Career-Ops for beginners: Let an AI agent screen 700+ roles and generate 100+ tailored resumes

One-line summary: If you’re already using Claude Code to write code, why not use it to help you get hired? Career-Ops turns an AI programming CLI into a job-search command center — paste a JD URL, automatically evaluate it, generate ATS-friendly resumes, and track progress. The author personally screened 740+ offers, generated 100+ tailored resumes, and ultimately landed an offer as Head of Applied AI.

Project overview

Career-Ops (a.k.a. careerops) is an AI-driven job-hunting pipeline. GitHub: santifer/career-ops, MIT License. Its core idea is:

Turn any AI programming CLI into a complete job-search command center.

It’s not “yet another AI resume-rewriting tool,” and it’s not a bulk-application bot. Career-Ops is built around the philosophy of doing less: out of hundreds of offers, it helps you filter to the few that are genuinely worth applying to, using A-F scoring, 6 major evaluation dimensions, ATS-optimized PDF output, and fully automated tracking.

Core capabilities (15 slash commands):

Supported AI programming CLIs: Claude Code / OpenCode / Gemini CLI / Codex / Qwen / GitHub Copilot / Kimi — as long as it follows open agent skill standard.

Multilingual support: Default is English. Also available: German (modes/de/) / French (modes/fr/) / Japanese (modes/ja/) / Turkish (modes/tr/) / Simplified Chinese (README.cn.md) / Traditional Chinese (README.zh-TW.md).

Track record: The author santifer used it to evaluate 740+ offers, generate 100+ tailored resumes, and ultimately received an offer for Head of Applied AI.

Difficulty / time / outcomes

• Difficulty: ⭐⭐ (moderate; you need to be comfortable using an AI programming CLI like Claude Code)
• Time: 30–60 minutes for first onboarding (including cv.md prep + profile configuration)
• Outcomes:
  • Get a real, end-to-end AI job-search pipeline working
  • Understand the agent architecture design: “User Layer vs System Layer”
  • Learn to do precise job hunting using A-F scoring + STAR+R stories
  • See how to assemble an open-source AI agent project that’s been covered by WIRED / Business Insider
  • Bonus: pick up 15 slash commands and use them every day during your job search

Target audience

• Developers who are currently job hunting / switching jobs / exploring opportunities
• Already using Claude Code / OpenCode / Gemini CLI
• Want to switch from “spray-and-pray” to “apply precisely,” and don’t want to waste time on junk offers
• Want to see what a production-grade AI agent project with 7k+ stars looks like

# Assuming you use Claude Code, edit in ~/.claude/settings.json or in env
ANTHROPIC_BASE_URL=https://api.defapi.org
ANTHROPIC_API_KEY=sk-xxxxx   # Get it from the defapi.org dashboard

Core dependencies and environment

Minimum requirements:

Optional / enhancements:

Complete project structure tree

career-ops/
├── AGENTS.md                    # Agent common instructions (across CLIs)
├── CLAUDE.md                    # Claude Code specific (imports AGENTS.md)
├── GEMINI.md                    # Gemini CLI specific
├── cv.md                        # Your resume (created by you; key file)
├── article-digest.md            # Your proof of highlights (optional)
├── config/
│   └── profile.example.yml      # Your personal profile template
├── modes/                       # 15 skill modes
│   ├── _shared.md               # Shared system-layer context
│   ├── _profile.template.md     # User-layer customization (put your personal archetype here)
│   ├── oferta.md                # Single-offer evaluation
│   ├── pdf.md                   # PDF generation
│   ├── scan.md                  # Role scanning
│   ├── batch.md                 # Batch processing
│   ├── auto-pipeline.md         # URL → evaluation + PDF + tracker
│   ├── de/                      # German mode
│   ├── fr/                      # French mode
│   ├── ja/                      # Japanese mode
│   └── tr/                      # Turkish mode
├── templates/
│   ├── cv-template.html         # ATS-optimized CV template
│   ├── cv-template.tex          # LaTeX template
│   ├── portals.example.yml      # 45+ preconfigured companies
│   └── states.yml               # Application status dictionary
├── data/                        # Application data (gitignored)
│   ├── applications.md          # Main tracking table
│   ├── pipeline.md              # Inbox for pending URLs
│   ├── scan-history.tsv         # Scan de-duplication
│   └── follow-ups.md            # Follow-up history
├── reports/                     # Evaluation reports (gitignored)
├── output/                      # Generated PDFs (gitignored)
├── jds/                         # JD text stored here (optional)
├── interview-prep/              # Interview preparation
│   ├── story-bank.md            # Accumulated STAR+R stories
│   └── {company}-{role}.md      # Company-specific intel
├── batch/
│   ├── batch-prompt.md          # Worker prompt
│   ├── batch-runner.sh          # Orchestration script
│   └── tracker-additions/       # Items to merge into the main tracking table
├── dashboard/                   # Go TUI (Bubble Tea)
│   ├── main.go
│   └── internal/
├── fonts/                       # Space Grotesk + DM Sans
├── providers/                   # LLM provider configuration
├── docs/                        # Multilingual docs + media coverage
├── examples/                    # Sample CV / reports / highlights
└── *.mjs                        # Scripts for scanning / verification / merging / health checks

As a user, you only need to touch four files: cv.md / config/profile.yml / modes/_profile.md / portals.yml — that’s your User Layer. System updates won’t change them. Everything else is handled by the agent.

