jd scoring · tailored cv pdfs · portal scans · markdown tracker · six standalone clis · web + terminal dashboards

paste a JD into Claude Code, get a scored evaluation, a tailored CV PDF, and a tracker entry

A fork of santifer/career-ops with a Next.js landing page and coverage extensions for non-AI job categories.

Quick start | Features | Use | Dashboards | Fork map | Credits

🏗️ Clone it, then type /career-ops in Claude Code

Runs from your terminal against files you own. No server, no account, no database.

❤️ Sponsor this project ❤️

---

The /admin pipeline view running on sample data.

What it does

Paste a job description and get a scored fit evaluation against your CV, plus a tailored ATS-clean PDF ready to submit. It scans your configured job boards for new postings and tracks every application in a plain markdown pipeline. The whole loop runs from your terminal against files you own.

flowchart LR
    A[Job description] --> B[Scored evaluation]
    B --> C[Tailored CV PDF]
    C --> D[Tracker entry]
    D --> E[Dashboards]

Loading

✨ Features

• 📋 JD in, decision out: the /career-ops skill evaluates a pasted job description against your cv.md and profile, writes a scored report, and appends the tracker.
• 📄 ATS-clean PDF: generate-pdf.mjs renders a tailored CV variant per JD; verify-pipeline.mjs checks the loop end to end.
• 🗂️ Portal scans by name: configured queries for Ashby, Greenhouse, and Lever boards ship in templates/portals.example.yml, with non-AI board coverage in templates/portals.extensions.yml.
• 🧰 Six standalone CLIs: rehearse, tailor, negotiate, outreach, timeline, and learn-rejection each run as a plain npm script, no Claude session needed.
• 🖥️ Two dashboards, one data model: a web /admin surface and a Go TUI, both reading the same markdown tracker, both falling back to clearly labeled sample data when no tracker exists.
• 🎨 One token file: the web layer styles from tokens/tokens.css and the TUI's wranngle theme maps the same palette.
• 🧪 Tested: 126 unit tests and 82 integrity checks pass on npm test.

🚀 Quick start

1. Clone and install

   git clone https://github.com/wranngle/career_architect.git
   cd career_architect
   
   # Prereqs: Node >=20.19, Go >=1.24.2, Python >=3.11
   npm install
   npx playwright install chromium
   pip install -r requirements.txt

2. Configure your profile

   cp config/profile.example.yml config/profile.yml
   cp modes/_profile.template.md modes/_profile.md
   cp templates/portals.example.yml portals.yml
   # Optional: append sections from templates/portals.extensions.yml
   #           into portals.yml for non-AI board coverage.
   # Create cv.md with your master resume content.

3. Validate the environment

   npm run doctor

4. Open Claude Code in the repo and type /career-ops.

Personal data can also live in a separate directory instead of the repo: run any script from that directory and it resolves user-layer paths (cv.md, config/, data/, reports/, output/) against your CWD. See "Split-repo layout" in DATA_CONTRACT.md.

🧰 Use

Get the job-search pipeline from JD intake to evaluation, PDF generation, tracker status, portal scans, follow-up checks, recruiter practice, CV tailoring, negotiation drafts, outreach messages, timelines, and rejection lessons.

Inside Claude Code

/career-ops                       # show all subcommands
/career-ops <JD text or URL>      # auto-pipeline: evaluate -> PDF -> tracker
/career-ops scan                  # scan all enabled portals
/career-ops pdf                   # tailor cv.md for one JD, render PDF
/career-ops tracker               # pipeline status
/career-ops followup              # flag overdue follow-ups

Full list: .claude/skills/career-ops/SKILL.md.

Standalone npm scripts

Six of the steps also run standalone, no Claude session needed. With npm scripts, CLI args go after --, e.g. npm run tailor -- <jd.md>.

Worked example: tailor a CV against the sample JD

npm run tailor -- fixtures/jd-sample.md

Every CLI runs the same way against the sample files in fixtures/, so you can try the whole toolbelt before touching your own data:

• npm run rehearse -> node bin/rehearse.mjs: 5-turn mock recruiter call. Usage: npm run rehearse -- --company <slug> --mock <fixture.json> [--turns 5] [--root <dir>]. The --mock flag is currently required (live recruiter client not wired).
• npm run tailor -> node bin/tailor.mjs: per-JD CV variant. Usage: npm run tailor -- <jd.md> [--root <dir>] [--cv <cv.md>] [--out-dir <dir>].
• npm run negotiate -> node bin/negotiate.mjs: offer negotiation script generator. Usage: npm run negotiate -- <offer.json> [--root <dir>] [--out-dir <dir>]. Offer JSON must include compensation.base_salary_usd.
• npm run outreach -> node bin/outreach.mjs: cold-message generator. Usage: npm run outreach -- <person.json> <jd.json> [--write] [--root <dir>] [--out-dir <dir>]. Prints to stdout; --write also writes a file.
• npm run timeline -> node bin/timeline.mjs: application calendar built from data/applications.md. Usage: npm run timeline -- [--root <dir>] [--out <path>] [--today YYYY-MM-DD] [--include ...] [--stdout]. Writes out/timeline.md by default.
• npm run learn-rejection -> node bin/learn-rejection.mjs: rejection-feedback learner; extracts lessons from a rejection email into data/lessons.md. Usage: npm run learn-rejection -- <rejection.md> [--root <dir>] [--lessons <path>] [--today <YYYY-MM-DD>] [--stdout].

📊 Dashboards

Two surfaces, one data model. Both read local career-ops files when they exist and fall back to clearly labeled sample data when the tracker is not initialized, so the first run always renders something honest.

Web (/admin)

npm run dev
# then open http://localhost:3000/admin

Pipeline, Progress, and Scans screens plus a Profile readiness view, with a data banner on every page stating whether you are looking at live or demo rows.

Scans screen, demo mode, banner on.

Terminal (Go TUI)

cd dashboard && go build -o ../career-dashboard .
../career-dashboard --path .. --theme=wranngle   # or catppuccin-latte / catppuccin-mocha / auto

See dashboard/README.md for the full theme and flag reference.

🧭 One pipeline, four surfaces

Claude Code the /career-ops skill drives the whole loop conversationally

Six CLIs rehearse, tailor, negotiate, outreach, timeline, learn-rejection

Web /admin pipeline, progress, and scans in the browser

Go TUI the same tracker, terminal-native

Every surface reads and writes the same plain markdown files, so nothing is locked in.

🔀 Why a fork

Upstream Career-Ops is tuned for senior AI/ML engineers searching Greenhouse, Ashby, and Lever. This fork adds non-AI portal coverage (templates/portals.extensions.yml), the Next.js landing page and /admin dashboard, and keeps upstream's architecture and license. UPSTREAM.md maps what is upstream and what is local.

🗺️ Roadmap

Item	State
CI gate running npm test on every push	next
examples/ fixtures feeding the /admin demo rows directly	next
Live recruiter client for rehearse (mock fixtures drive it today)	next

Non-goals

Same as upstream: no database, no auto-submit, no background/job queues, no vector DB. Claude evaluates and tailors; you submit via Simplify.jobs or any other manual path.

License

MIT, matching upstream. See LICENSE.

Credits

All non-trivial design credit belongs to Santiago Fernández de Valderrama (@santifer). This fork layers coverage extensions and a landing page on top. Fork-specific bug reports and feature requests go to this repo's issues; upstream questions route per SUPPORT.md.