<- all tokdocs
2026.10.19@vibewithkevin

The 50,000-star claim holds (56,232 at fetch), but four of the repo's six default job portals are Danish

Career
Watch on TikTok

View on TikTok ->

The artifact is a 69.20-second vertical clip at 1080x1920 in an MP4 container (ffprobe reports the format family as mov,mp4,m4a,3gp,3g2,mj2), video encoded as HEVC Main profile in yuv420p at 30 fps and 510,427 bps, audio as HE-AACv2 at 44,100 Hz stereo and 64,167 bps, for an overall bitrate of 581,378 bps and a file size of 5,028,927 bytes. The extracted audio.mp3 runs 69.25 seconds. The post timestamp is 1792379214, which is 2026-10-19 03:06:54 UTC. The uploader handle is vibewithkevin, the channel nickname is "Vibe with Kevin", and the audio track is credited as "original sound" by "Vibe with Kevin", so there is no licensed music. The description is 130 characters: "Your next resume, cover letter, and interview prep, all from one Claude Code repo. #claudecode #jobsearch #ai #aitools #jobhunting". At capture on 2026-10-19 the metadata recorded 1,204 views, 44 likes, 2 comments, 11 reposts, and 59 saves. I read 13 of the 35 extracted frames (001, 004, 007, 010, 013, 016, 019, 022, 025, 028, 031, 033, 035). Seven of them show a bearded man in a black henley speaking to camera in a home office with glass shelves, a galaxy print, Scrabble and Monopoly boxes, a book spine reading "SUPERAGENCY", and a nameplate reading "Kevin J. Magnan". Frame 001 carries a white caption box reading "Before your next job search, use this repo". Frames 010 and 013 are a full-screen terminal mockup headed "STEP 1 · SET IT UP" showing "Claude Code v2.1.267", "Opus 5 · ~/projects", the prompt "Copy https://github.com/MadsLorentzen/ai-job-search into a new private repo of mine. Don't fork it. Install it following the repo's setup guide, then walk me through /setup." and a status bar reading "Opus 5 | projects | ~2% of 1000k tokens". Frame 016 overlays a panel titled "README · ai-job-search" with three columns, "/setup", "/scrape", and "/apply ". Frames 019, 022, and 025 use a stock illustration of two marked-up documents tagged "/apply ", "FIT SCORE", "score + recommend", "2ND AGENT REVIEWS IT", "CV", "COVER LETTER", and "RESUME". Frames 028 through 033 show a white card reading "THE BUILDER'S OWN SEARCH", "69 applications", "20 interviews", "1 job", with a "/interview" tag added at 031. The transcript is 367 words. The spoken argument ends at 67.9 seconds with "what they're looking for", a sign-off that Whisper renders as "Follow up for more. All right. All right. Follow up for more." runs from 68.2 to 69.3 seconds, and Whisper then emits 31 more segments from 69.8 to 90.5 seconds, 98 words past the 69.25-second end of the audio, mostly "Thank you" repeated plus "Bye-bye", "90 minutes for you", and a stray CJK character. I treat those 98 words as invented and do not quote them. Whisper also renders "Claude" as "Cloud" throughout; I quote the transcript as written.

The 50,000-star figure is real and already low

The transcript opens at 0.0 to 4.3 seconds with "Before searching for your next job, use this repo that already has 50,000 stars on GitHub." I fetched the GitHub API record for MadsLorentzen/ai-job-search on 2026-10-19. It reported 56,232 stars, 17,942 forks, and 51 open issues, a creation date of 2026-03-18T14:33:04Z, a last push of 2026-10-19T16:42:33Z, and an MIT license. The star-history page I fetched recorded the repo trending for 19 days with a best rank of #1 across all languages on 2026-07-08. The video rounds down a number still climbing on the day it posted.

The on-screen card at frames 028 through 033, "THE BUILDER'S OWN SEARCH, 69 applications, 20 interviews, 1 job", is accurate to the source and correctly attributed. The README says: "I'm a geophysicist by training. When my position was cut in late 2025, I built this framework to run my own job search" and "Sixty-nine tailored applications, twenty first interviews, and one signed contract later, I started as an AI engineer in June 2026." The audio never speaks those numbers; they appear only as a graphic, and the graphic names the builder rather than the speaker.