Step-by-step instructions

Step 1: Clone + install

git clone https://github.com/santifer/career-ops.git
cd career-ops
npm install
npx playwright install chromium   # needed for PDF generation

Step 2: Run health check

npm run doctor

It will verify the Node.js version, Playwright, key files, and directory permissions. If it fails, the “doctor” will tell you what’s missing — fix it, then run again.

Step 3: Create your cv.md

This is the source of the entire pipeline — without it, the agent can’t evaluate offers.

Create cv.md in the project root:

# Your Name

**Email:** [email protected]
**Location:** Berlin, Germany (CET)
**LinkedIn:** linkedin.com/in/you
**GitHub:** github.com/you

## Summary

Senior backend engineer with 6 years building distributed systems...
(Target: 2–3 paragraphs; write what you do + your strongest one or two things)

## Experience

### Staff Engineer — Company A (2023 - Present)
- Led migration of X to Y, reducing Z by 40%
- (Try to include numbers; this section will be used by the agent to match the JD)

### Senior Engineer — Company B (2020 - 2023)
- ...

## Education
...

## Skills
Backend: Go, Python, PostgreSQL, Kafka
AI/ML: PyTorch, LangChain, RAG
...

Step 4: Configure config/profile.yml

cp config/profile.example.yml config/profile.yml

Open config/profile.yml and fill in:

personal:
  name: "Your Name"
  email: "[email protected]"
  location: "Berlin, Germany"
  timezone: "CET"

targeting:
  roles:
    - "Senior Backend Engineer"
    - "Staff Engineer"
    - "AI Platform Engineer"
  comp_target:
    base_min: 130000
    base_target: 160000
    currency: "EUR"
  remote_preference: "remote-first"
  deal_breakers:
    - "no on-call"
    - "no Java"
    - "no < 20 person startups"

# If you want to use Chinese / German / French / Japanese / Turkish modes, add this line
# language:
#   modes_dir: modes/de

Step 5: Configure portals.yml

cp templates/portals.example.yml portals.yml

By default it includes 45+ companies and 19 search queries, covering:

# AI Labs
- anthropic
- openai
- mistral
- cohere

# Voice AI
- elevenlabs
- vapi
- deepgram

# AI Platforms
- retool
- airtable
- vercel
- temporal

If you want to add companies you’re specifically targeting, add them to portals.yml. Job board coverage includes Ashby / Greenhouse / Lever / Wellfound / Workable / RemoteFront.

Step 6: Have the agent “Onboard” you

# Choose one of the AI CLIs you installed
claude       # Claude Code
# or
gemini       # Gemini CLI
# or
opencode     # OpenCode
# or
codex        # Codex

In the CLI, say:

Help me set up career-ops. I just cloned it — cv.md and config/profile.yml are filled in.
Please check whether my onboarding is complete, and what else I should add.

The agent will run a boot check:

1. Is cv.md there? ✅
2. Is config/profile.yml real (not .example)? ✅
3. Is modes/_profile.md copied from _profile.template.md? ⚠️
4. Is portals.yml real (not .example)? ✅

Any ❌ will put you into guided mode, prompting you step-by-step, one-on-one.

Step 7: First evaluation

Paste a JD URL to the agent:

/career-ops https://jobs.anthropic.com/staff-software-engineer-applied-ai

The agent runs a fully automated pipeline:

1. Archetype detection — categorize into LLMOps / Agentic / PM / SA / FDE / Transformation
2. A-F evaluation — 6 dimensions (Role summary / CV match / Level strategy / Comp research / Personalization / Interview prep)
3. Generate report — write to reports/001-anthropic-2026-06-09.md
4. Generate PDF — run Playwright rendering for an ATS resume; output to output/001-anthropic-cv.pdf
5. Write to tracker — add a line to data/applications.md
6. Update pipeline integrity — run node verify-pipeline.mjs to check

You should see, after 30–60 seconds, a complete evaluation report plus an ATS resume PDF.

Step 8: First scan

/career-ops scan

Or equivalently:

node scan.mjs                  # zero tokens; pure API fetch
node scan.mjs --verify         # API + Playwright second verification; filters expired roles

With 45+ preconfigured companies and 19 search queries, one scan typically yields 200–500 new roles, directly adding them to the inbox data/pipeline.md. Then claude -p workers run evaluations concurrently — processing 100+ offers in one night is not a dream.

Step 9: Run the TUI dashboard (optional)

If you want to view progress in a terminal UI:

cd dashboard
go build -o career-dashboard .
./career-dashboard --path ..

It has 6 filter tabs, 4 sorting modes, grouped/flat views, lazy-loaded previews, and selecting an item changes its status. Catppuccin Mocha theme — it looks great.

