots. And tens of thousands of people are starring them.
The 77,000 Star AI Job Application Movement
Two projects are driving this. The big one is Career-Ops, an open source AI job application system sitting at 59,000 stars. The fast one is AI Job Search, a fork and go framework that pulled in nearly 10,000 stars this week alone. Over 5,000 of those landed in a single day.
Both are built on Claude Code, the terminal tool we covered in our roundup of the best GitHub repos for Claude Code productivity. Neither one is a product. There’s no signup page, no pricing tier, no waitlist. You clone a folder, feed it your resume, and your AI subscription becomes a job search department.
Career-Ops came from a developer named Santiago Fernandez, who spent months applying to jobs the hard way and decided to engineer his way out. His pitch sits right at the top of the README: companies use AI to filter candidates, so he gave candidates AI to choose companies. He told Business Insider how his AI job application pipeline chewed through over 740 listings before he landed a Head of Applied AI role.
Then he open sourced the whole thing. Now it has a Discord community, a Product Hunt launch, translations of the documentation into 15 languages, and press coverage in two countries. For a folder of markdown files, that’s a career.
One housekeeping note before the details: neither project is affiliated with Anthropic. These are independent builders who happened to pick Claude Code as the engine. Career-Ops runs on other AI coding tools too (OpenCode, Codex, Copilot, and Grok’s CLI among them). The movement is bigger than any one vendor.
What the Machine Does All Day
Career-Ops turns Claude Code into a pipeline. It scans job portals automatically, covering Greenhouse, Ashby, Lever, and 45 preconfigured company boards including Anthropic and OpenAI. Paste a URL and the auto-pipeline kicks off: a full evaluation, a keyword injected PDF resume tuned to that exact posting, and a tracker entry. All from one action.
The evaluation is the core of it. Every listing gets graded from A to F across ten weighted criteria. The report comes back in six blocks: a role summary, a CV match analysis, a leveling strategy, compensation research, personalization notes, and interview prep built around structured behavioral stories.
That last one deserves a pause. The system maintains an interview story bank, accumulating your strongest behavioral answers across every evaluation it runs. After a few weeks you end up with a small library of polished stories that can answer nearly any “tell me about a time when” question. Most people rebuild those answers from scratch the night before every interview. This thing versions them like code.
It Finds the Human Behind the Posting
Applications get you in the queue. Conversations get you hired. Career-Ops knows the difference, which is why it goes past the paperwork.
Its research mode digs into the company itself. It surfaces recent moves, engineering culture, AI strategy, and the specific angle your profile should take. A separate contact discovery mode identifies the recruiter, hiring manager, or team peer worth reaching out to. Then it drafts a LinkedIn message under 300 characters tuned to that contact type. There’s also a formal email drafter that turns an evaluated posting into a subject line, body, and attachment checklist.
Draft only, though. The system never sends, submits, or clicks anything on your behalf. More on why that matters in a minute.
The Scam Filter Might Be the Best Part
My favorite feature is Block G, a legitimacy check that flags scams and ghost jobs before you waste an evening on them. Considering how much of the modern job board is listings nobody intends to fill, an AI that calls that out might save more time than the CV generator does.
There are negotiation scripts too, including one specifically for pushing back on geographic pay discounts. And a terminal dashboard lets you browse, filter, and sort your whole pipeline without opening a spreadsheet ever again.
The Twist: It Refuses to Spam
Everyone’s first assumption about an AI job application tool is a spam cannon, a bot blasting 500 identical applications into the void. That assumption is wrong, and the wrongness is the best part.
Career-Ops is built as the opposite. The README warns against spray and pray in bold, then behaves like a picky recruiter. The system recommends skipping anything that scores below 4 out of 5. And it never submits an application on its own. You always click send. The AI’s job is finding the five listings worth your energy out of the 200 that aren’t.
Which is a funny inversion when you think about it. The candidate’s robot has stricter quality standards than most corporate screening software ever did.
The documentation is also honest about the learning curve, and the honesty reads like relationship advice. The first evaluations won’t be great, it warns, because the system doesn’t know you yet. Feed it your CV, your career story, your proof points, what you want to avoid. The comparison it uses: onboarding a new recruiter who needs a week to learn you before becoming invaluable. That framing is doing something no SaaS landing page ever does. It’s lowering your expectations on purpose.
Setting It Up Takes One Command
Here’s the part that surprised me, because open source tools usually punish regular people at the install step. Career-Ops doesn’t.
The whole thing starts with a single command: npx @santifer/career-ops init. That clones the latest release and installs everything. You open Claude Code inside the folder, and on first launch it walks you through setup conversationally. Your CV, your profile, your target roles, all configured by chatting. Nothing to edit by hand.
Customizing it works the same way. Want the scoring weights changed, the target roles switched to marketing instead of engineering, or five new companies added to the portal scanner? You just ask. The system is designed to be modified by the AI running it, since Claude reads the same files it operates from.
The prerequisites are mild by open source standards. You need Node installed (five minutes, one download) and an AI coding CLI, which most readers of this site already have. If Claude Code is new to you, it installs with one command and the free tier is enough to test drive the system before committing.
Cost wise, you’re mostly covered by whatever Claude plan you already have. We break down what the subscription gets you in our Claude Pro review, and Career-Ops even ships a guide for running on a budget with cheaper or local models. For a tool this capable, the marginal cost of using it rounds to zero.
