JumpingJobs

A job-search and resume toolkit for AI coding agents. Build a durable profile, find and assess roles, tailor your resume to each posting, and generate interview prep, all from your agent's chat.

Works in Claude Code, Cursor, GitHub Copilot, Gemini CLI, Codex CLI, OpenCode, and Kiro.

---

What it is

One onboarding command, eight workflow commands, and two supporting skills:

Command	What it does
/init-resume	Start here. Scaffolds your workspace, adds the resume template, and builds your profile.md by reading your existing resume and interviewing you.
/find-jobs	Sweeps your configured job boards for new, on-profile roles, deduped against what you have saved.
/scrape-job <url>	Saves a posting as structured Markdown.
/assess-job	Scores a posting's fit against your profile and hard constraints.
/tune-resume	Tailors your resume to a posting (2 pages, honest, targeted).
/interview-prep	Deep company research into a full dossier plus a condensed cheat sheet.
/interview-debrief	Captures a round right after it happens — questions, stumbles, signals — and turns stumbles into next-round prep.
/assess-offer	Extracts every term from an offer letter, flags gaps and red flags, checks written terms against verbal promises, plans the negotiation and drafts the response.
/move-job	Moves a posting through the numbered pipeline stages, taking its whole file family along.

The two skills the commands rely on: applicant-profile (maintains profile.md, your source of truth) and interview-cheatsheet (a tight, in-the-room prep sheet).

---

Install

Claude Code

/plugin marketplace add jumpingjobs/jumpingjobs
/plugin install jumpingjobs@jumpingjobs

Reload plugins. Commands appear namespaced, e.g. /jumpingjobs:init-resume.

Cursor, Copilot, Gemini, Codex, OpenCode, Kiro

npx jumpingjobs install          # auto-detects your harness from the current project

Or target one explicitly / see what is supported:

npx jumpingjobs install --harness cursor
npx jumpingjobs list

This copies the skills into your tool's directory (.cursor/skills/, .github/skills/, .gemini/skills/, ...). Run npx jumpingjobs update later to refresh them.

---

Quick start

1. Onboard. Run /init-resume (Claude Code) or invoke init-resume after installing.
2. Build your profile. Hand it your existing resume and/or LinkedIn. It extracts what it can, then interviews you to fill the gaps, writing resume/<your-slug>/profile.md.
3. Get a starter resume. It produces resume/<your-slug>/<your-slug>-resume.html from the template.
4. Work the pipeline. /find-jobs -> /scrape-job -> /assess-job -> /tune-resume -> /interview-prep, moving postings with /move-job as their status changes.

---

How it works

Your workspace

resume/<your-slug>/
  profile.md                 # source of truth (verbose; you maintain this)
  <your-slug>-resume.html    # layout skeleton tune-resume fills per posting
  job-boards.md              # your boards + search config (edit for your market)
  job-postings/              # numbered so folders sort in pipeline order
    1-scraped/  2-applied/  3-interview/  4-lost/  5-archived/  6-won/

profile.md is the source of truth

Your resume is a lossy projection of profile.md. Keep the detail (real metrics, stories, constraints) in the profile; let /tune-resume trim a 2-page resume from it per job. After every tune, new facts surfaced in the conversation are folded back into profile.md, so it never goes stale. Maintain the profile, not the HTML.

---

Prerequisites and localization

• Browser automation makes /find-jobs and deep /interview-prep research work well, since job-board search pages and many company/registry pages are JavaScript-rendered or gated. Without it you can still scrape single postings and run everything else.
• Job boards are market-specific. job-boards.md ships with working examples for two Norwegian boards (Finn.no, The Hub) so you can see the shape. Replace them with your own boards and write the matching extractor (each board structures its result cards differently).
• Company registries are jurisdiction-specific. /interview-prep references "the registry for the company's jurisdiction" (SEC EDGAR, Companies House, your national register, ...). Point it at yours.

---

Supported harnesses

Claude Code, Cursor, GitHub Copilot (VS Code), Gemini CLI, Codex CLI, OpenCode, Kiro, and the generic Agent Skills layout. Adding another is one config entry, see docs/DEVELOP.md.

---

Contributing

The single source of truth is source/skills/<name>/SKILL.md. A small, dependency-free Node build transforms it into per-harness output:

npm run build      # source/ -> all harness trees
npm run check      # validate manifests + skills
npm run rebuild    # clean + build

Edit source/ only — the per-harness trees are generated and gitignored, rebuilt automatically when the package is published and by the installer when missing. CI builds and validates them on every push. See docs/DEVELOP.md for the architecture and how to add a harness.

License

MIT, see LICENSE.