The audio says "local copy"; the README says private repository, because forks are public

At 15.6 to 27.7 seconds: "First, tell Cloud to copy the repo on your computer as a local copy. You do this because in the next step, you're going to run through the setup that helps Cloud fill in your entire profile, aka your resume, and you don't want that floating around the internet." The on-screen prompt at frame 013 says something more specific: "Copy https://github.com/MadsLorentzen/ai-job-search into a new private repo of mine. Don't fork it."

The screen matches the README, which warns that a fork of the repo is always public, because GitHub does not allow private forks of public repositories, and that /setup writes personal data (it lists name, contact details, employment history, and salary expectations) into tracked files. Its instruction for anyone using the copy for their own search rather than contributing back is to use a private repository with the original as upstream. SETUP.md section 8 gives the steps: create a private repository, push there, and run git remote add upstream https://github.com/MadsLorentzen/ai-job-search.git to keep receiving updates.

A plain local clone also keeps the profile off the internet as long as you never push it, so the audio's advice works. What it skips is the mechanism the README cares about: a private remote lets you keep pulling framework updates without exposing the profile files. The GitHub page title ends with "Fork it and own it.", which reads as the opposite of the README's guidance for personal use; the README's warning is the operative instruction.

"Multiple job portals" means four Danish boards, LinkedIn, and one aggregator

At 27.7 to 36.5 seconds: "Then, Cloud Code will use a scrape feature that searches multiple job portals for positions matching the profile you just set up with it, and it will rank them for you." The README's one-line description of /scrape is "Search multiple job portals, deduplicate, and rank by fit", so the verbs are accurate.

The video does not say which portals. The README lists Jobindex, Jobnet (the Danish government portal), Akademikernes Jobbank, and Jobdanmark, all Danish, plus LinkedIn public job listings and freehire.me, a tech-jobs aggregator the README describes as multi-market. For a US viewer, LinkedIn and freehire.me are the only defaults that will return relevant results without extra work. The README provides /add-portal, which generates a job-portal search skill for a local job board, so the gap is closable. It is still work that "searches multiple job portals" leaves out for a video pitched at a US audience.

/apply produces two documents, and the video counts three

At 36.8 to 50.7 seconds: "Next, run the apply command to use the job description, the requirements, and your profile to score the job for fit, create all the necessary assets like a cover letter, a CV, and a resume, and then even run a second agent to review and present you the final output." Frame 025 reinforces the count with three separate labels, "CV", "COVER LETTER", and "RESUME".

Most of this matches the source. The README's /apply <url> is a full pipeline: evaluate fit, draft a tailored CV and cover letter, have a second agent review, revise, compile PDFs, and verify ATS readability. The reviewer step reads: "a second Claude agent, spawned with a fresh context, researches the company and critiques the drafts. The drafter then revises." The "FIT SCORE" tag on screen matches "Evaluate fit, Score & recommend" in the README's workflow diagram, the one shown at frame 016.

The document count does not match. The README produces a CV (compiled with lualatex, with a two-page limit and a layout check) and a cover letter (compiled with xelatex). The word "resume" appears once in the README, in a sentence about generic "LaTeX-resume templates" that break in the PDF, and never as an output. A viewer who treats the CV as the resume gets what the description promises, in two files. The video's own description, "Your next resume, cover letter, and interview prep", has the count right.

/interview matches the README closely, and the video undersells it

At 50.7 to 67.9 seconds: "when you land the interview, you can come back to this repo and run the interview command for it to prep you for the interview, since it intimately knows both the job you apply for and your resume. So, it knows questions they ask you, like where to ask for strengths and gaps based on the company and what they're looking for."

The README's /interview entry says it "preps you for a scheduled interview on a tracked application", builds "a stage-specific prep pack from the application's archive (the exact posting, the CV and cover letter the interviewer actually read, feedback recorded from earlier rounds)", "researches the company and interviewers with a verify-before-use rule, maps likely questions to your STAR examples, and offers a mock interview", and closes with "Gaps get honest bridge answers, never invented experience." The command file at .claude/commands/interview.md adds that it reads feedback from earlier stages, researches recent news and public profiles of known interviewers, ranks probable questions by likelihood, builds a consistency brief of claims the interviewer will probe, and ends by prompting you to log the result with /outcome.

