Job Hunter Skill
lomartins/job-hunter-skill
Central de busca de emprego self-hosted com scraping, rastreamento e automação ATS
Ver no GitHubSkill Python local-first para automação completa de busca de empregos: scraping de múltiplas job boards, rastreamento kanban de candidaturas em SQLite, preenchimento inteligente de formulários ATS e adaptação automática de currículo por JD. Mantém dados sensíveis (PII) 100% locais, sem enviar informações para a IA.
Instalação
npx convez-ai lomartins/job-hunter-skill1 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
Casos de uso
- Automatizar coleta de vagas de múltiplas job boards (LinkedIn, Indeed, Glassdoor, Gupy, etc)
- Rastrear pipeline completo de candidaturas com kanban visual e métricas
- Preencher formulários ATS automaticamente ou assistido sem enviar dados sensíveis
- Adaptar currículo e carta de apresentação por descrição de vaga
Veredito

Skill robusto e pragmático para quem faz busca de emprego ativa. A arquitetura local-first com SQLite e webapp FastAPI + HTMX é bem pensada; suporte a 11+ fontes de vagas e integração com Claude Code o tornam eficiente. Recomendado para perfis técnicos que precisam escalar candidaturas sem comprometer privacidade.
README
A tradução para o português deste README ainda está na fila.
job-hunter
A self-hosted, PII-safe job-search command center — for Claude Code, Codex, Cursor, and the terminal.
Job hunting buried under a hundred tabs? job-hunter scrapes job boards, tracks every application kanban-style in a local SQLite DB, fills ATS forms in shadow or auto mode, surfaces salary distributions from your own pipeline, and tailors a Reactive Resume per JD — all without ever sending your phone, ID, or session cookies to a model.
Works for any tech role — backend, frontend, mobile, data, ML, infra, devrel, security, design-eng. Drop your role keywords in profile.yaml once and the same skill works for a senior platform engineer, a junior data analyst, or a staff designer.
Webapp
Launched with job-hunter web (localhost-only, no auth):
What you get
- 11 sources out of the box — LinkedIn, Indeed, Glassdoor, Gupy, RemoteOK, Job na Gringa, Remotive, We Work Remotely, Himalayas, Programathor, Coodesh. New scrapers via a
learn.pyadapter that drafts YAML from any unknown form. - Full pipeline tracking —
discovered → queued → applying → applied → screening → technical → behavioral → offer / rejected / withdrawn, with note-bearing transitions, flagging, and structured tags. - Local FastAPI + HTMX webapp at
127.0.0.1:8765— kanban tracker with drag-and-drop stage moves, filter by tag/source/stage/flag, sort by match/date/salary, currency conversion (BRL/USD/EUR via free ECB rates), markdown-rendered JDs, daily/weekly application charts. - Match score per job derived from your profile + JD heuristics (seniority, country lock, on-site mismatch, US-clearance, etc.).
- Salary aggregator — p25/median/p75 from your own scraped data, no live Glassdoor scraping needed.
- AI form-fill in two modes —
shadowpauses before submit,autorequires 5 reliability gates and explicit consent. - Resume tailoring —
/job-hunter:tailor-resume <id>emits Reactive Resume JSON tuned to the JD. - Claude Code slash commands —
discover,apply,dig,tailor-resume,validate,web,tracker,status,review,doctor.
PII-safe by design
Your personal data (national ID, phone, address, salary expectations, session cookies) lives in ~/.config/job-hunter/secrets/personal.env — chmod 600, and the model is explicitly forbidden to read it via Read tool, cat/head/tail/grep, or any pipe whose output enters the conversation. The CLI loads it via python-dotenv in a child process; the model only ever sees field schemas. CI enforces this: lint_secret_leaks.py blocks any PR that would echo PII into logs, screenshots, or artifacts.
Install
Recommended: Claude Code plugin marketplace
/plugin marketplace add lomartins/job-hunter-skill
/plugin install job-hunter@lomartins-skills
Manual (Claude Code)
git clone git@github.com:lomartins/job-hunter-skill.git ~/.claude/skills/lomartins-job-hunter-skill
ln -s ~/.claude/skills/lomartins-job-hunter-skill/skills/job-hunter ~/.claude/skills/job-hunter
Codex CLI / Cursor / Gemini CLI
job-hunter follows the open Agent Skills standard. Clone into the host's skill dir:
git clone git@github.com:lomartins/job-hunter-skill.git ~/.codex/skills/job-hunter-skill
# Cursor: ~/.cursor/skills/
# Gemini: ~/.gemini/skills/
Install the CLI binary
The skill ships a Python CLI that does the actual work. Install once:
uv tool install --from git+https://github.com/lomartins/job-hunter-skill.git job-hunter
playwright install chromium
This puts job-hunter and the shorter alias job on your PATH.
60-second quickstart
# 1. Initialize XDG dirs and copy templates
job init
# 2. Edit your profile — role keywords drive matching (no PII goes here)
$EDITOR ~/.config/job-hunter/profile.yaml
# roles: [backend engineer, golang, distributed systems]
# roles: [senior frontend, react, typescript]
# roles: [data engineer, dbt, snowflake]
# roles: [staff android, kotlin, kmp] # whatever fits you
# 3. Add your secrets (kept out of model context)
$EDITOR ~/.config/job-hunter/secrets/personal.env
chmod 600 ~/.config/job-hunter/secrets/personal.env
# 4. Discover roles from RemoteOK (no auth needed)
job discover --source remoteok
# 5. See what landed
cat ~/.local/share/job-hunter/tracking.md # human-readable mirror
job list --stage discovered # CLI view
job web # or: visual triage in a browser
# 6. Queue something and apply in shadow mode
job queue 7
job apply 7 --mode shadow
The webapp listens on http://127.0.0.1:8765 by default (localhost-only, no auth, no PII exposed). Pass --port to override.
