convez.ai
AgenteAutomaçãoAvançado
R1chi33333

Job Hunter — Automação de Candidaturas com Multi-Agent

R1chi33333/job-hunter

5 agentes de IA automatizam busca, scoring, escrita e envio de candidaturas

Ver no GitHub
InicialPluginScore · de 0 a 100

Sistema de 5 agentes que automatiza buscas de emprego, filtragem, escrita de cartas de apresentação personalizadas, revisão de qualidade e submissão de candidaturas. Funciona inteiramente no VS Code via Claude Code, sem necessidade de API key, integrando Playwright para scraping em Seek/Indeed e geração de rascunhos por email.

Instalação

$npx convez-ai R1chi33333/job-hunter

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Automação completa do pipeline de candidaturas a múltiplas vagas
  • Scraping inteligente de vagas em Seek, Indeed e career pages
  • Geração e revisão automática de cartas de apresentação personalizadas
  • Submissão direta de candidaturas ou via rascunhos de email

Veredito

Projeto ambicioso e bem arquitetado para quem busca automação completa de candidaturas. Requer Python avançado, Claude Code CLI e Chrome local; ideal para desenvolvedores no macOS dispostos a customizar o pipeline completo.

README

A tradução para o português deste README ainda está na fila.

Job Hunter — Multi-Agent Job Application Automation

A 5-agent automated job hunting system that runs entirely inside VS Code via Claude Code (no API key needed). Scrapes Auckland job listings, scores fit, writes tailored cover letters, QC-reviews them, and submits applications — either directly on Seek/Indeed via Playwright or via email draft.

Scout → Filter → Writer → Reviewer → [Human approval] → Sender

Architecture

┌─────────────────────────────────────────────────────────┐
│                     ORCHESTRATOR                        │
│                (main.py — runs the pipeline)            │
└───┬──────────┬──────────┬────────────────┬──────────────┘
    │          │          │                │
[1]Scout  [2]Filter  [3]Writer ──→ [4]Reviewer ──→ [5]Sender
  Scrape    Score      Write          QC + rewrite    Submit
  listings  1-10       cover          (max 2x)       Seek or
  Seek/     via        letters        NEEDS_HUMAN_    email
  Indeed/   claude     via claude     REVIEW flag
  direct    -p CLI     -p CLI

Agent Details

AgentFileWhat it does
Scoutagents/scout.pyPlaywright scraper — Seek (with saved session), Indeed, company career pages. Listing-only (no detail page visits) for speed. Full pagination. Deduplication across runs.
Filteragents/filter.pyScores each job 1-10 via claude -p. Caches scores in output/seen_scores.json. Saves markdown report to output/filter_report.md.
Writeragents/writer.pyWrites 4-paragraph cover letter via claude -p for every job that scored ≥ min_score. Fetches full JD lazily before writing.
Revieweragents/reviewer.pyQC-checks 8 criteria (hook, company-specific, metrics, portfolio, length, no-AI-cliches, salutation, JD-alignment). Rewrites up to 2x, then flags NEEDS_HUMAN_REVIEW.
Senderagents/sender.pyMode A: Playwright submits directly on Seek/Indeed. Mode B: Generates email draft + Claude Code Gmail command for company-direct roles.

Requirements

  • macOS (uses channel="chrome" — real Chrome must be installed)
  • Claude Code — claude CLI must be on your PATH and logged in
  • Python 3.11+
  • GitHub CLI (gh) — optional, only needed for setup

Setup

1. Clone and create venv

git clone https://github.com/R1chi33333/job-hunter.git
cd job-hunter
python3 -m venv venv
source venv/bin/activate

2. Install dependencies

pip install -r requirements.txt
playwright install chromium

3. Configure your profile

cp config/profile.example.json config/profile.json

Edit config/profile.json with your real name, email, phone, portfolio URL, and email signature.

4. Configure targets

Edit config/targets.json to set your search keywords, target companies, and minimum fit score.

5. Add your resume

# Place your resume PDF here (used by Sender for file uploads):
cp /path/to/YourResume.pdf config/YourResume.pdf

# Place your resume as plain text here (used by Filter/Writer for context):
# Create or copy: config/resume.md

6. Log in to Seek (one-time)

Seek uses Google OAuth, so a password-based login is not possible. This script opens a real Chrome window for you to log in manually, then saves the session cookie:

python3 seek_login.py

After logging in and seeing the Seek home page, press Enter in the terminal. The session is saved to config/seek_session.json (gitignored).


Usage

Full pipeline (Scout → Filter → Writer → Reviewer)

source venv/bin/activate
python3 main.py

Options

python3 main.py --dry-run       # Scout only — no scoring, no letter writing
python3 main.py --skip-scout    # Use last Scout output, re-run Filter→Writer→Reviewer
python3 main.py --single-job "https://seek.co.nz/job/..."  # Test one job end-to-end

Review output

After the pipeline finishes:

  • Filter report → output/filter_report.md (scored table, passed/failed)
  • Cover letters → output/letters/*_final.md (one per passed job)
  • Job data → output/jobs/*.json

Read the cover letters, then approve the ones you want to send.

Send applications

# Submit a specific job
python3 send.py --job "rocketwerkz-unity-developer"

# Submit all jobs marked as approved
python3 send.py --all-approved

# Test Seek submission flow (no actual submit)
python3 send.py --test-seek

# List all jobs ready to send
python3 send.py --list

How Claude Code CLI is used (no API key needed)

All AI calls use claude -p "<prompt>" subprocess calls via agents/_claude.py. This means:

  • No ANTHROPIC_API_KEY needed — uses your existing Claude Code subscription
  • Claude Code must be installed and authenticated (claude --version should work)
  • If you hit your usage limit, the system gracefully returns score 0 and logs the error — just re-run after the limit resets

Output directory structure

output/
├── filter_report.md          # Scored table — all jobs, pass/fail
├── seen_jobs.json            # Dedup index (never re-scrape same job)
├── seen_scores.json          # Score cache (never re-score same job)
├── jobs/
│   └── 2026-06-09_scout.json
├── letters/
│   ├── rocketwerkz-unity-dev_draft.md
│   ├── rocketwerkz-unity-dev_final.md
│   └── rocketwerkz-unity-dev_email_draft.txt
└── receipts/
    └── rocketwerkz-unity-dev.json   # Submission receipt + screenshot

Config files

FileStatusPurpose
config/profile.jsongitignoredYour name, email, phone — copy from profile.example.json
config/resume.mdgitignoredResume plain text for AI context
config/*.pdfgitignoredResume PDF for file upload
config/seek_session.jsongitignoredSeek browser session cookie
config/targets.jsoncommittedKeywords, target companies, platforms
config/profile.example.jsoncommittedTemplate — safe placeholder values only

Seek session notes

  • Session cookies expire periodically. If Sender detects a login redirect, it prints: ✗ Seek session expired. Run: python3 seek_login.py
  • Re-run seek_login.py to refresh the session.

License

MIT

Ver o original no GitHub

Bom para

Relacionados