Step 10: Update the system (important!)

Career-Ops iterates fast; new modes and company configs come out every few weeks.

# In the project root, ask the agent to check
claude  # tell it: "check whether there are new career-ops versions"

The agent will automatically run node update-system.mjs check and output JSON like this:

{"status": "update-available", "local": "1.3.0", "remote": "1.4.0", "changelog": "..."}

It will tell you: “v1.3.0 → v1.4.0 is available. Your data (CV / profile / tracker / reports) won’t be touched. Do you want to update?” To update:

node update-system.mjs apply       # update
node update-system.mjs rollback    # roll back if you’re unhappy

Common troubleshooting

Q1: cv.md not found / agent stuck on onboarding

The most common reason is that cv.md isn’t placed in the project root:

# confirm it’s in the root
ls -la cv.md
# you should see something like -rw-r--r-- cv.md

# if it’s missing, create it (in the project root)
cd career-ops
touch cv.md
# then fill in the content

Q2: PDF generation fails: "browser not found"

Playwright isn’t installed, or Chromium wasn’t installed:

npx playwright install chromium
# if you’re in China and the network is slow, you can add a mirror
PLAYWRIGHT_DOWNLOAD_HOST=https://npmmirror.com/mirrors/playwright npx playwright install chromium

Q3: Evaluation scores are too high / too low

This happens in the first week. Fix it by:

1. Run 3–5 more evaluations; each time tell the agent “I don’t agree with this score because XX.”
2. The agent will write what it learns into modes/_profile.md.
3. Pay special attention to what’s been filled in modes/_profile.md: your compensation floor, remote preference, role positioning, and reverse cases.

Important: Put personalized content into modes/_profile.md or config/profile.yml. Don’t edit modes/_shared.md — system updates will overwrite the latter.

Q4: WebSearch / WebFetch verification accuracy

Not accurate. Career-Ops has an internal rule: you must verify offers with Playwright to confirm they’re still active:

# use slash command; internally it runs Playwright
/career-ops verify --url <JD_URL>

In batch mode (claude -p), Playwright isn’t available, so it will be marked **Verification:** unconfirmed (batch mode). You’ll need to manually review.

Q5: Batch evaluation token usage explodes

If you evaluate 100+ offers, token bills can easily exceed $20 in one night. There are three ways to fix it:

1. Reduce the batch size — run 10–20 at a time; don’t scan 100+ in one go.
2. Local model fallback — structured tasks like ofertas (compare) / tracker / pdf can be swapped to local small models.
3. Get a half-price API by using Defapi — protocol-compatible; it uses the same model as the official side, so your monthly bill is cut dramatically in half (see the TIP after the “Target audience” section for the specific configuration).

Q6: Switching to German / French / Japanese / Turkish modes

# 1. Edit config/profile.yml
language:
  modes_dir: modes/de   # or fr / ja / tr

Or simply tell the agent: “Run in German mode.” The agent will switch to the evaluation logic under modes/de/ (including localized DACH-market terms like 13. Monatsgehalt / Probezeit / AGG / Tarifvertrag).

Q7: Tracker status fields contain dirty formatting like **Applied**

Tracker statuses were bolded manually. Fix it:

node normalize-statuses.mjs     # normalize statuses
node dedup-tracker.mjs          # de-duplicate
node verify-pipeline.mjs        # health check

After that, edit templates/states.yml to add canonical statuses.

Further reading / advanced directions

• User Layer vs System Layer: This is the most interesting engineering design in Career-Ops — modes/_profile.md (user) is never overwritten by modes/_shared.md (system). All your personalization from onboarding will survive 100 system updates. Recommended: read DATA_CONTRACT.md for the full picture.
• Change archetypes: Tell the agent “change the archetypes to data engineering roles.” It will modify modes/_profile.md, not modes/_shared.md.
• Multilingual switching: When applying to DE/FR/JP/TR roles, switch to modes/{de,fr,ja,tr}/; evaluation logic will be localized automatically (local terms like CDI / 賞与 / commuting allowance).
• TUI dashboard advanced: The dashboard is written in Go + Bubble Tea + Lipgloss, with a Catppuccin Mocha color scheme. Source is in dashboard/internal/. If you want custom filters/sorting, just edit it.
• STAR+R story bank: interview-prep/story-bank.md accumulates your “Situation + Task + Action + Result + Reflection” stories across evaluations. After 3–5 evaluations, you’ll have 5–10 stories that can handle almost any behavioral interview question.
• Fork the portfolio companion: The author’s portfolio is also open source — santifer/cv-santiago — with an AI chat bot + an LLMOps board + case studies. Fork it and customize it as your own.

GitHub repo: santifer/career-ops. Multilingual docs are in README.{cn,es,ja,ko-KR,pt-BR,ru,ua,zh-TW}.md. The design philosophy is explained in DATA_CONTRACT.md. Media coverage is in docs/press/. The roadmap is in ROADMAP.md (if you can’t find it in the repo, ask the agent: “show me your roadmap”).