Both of the video's claims, that the command knows the job and the submitted documents and that it anticipates questions including gaps, are in the source. What the video undersells is the mock interview, a live roleplay with feedback that the README treats as a core part of the command. The video also presents four commands as the whole tool. The README lists fourteen, including /rank, /outcome, /upskill, /add-portal, and /gmail-sync.

What the terminal shows, and what "under a minute" leaves out

Frames 010 and 013 show "Claude Code v2.1.267" and "Opus 5", with a status bar reading "~2% of 1000k tokens". I checked both against live sources on 2026-10-19. The Claude Code changelog on GitHub has a 2.1.267 entry (its first bullet adds a maxEffortLevel setting) and its top entry when I fetched was 2.1.303, so the version on screen is a real, slightly older build. Anthropic's model page for Claude Opus 5 lists it as "Legacy. Released July 24, 2026." with model ID claude-opus-5, a 1M-token context window, and the note "Although Claude Opus 5 is still available, you should consider migrating to Claude Opus 5.5." The models overview lists the current lineup as Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5, and Claude Haiku 5.5, all with 1M-token context windows, which matches the "1000k tokens" on screen. The README does not recommend any particular model.

At 4.7 to 5.9 seconds the transcript says "Here's how to do it in under a minute." The explanation runs under a minute. The install does not. The README states "Claude Code has no free tier: you need a Claude Pro/Max/Team subscription or Anthropic API credits", and Anthropic's Claude Code overview says most surfaces require a Claude subscription or a Console account. SETUP.md lists Python 3.10+ for the salary lookup tool, Bun for the TypeScript job-portal CLIs, and a LaTeX distribution to compile the generated .tex files, naming MiKTeX on Windows, MacTeX on macOS, and texlive-full on Linux, with pypdf or Poppler as optional extras for the ATS check. The video mentions none of these. For a viewer who has never installed LaTeX, that prerequisite is the longest step.

On the speaker: the nameplate in frame 001 reads "Kevin J. Magnan". His site, kevinjmagnan.com, describes him as an "AI engineer and educator" who leads AI strategy at Slalom and links the @vibewithkevin account with "324 videos" and "13,000+ followers". The TikTok profile page returned only its title when I fetched it.

Key Takeaways

  • Verified: The artifact facts are confirmed by ffprobe and the post metadata. 69.20 seconds, 1080x1920, HEVC Main at 30 fps, HE-AACv2 audio, 5,028,927 bytes, posted 2026-10-19 03:06:54 UTC, 1,204 views and 44 likes at capture on 2026-10-19.
  • Verified: The repo exists at github.com/MadsLorentzen/ai-job-search and the "50,000 stars" claim is low. The GitHub API reported 56,232 stars and 17,942 forks on 2026-10-19 for a repo created 2026-03-18.
  • Verified: The on-screen "69 applications, 20 interviews, 1 job" card matches the README, which reports sixty-nine applications, twenty first interviews, and one signed contract leading to an AI engineer role in June 2026, and the card credits the builder rather than the speaker.
  • Correction: The video says /apply creates "a cover letter, a CV, and a resume" and frame 025 labels three documents. The README produces two, a LaTeX CV and a cover letter; "resume" appears in the README only in a remark about generic templates.
  • Partial correction: The audio says to make "a local copy" on your computer. The README's instruction is a private GitHub repository with the original as upstream, because a fork of a public repo is always public and /setup writes name, contact details, employment history, and salary expectations into tracked files. The on-screen prompt at frame 013 follows the README.
  • Clarification: "Searches multiple job portals" means Jobindex, Jobnet, Akademikernes Jobbank, Jobdanmark, LinkedIn, and freehire.me by default. Four of the six are Danish; /add-portal exists for other boards.
  • Clarification: "Under a minute" describes the explanation. Setup requires a paid Claude plan or API credits, Python 3.10+, Bun, and a LaTeX distribution, none of which the video mentions.
  • Context: The terminal on screen shows Claude Code v2.1.267 and Opus 5. On 2026-10-19 the Claude Code changelog had 2.1.267 as a real entry below a current 2.1.303, and Anthropic's docs listed Claude Opus 5 as a legacy model released July 24, 2026 with a 1M-token context window, which matches the "1000k tokens" status bar.
  • Context: The speaker is Kevin J. Magnan per the nameplate in frame. kevinjmagnan.com describes him as an AI engineer and educator who leads AI strategy at Slalom.

Resources

Published October 19, 2026. Writeup generated from a favorited TikTok.