Job Application Automation

An end-to-end pipeline that scrapes job postings, tailors your resume for each one with an AI agent, and fills out the application form

scraper  →  CSV (status="new")
         →  tailoring agent  →  CSV (status="tailored")  +  tailored PDF resume
         →  submission agent →  CSV (status="applied")   +  form filled, waiting for manual submit

All state lives in a single CSV file (job_tracker.csv) — there's no database, no server, no accounts. Everything runs locally.

---

⚠️ Disclaimer

This project automates interactions with third-party sites (LinkedIn, Indeed, and various job boards/ATS platforms) via scraping and browser automation. That may be against the terms of service of those sites, and aggressive use can get your account flagged or rate-limited.

• Use this for your own job search, on your own accounts, at a reasonable pace. Don't run it as a mass-application spam tool.
• The submission step never clicks "Submit" and never logs in for you — you review and submit every application yourself.
• You are responsible for how you use this. The author provides no warranty and is not liable for account bans, rate limiting, or anything else that results from running this code.

---

How it's meant to be run

This repo is built around AI coding agents driving the pipeline, not just providing library code for you to script against.

• Primarily built and tested with Claude Cowork (the hosted "cowork" sessions referenced throughout this README and in CLAUDE.md). This is the supported, tested path.
• The scraper and render-cv.js are plain Python/Node scripts and should work fine from Claude Code, Codex, or any other coding agent / terminal — but the tailoring and submission prompts are written specifically for a Claude Cowork session (they assume tool access like mcp__claude-in-chrome__* for browser control), though they will probably also work on Codex or Claude Code.
• If you adapt them to another agent, expect to do some translation work.
• To run steps on a schedule from Claude Code, use the built-in /loop command, e.g. /loop 24h <prompt>.
• To run the scraper periodically you can cron it from your mac.

---

Repo layout

scraper/                  Python scraper (LinkedIn + Indeed -> job_tracker.csv)
  main.py
  CLAUDE.md / README.md

tailoring/                Resume tailoring (Claude Cowork prompts + renderer)
  COWORK_PROMPT.txt        Batch prompt: processes all status="new" rows
  TAILOR_PROMPT.txt        Single-job prompt: paste a URL or JD
  render-cv.js             resume.json -> HTML + PDF (via Playwright)
  Base-CV.html / .pdf      Reference copy of your base CV
  candidate_context.md     Background facts not in the CV (skills, adjacency, etc.)
  resume-template.json     Generic personal-data template

application/              Application submission (Claude Cowork prompt)
  SUBMISSION_PROMPT.md     Opens tabs, triggers autofill, uploads PDF, stops before submit

autofill-chrome-extension/  Manifest V3 extension that fills form fields
  content.js, bridge-main.js, manifest.json, popup.*
  candidate-data.json      Placeholder — replace with your real data
  selector-overrides.json  Per-hostname CSS selector overrides

setup/                    One-time setup
  SETUP_PROMPT.md          Guided setup prompt (run this first)
  linkedin-skills/         Helper script to pull your full LinkedIn skills list

private-files/            Gitignored — your real personal data lives here
package.json              Playwright dependency (repo root)
CLAUDE.md                 Full architecture reference for AI agents working in this repo

Runtime data (not committed to the repo) lives in ~/Documents/job-application-automation/:

job_tracker.csv           The central state file
autofill_issues.md         Log of fields the extension couldn't fill
resume-template.json       Your filled-in personal template
CVs/base/                   Your base CV (HTML)
CVs/tailored/{job_id}/      Tailored resume.json + HTML + PDF per job

---

Prerequisites

• Python 3.11 + python-jobspy + pandas (for the scraper)
• Node.js + npm (for render-cv.js / Playwright)
• Google Chrome, with the bundled autofill extension loaded as an unpacked extension
• A Claude Cowork session (claude.ai) with this repo's directory open, for the tailoring and submission steps

# Python deps
pip install python-jobspy pandas

# Node deps (run from repo root)
npm install playwright
npx playwright install chromium-headless-shell

---

Setup (do this first, once)

Run setup/SETUP_PROMPT.md as a prompt in Claude Cowork (or Claude Code/Codex — it's plain instructions, no Cowork-specific tools required). It will:

1. Ask for your personal info, professional profile, and CV content.
2. Create the runtime directories under ~/Documents/job-application-automation/.
3. Install Playwright and the headless Chromium build.
4. Generate your personal resume-template.json and candidate_context.md.
5. Personalize tailoring/TAILOR_PROMPT.txt and tailoring/COWORK_PROMPT.txt with your name, title, experience, and the repo's absolute path.
6. Tell you where to put your base CV (CVs/base/{name-slug}-CV.html).
7. Walk you through loading the autofill extension and filling in autofill-chrome-extension/candidate-data.json.
8. Run a final checklist and confirm everything is wired up.

Optionally, run setup/linkedin-skills/scrape-linkedin-skills.js to pull your full, per-role-attributed skills list from LinkedIn — useful for building an accurate candidate_context.md. See setup/linkedin-skills/README.md for details.

