<- all tokdocs
2026.09.17@github.signals

career-ops Scores Job Listings Against Your CV Inside Claude Code, and Never Clicks Submit

Career
Watch on TikTok

View on TikTok ->

career-ops is a real, MIT-licensed repo at 72,113 stars that scores job listings 1 to 5 against your CV and refuses to submit anything on your behalf. The video scrolls the GitHub README for career-ops-hq/career-ops and narrates the pitch: scan portals, evaluate every listing, get a score, find out if you actually fit before you apply. Almost everything on screen checks out against the repo and the project site. The word that needs qualifying is "private," and the headline outcome number rests on one person's job search.

What the tool actually does

career-ops installs with npx @santifer/career-ops init, which clones the repo into ./career-ops and installs dependencies. You then open an AI coding CLI in that directory. It is not a standalone binary with its own model. The CLI you already pay for does the reasoning, and career-ops supplies the prompts, modes, scanners, and file layout.

Paste a job URL and it produces a structured evaluation split into blocks A through H: role summary, CV match, level strategy, compensation research, personalization, interview prep in STAR plus Reflection format, a posting-legitimacy check that flags scams and ghost jobs, and a work-authorization signal. The global 1 to 5 score comes from holistic judgement across five dimensions rather than an arithmetic average, per the README. Two signals are explicitly score-neutral: the per-requirement importance column in block B and the legitimacy assessment in block G. Block H is only drafted at 4.5 and above.

The README states the system "strongly recommends against applying to anything scoring below 4.0/5." That is the actual mechanism behind the video's "only chase the opportunities that match." It is a threshold with a stated number, not a vibe.

The "private" claim needs a footnote

The transcript calls it "the smart, private way to filter the noise." The repo's own framing is local-first, which is a narrower claim. Your CV, profile, pipeline, and reports live on your machine as Markdown and YAML files. There is no career-ops account, no career-ops server, and no telemetry going to the project.

Your data still leaves your machine. Every evaluation is a prompt sent to whatever AI CLI you pointed at the directory, which for most users means Anthropic, OpenAI, or Google receives your resume and the job descriptions you are considering. The project site says as much: no uploads to career-ops servers, only data sent to AI CLI providers. The repo documents a genuinely private path in docs/RUNNING_ON_A_BUDGET.md, running through Ollama or any OpenAI-compatible endpoint, so local inference is supported. It is not the default, and the video does not mention the distinction.

The human-in-the-loop design is the point

The README carries an unusual instruction in an HTML comment next to the human-in-the-loop row, telling translators not to hedge the guarantee with words like "automatically" or "without your permission." The system never submits an application. It drafts cover letters, application emails, and LinkedIn messages, and it stops there. The FAQ repeats it: "It never submits, sends, or clicks anything."

That design choice rules out the mass-apply behavior that most AI job tools sell. It also means career-ops does not save you the last mile. You still paste, upload, and click. What it removes is the reading, the comparison, and the spreadsheet.

What is verified and what is one person's result

Verified from primary sources on 2026-09-19:

  • career-ops-hq/career-ops has 72,113 stars and 13,589 forks, MIT licensed, created 2026-04-04, still pushing commits today.
  • Release career-ops-v1.33.0 shipped 2026-09-16, one day before this video. The npm package @santifer/career-ops published 1.33.0 at the same time. The v1.33.0 badge visible in the video was current.
  • The maintainer is Santiago Fernández de Valderrama Aparicio, GitHub handle @santifer.
  • The Business Insider article linked from the README resolves and loads. The WIRED link points to wired.com.gr, the Greek edition, and returned a 403 to automated requests, so I confirmed the link target but not the article content.
  • The hired-count JSON the README badge reads showed 8 in the video and reads 10 as of 2026-09-18.

The banner stat, "740+ job listings evaluated · 100+ personalized CVs · 1 dream role landed," is the creator's own 2026 job search. The project site breaks it down further: 740 listings, 68 applications, 12 interviews, 1 offer. It is a sample of one, self-reported. The 10 verified hires are self-submitted GitHub issues on a public wall you can read, which is more transparent than most such counts and still not an audit.

The fork problem

Searching for career-ops surfaces a wall of forks before it surfaces the source. Individual user forks appear high in results carrying stale descriptions from older releases, some still advertising an A-F rubric and Gemini CLI support that the current README has moved past. With 13,589 forks, a person who installs from the first GitHub result they see can end up months behind on a project shipping releases weekly. The canonical repo is career-ops-hq/career-ops, which is what the video shows.

Key Takeaways

  • career-ops runs inside an AI coding CLI you already have. It is prompts, modes, and scripts, not its own model, and the only cost is your existing CLI subscription.
  • Every listing gets an A through H report and a 1 to 5 score, and the system tells you not to apply below 4.0.
  • Local-first is accurate. Fully private is not, unless you configure a local model through Ollama or an OpenAI-compatible endpoint.
  • The tool never submits an application. It drafts and stops, by explicit design.
  • The star count, fork count, license, maintainer, and v1.33.0 release date all check out against the GitHub API and npm registry.
  • Unverified: the WIRED Greece article content, since the site blocked my request. I confirmed only that the README links there.
  • Unverified: whether career-ops works for anyone other than its author at scale. The 740-listing result is one search by one person, and the 10 verified hires are self-reported.

Resources

Published September 17, 2026. Writeup generated from a favorited TikTok.