The Fork It and Go Version
AI Job Search, the one currently rocketing up GitHub trending, takes a scrappier path. Fork the repo, run /setup, and Claude either interviews you or reads a documents folder (CV, LinkedIn export, diplomas, old applications) to build your profile. After that the loop is two commands: /scrape searches job portals and ranks matches by fit, /apply drafts a tailored CV and cover letter for whichever posting you pick.
The clever part is the drafter and reviewer setup. One Claude agent writes your application, then a second agent critiques it and forces a revision before anything reaches you. Built in quality control, because the author knew first drafts are where AI writing goes to embarrass people.
It keeps going after you hit send, too. A /rank command batch scores a pile of scraped listings into a shortlist with honest per job strengths and gaps. An /interview command builds a prep pack from the exact CV and cover letter the interviewer read. It researches your interviewers and runs a mock interview, with an explicit rule against inventing experience to cover gaps. An /outcome command records what happened, and once a few applications resolve, the system recalibrates its fit scoring based on which ones got interviews.
Read that last sentence again. The framework learns which of your applications convert and adjusts. That’s a feedback loop most paid career services don’t have.
Two caveats keep it from being the default recommendation. The portal integrations target the Danish job market out of the box (there’s a command to generate a scraper for your local boards, but that’s an extra step). And it compiles CVs with LaTeX, which means installing a typesetting system that has ruined stronger nerves than yours. Career-Ops sticks to friendlier PDF generation. Regular people should start there.
The Ecosystem Growing Underneath It
Zoom out from the two headliners and the whole category is sprouting. A plugin called job-apply auto fills application forms on LinkedIn, Greenhouse, Ashby, and Workday. A project called jobpilot searches boards, generates cover letters, and preps interviews as a plugin for both Claude Code and Codex. A resume skills collection built for job seekers and career changers has crossed a thousand stars. There’s even a Chinese framework doing batch applications on Boss Zhipin, the biggest hiring platform in China, because this pressure is global.
Different builders, different countries, same month. When strangers on opposite sides of the planet independently build the same tool, that’s demand talking.
The pattern underneath all of them is the one we flagged in our GitHub repos pillar: Claude Code stopped being a coding tool a while ago. It’s a general purpose agent that happens to live in a terminal, and non coding use cases like this one are where the weird growth is happening.
The Part Nobody Wants to Say Out Loud
Time for the cold water, because every article hyping these tools skips it.
If 77,000 people run the same AI job application system, the applications start converging. Same evaluation logic, same keyword mirroring, same forward looking cover letter framing. Recruiters already complain about a flood of AI written applications that read identically. Some screening vendors now advertise AI detection as a feature. The tailoring these frameworks do per posting helps them stand out from lazy ChatGPT copy paste. But an arms race only escalates. Whatever edge early adopters have right now shrinks as adoption grows.
There’s a terms of service gray zone too. The tools that stop at drafting are clean, but the ones that auto fill forms on LinkedIn brush against platform rules that prohibit automation. LinkedIn has suspended accounts for less. The two headline frameworks stay on the safe side of that line by never touching the submit button, which is one more reason the human in the loop design matters beyond ethics.
There’s also the accuracy problem. These systems draft claims about your experience, and AI drafts drift. Both projects build in guardrails (the reviewer agent, the human approval gate, the no invented experience rule), but the person clicking send owns every word. Submitting a hallucinated qualification isn’t the robot’s problem in the interview. It’s yours.
And the quiet irony: the best feature in either tool might be the one that tells you not to apply. A system that vetoes 195 of 200 listings is really a system for protecting your time and your dignity from a job market that stopped respecting both. That’s less a productivity hack and more a survival tool.
Why This Blew Up Now
The timing isn’t a mystery. Tech has logged over 100,000 layoffs in 2026, and companies keep name checking AI as the reason. Meanwhile the Stanford HAI 2026 AI Index found entry level hiring already declining in the exact fields where AI boosts productivity most.
So the market is brutal and the screening is automated. Every applicant knows their carefully written cover letter gets keyword scanned by software before any human sees it. For years the asymmetry only ran one direction: companies had the machines, candidates had Sunday afternoons and a Word template.
That asymmetry just ended. For the price of a Claude subscription, anyone can run the same caliber of automation the HR department uses. You don’t need to be a developer. Both projects are built for the fork it, fill it in, run it crowd, and Career-Ops in particular installs with less friction than most WordPress plugins.
77,000 stars in a hiring market this cold says people got the message.
What Hiring Looks Like From Here
Play this forward. The company’s AI screens applications. The candidate’s AI wrote them, after grading the company, researching the team, and picking the hiring manager to charm. Two pieces of software, negotiating on behalf of humans who haven’t met yet.
The corporate world will complain about this, loudly, and the complaints will be hilarious. Companies spent a decade making candidates perform keyword gymnastics for parsing software, then rejected them without a human glance. After all that, calling the candidate’s automation unfair takes a special kind of nerve.
The resume was always a formatting exercise pretending to be a merit test. Now both sides have machines for the formatting. What’s left is the part that always mattered: two people, in a room, figuring out if they want to work together.
The robots just handle the paperwork on the way there.