---

Running the pipeline

1. Scrape jobs

python3.11 scraper/main.py            # incremental: jobs from the last 24h
python3.11 scraper/main.py --full-sync  # full sync: jobs from the last 7 days

Appends new jobs to job_tracker.csv with status = "new". Config (search terms, title filters, blacklisted companies, allowed locations) lives at the top of scraper/main.py — see scraper/README.md for details.

Running it on a schedule: from Claude Code, use /loop, e.g.

/loop 6h Run python3.11 scraper/main.py and report how many new jobs were found.

Or, since the scraper is a plain script, schedule it directly with cron on your Mac (no Claude session needed):

0 */6 * * * cd ~/repos/job-application-automation && python3.11 scraper/main.py >> ~/Library/Logs/job-scraper.log 2>&1

2. Tailor your resume

In Claude Cowork, open this repo's directory and paste the contents of:

• tailoring/COWORK_PROMPT.txt — processes all status="new" rows (up to 15 at a time): reads your base CV + candidate_context.md, qualifies the role, writes a tailored resume.json, and renders it to PDF via render-cv.js.
• tailoring/TAILOR_PROMPT.txt — for a single job: paste a job URL or raw job description text.

Rows that pass become status = "tailored" with a PDF at CVs/tailored/{job_id}/. Underqualified or off-domain roles become status = "skipped".

You can also run this on a recurring schedule with /loop from Claude Code (e.g. once a day, after the scraper runs) — paste COWORK_PROMPT.txt as the loop prompt. Alternatively, set it up as a scheduled task in Claude Cowork to run COWORK_PROMPT.txt automatically on a recurring basis.

3. Fill out applications

In Claude Cowork, paste the contents of application/SUBMISSION_PROMPT.md. It will:

• Take up to 15 rows with status = "tailored".
• Open each job's application page in Chrome (in batches of 3).
• Trigger the autofill extension via DOM data attributes.
• Upload the tailored PDF resume.
• Fill any remaining fields the extension can't handle (years of experience, salary expectations, notice period, etc.).
• Stop at the submit button and alert you.
• Update the CSV to status = "applied" and log any recurring missed fields to autofill_issues.md.

Hard rules baked into this prompt: never click submit, never log in on your behalf, never write a cover letter.

Known limitation: the resume upload step currently does not work end-to-end via Cowork — mcp__claude-in-chrome__file_upload cannot attach the tailored PDF to the file input on most ATS pages. Until that's fixed, use the autofill Chrome extension (autofill-chrome-extension/) for field-filling: click the extension icon and press "Autofill This Page" to fill the visible form, then attach the resume PDF yourself from CVs/tailored/{job_id}/ and review before submitting.

4. Review and submit

Go through each opened tab, double-check the filled fields, and click Submit yourself.

---

The autofill extension

A Manifest V3 Chrome extension (autofill-chrome-extension/) that fills common application fields (name, email, phone, LinkedIn URL, location, work authorization, sponsorship). It does not run automatically — it's triggered either:

• By the submission agent, via data-* attributes on document.body, or
• Manually, via the extension popup's "Autofill This Page" button.

It does not fill years of experience, salary, notice period, willingness to relocate, or cover letters — those are handled by the submission prompt or left for you.

Install:

1. Open chrome://extensions, enable Developer Mode.
2. Click Load unpacked, select autofill-chrome-extension/.
3. Replace the placeholder values in autofill-chrome-extension/candidate-data.json with your real data (the setup prompt walks you through this).

Platform support, the trigger protocol, and selector-override format are documented in autofill-chrome-extension/README.md.

---

CSV status lifecycle

job_tracker.csv has a status column that drives the whole pipeline:

Status	Set by	Meaning
new	Scraper	Just found, not yet assessed
skipped	Tailoring	Underqualified or wrong domain
tailored	Tailoring	PDF ready, awaiting submission
tailor_error	Tailoring	PDF render failed
applied	Submission	Form filled, stopped before submit
submit_error	Submission	Login wall, upload failure, etc.

---

Privacy

Anything containing your personal data is gitignored:

• private-files/ — your real candidate-data.json, candidate_context.md, resume template, etc.
• ~/Documents/job-application-automation/ — runtime CSV, tailored CVs, resumes.
• autofill-chrome-extension/candidate-data.json and */selector-overrides.json overrides containing personal selectors.

Only placeholder/template versions of these files are committed.

---

More details

For the full architecture reference (used by AI agents working in this repo, but useful for humans too), see CLAUDE.md. Component-level docs:

• scraper/README.md
• autofill-chrome-extension/README.md
• setup/linkedin-skills/README.md

---

Planned updates

• Migrate state from CSV to SQL. job_tracker.csv currently has to be read/written in full by every agent step, which burns tokens as it grows. A small local database (e.g. SQLite) would let each step query/update only the rows it needs.
• Improve setup/SETUP_PROMPT.md. Make the guided setup more robust and reliable for new users/repos.
• Fix the submission resume-upload bug. Resolve the mcp__claude-in-chrome__file_upload issue described above so application/SUBMISSION_PROMPT.md can attach the tailored PDF end-to-end without falling back to the extension's manual button.