Job Search Ops
davidsanz21/job-search-ops
Busca de emprego coordenada por time de especialistas IA com documentos verídicos
Ver no GitHubUma skill Claude que orquestra sua busca por emprego como um time coordenado de especialistas em carreira. Integra intake de documentos, entrevista estruturada, CV master, avaliação de fit por vaga, geração de documentos de candidatura personalizados (PDFs) com base em fatos reais, banco de dados de aplicações e coaching de entrevistas com cheatsheet imprimível.
Instalação
npx convez-ai davidsanz21/job-search-ops1 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
- Estruturar e documentar histórico profissional de forma consistente
- Avaliar fit (capacidade e preferência) antes de aplicar em vagas
- Gerar CVs e cartas de apresentação personalizadas sem inventar experiências
- Preparar respostas para entrevistas com feedback e refinamento colaborativo
- Manter banco de dados de aplicações com ângulos, versões e decisões tomadas
Veredito

Skill robusto e bem pensado que resolve um problema real com rigor (verdade obrigatória, consistência entre documentos, controle do usuário). Ideal para quem busca emprego com seriedade e quer eliminar contradições entre CV, carta e entrevista.
README
A tradução para o português deste README ainda está na fila.
job-search-ops
A Claude skill that runs your entire job search as a coordinated team of career experts: document intake, a structured career-profile interview, a master CV, per-posting fit assessment, tailored truth-only application documents as PDFs, an application database, and interview coaching that ends in a printable cheatsheet.
Everything in here was used in production on real applications before it was packaged.
What it does — five phases
| Phase | Trigger | Output |
|---|---|---|
| 1 · Intake | "Start my job search" | document-inventory.md — every career document processed into a fact inventory |
| 2 · Career interview | "next" | career-profile.md — a structured interview (chronology, evidence mining, constraints, honest narratives) builds the single source of truth |
| 3 · Master CV | "next" | master-cv.md — the comprehensive superset every tailored CV is cut from |
| 4 · Application loop | paste any job posting | fit assessment (capability and preference scores, 1–10), apply/skip recommendation, tailored CV + cover letter as PDFs, a database record |
| 5 · Interview coaching | "I have an interview" | question-by-question practice (answer → feedback → co-craft → lock), ending in a cheatsheet PDF |
flowchart LR
subgraph ONCE[Set up once]
direction LR
P1[1 · Intake] --> P2[2 · Career interview] --> P3[3 · Master CV]
end
subgraph LOOP[Per posting]
P4{4 · Fit assessment} -->|apply| OUT[Tailored CV + letter PDFs]
P4 -->|skip| SKIP[Skip record]
end
subgraph PREP[Per interview]
P5[5 · Coaching] --> CS[Cheatsheet PDF]
end
P3 --> P4
OUT --> DB[(application database)]
SKIP --> DB
DB --> P5
The doctrine (why you can trust its output)
- Truth only. Tailoring means selecting, ordering, and phrasing your real facts for a specific audience. The system never invents titles, inflates scope, or claims tools you don't have.
- Honest-bridge framing. Real gaps are named once, early, plainly — then outweighed with evidence. The exact gap wording is recorded and reused, so your story never drifts between CV, letter, and interview.
- You decide and submit. The team assesses and recommends — including recommending you skip. It never applies on your behalf.
- Consistency enforced by data. Every application's angle, document versions, and verbatim gap language go into the database; interview coaching reads that record so it never coaches you into contradicting your own application.
- Private stays private. Your compensation floor and deal-breaker reasoning inform the team's judgment and never appear in an outward document.
Install
Claude.ai: download both .skill files from the latest release, then Settings → Capabilities → Skills → upload each one (claude.ai accepts one skill per upload). For best results, run it inside a dedicated Project and save the canonical files it produces (career-profile.md, master-cv.md, application-database.md) into the project knowledge — that's how state persists between sessions.
Claude Code / Cowork: clone into your skills directory:
git clone https://github.com/davidsanz21/job-search-ops ~/.claude/skills/job-search-ops
Recommended companion: companion-skills/humanaizer/ — install it alongside. Every recruiter-facing text gets a silent humanization pass through it (a condensed fallback is bundled for installs without it).
Use
"Start my job search" → Phase 1
"next" → advance phases
{paste a job posting} → fit assessment → apply/skip → documents
"They rejected me" / "Screen booked Friday" → database update
"I have an interview with {company}" → coaching → cheatsheet
PDF rendering
scripts/render_pdf.py converts the Markdown documents to print-quality A4 PDFs (presets: cv, letter, cheatsheet). Inside Claude's sandbox it runs automatically. Locally:
pip install -r scripts/requirements.txt
python scripts/render_pdf.py examples/sample-cv.md --style cv -o out/cv.pdf
Best with the Lora and Liberation Sans fonts installed; falls back to Georgia/Arial. The script reports page count so the agent can enforce length discipline (CV ≤ 2–3 pages, letter ≤ 1) by restructuring content instead of shrinking type.
What it deliberately does not do
- No auto-applying, no scraping job boards, no sending anything. Scouting automation is a known extension — by design it activates only when a search shifts from selective to volume mode, and it should never auto-apply either.
- No fabrication, ever. If a fact isn't in your profile, it doesn't ship.
Repository layout
SKILL.md orchestrator: doctrine, team, state protocol, phase router
references/ one procedure file per phase + humanizer fallback
assets/templates/ career profile, application database, cheatsheet templates
scripts/ render_pdf.py + print stylesheets (cv / letter / cheatsheet)
examples/ fully synthetic sample inputs and their rendered PDFs
companion-skills/humanaizer/ the text-humanization companion skill
All example data is synthetic. No real names, employers, or application records ship with this repo.
Notes
- The Python in this repository was written with Claude under my direction: I specified the behavior, reviewed the output, and I operate and maintain it. That division of labor is not a disclaimer — it is the working model this repo demonstrates.
License
MIT — see LICENSE. Built by David Sanz.