Architecture
flowchart LR
subgraph Sources
L[LinkedIn]
G[Gupy]
R[RemoteOK]
J[Job na Gringa]
E[7 more...]
end
subgraph Skill
D[(SQLite jobs.db)]
MD[tracking.md]
AD[Adapters YAML]
AP[apply.py]
LE[learn.py]
end
subgraph Forms
F1[Gupy form]
F2[Greenhouse form]
F3[Workday form]
end
subgraph PII[chmod 600 — never read by model]
S[personal.env]
end
L & G & R & J & E -->|scrape| D
D --> MD
D --> AP
AD --> AP
AP -->|fill| F1 & F2 & F3
AP -.->|secret.* lookups at runtime| S
F1 & F2 -.unknown form.-> LE
LE -->|draft adapter| AD
Source compatibility
| Source | Phase | Method | Auth | Firecrawl boost? |
|---|---|---|---|---|
| Job na Gringa | 1 | HTML | — | No |
| 1 | HTML + cookie | LINKEDIN_LI_AT | No (cookie auth) | |
| Gupy | 1 | HTML | — | No |
| RemoteOK | 1 | JSON API | — | No |
| Indeed | 0.9 | HTML, captcha-aware | — | Yes |
| Glassdoor | 0.9 | HTML + cookie or Firecrawl | GLASSDOOR_GD_ID + _UAC | Yes |
| Remotive | stub | JSON API | — | — |
| We Work Remotely | stub | RSS | — | — |
| Himalayas | stub | HTML/GraphQL | — | — |
| Programathor | stub | HTML | — | — |
| Coodesh | stub | HTML | Optional | — |
| Trampos.co | stub | HTML | — | — |
| Arc.dev | stub | HTML | Required | — |
Set FIRECRAWL_ENDPOINT=http://localhost:3002 in personal.env to route Indeed + Glassdoor through a self-hosted Firecrawl instance (handles JS rendering + anti-bot). Apply-path is hard-disabled when Firecrawl is set. See references/firecrawl.md.
Salary distribution from your own pipeline
After running discover across a few sources, get the role's salary distribution from the data you've actually seen — no live Glassdoor scraping needed:
job-hunter discover --source indeed
job-hunter discover --source remoteok
job-hunter salary --role "backend engineer" --location "united states"
job-hunter salary --role "kotlin" --location brazil
job-hunter salary --role "data engineer" --location remote
Outputs p25 / median / p75 per currency + a suggested expectation (p75 + 10% padding). The --role matches against title substrings, so use whichever keyword density matches your search.
Adapter platform support
| Platform | Adapter shipped | Auto-eligible default |
|---|---|---|
| Gupy | ✓ | ✗ |
| Greenhouse | ✓ | ✗ |
| Lever | ✓ | ✗ |
| Workday | ✓ | ✗ |
| Ashby | ✓ | ✗ |
| SmartRecruiters | (learned via learn.py) | — |
| Recruitee | (learned) | — |
| BambooHR | (learned) | — |
| Personio | (learned) | — |
| Jobvite | (learned) | — |
Auto-eligibility flips on per-signature after three consecutive successful shadow submits and an explicit job adapter mark-auto-eligible <sig>.
Slash commands (Claude Code plugin install)
| Command | What it does |
|---|---|
/job-hunter:discover [source] | Pull new postings (default: remoteok). |
/job-hunter:list [filters] | Pipeline table. |
/job-hunter:track <url> | Track a specific posting + queue it. |
/job-hunter:apply <id> | Preview the form-fill plan (dry-run). |
/job-hunter:dig <id> | Re-fetch JD + brief: fit, friction, tailoring angles, next move. |
/job-hunter:tailor-resume <id> | Generate a Reactive Resume JSON tailored to the JD. |
/job-hunter:validate [id] | Pre-flag obvious non-fits (junior, country-locked, on-site wrong country). |
/job-hunter:web | Launch the local triage webapp. |
/job-hunter:status | Pipeline summary + suggested next steps. |
/job-hunter:review | Paused adapters + inbox drafts. |
/job-hunter:sync | Regenerate tracking.md. |
/job-hunter:doctor | Validate install / perms / deps. |
PII isolation
The model is forbidden to read ~/.config/job-hunter/secrets/personal.env. It works against assets/personal.env.example (empty-value template) for schema, and the CLI loads real values via python-dotenv at runtime in a child process. The webapp reads the SQLite DB only — it never opens the secrets file.
CI runs lint_secret_leaks.py on every PR; pre-submit checks block any form fill that would echo PII into logs or screenshots. See skills/job-hunter/SKILL.md for the full forbidden-action list.
Contributing
See skills/job-hunter/references/decisions.md for the phase plan and skills/job-hunter/references/ for area-specific docs. New adapters welcome via job adapter contribute <signature>.
License
Apache-2.0. The skill never transmits PII to model servers — see PII isolation and the decisions doc for the architectural enforcement.