JobClaw Skills
jain777/jobclaw-skills
Automação completa da busca de emprego dentro de agentes IA
Ver no GitHubConjunto de 20 skills open-source em Python para automatizar toda a busca de emprego dentro de um agente de IA. Inclui onboarding via perfil único, busca multi-fonte com 13 adaptadores, scoring de fit, geração de resumes e cover letters otimizados, triagem de recrutadores, preparação de entrevistas e negociação de ofertas. Sem necessidade de chaves externas para a maioria das skills.
Instalação
npx convez-ai jain777/jobclaw-skills1 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
- Buscar vagas em múltiplas plataformas (Greenhouse, Lever, YC, HN) com deduplicação
- Gerar resumes e cover letters personalizados otimizados para ATS
- Preparar entrevistas com mock interviews e questões contextualizadas
- Classificar e responder emails de recrutadores de forma automatizada
- Negociar ofertas com coaching baseado em dados do mercado
Veredito

Ferramental robusto e bem-documentado para quem quer automatizar processos seletivos. A abordagem de perfil único + múltiplas skills interconnectadas é elegante, mas requer integração cuidadosa e compreensão da arquitetura para máximo proveito. Prático para quem domina Python e quer controle total do pipeline de candidatura.
README
A tradução para o português deste README ainda está na fila.
What is this
JobClaw by whatNxtAI is a set of open-source Agent Skills that run your entire job hunt from inside an AI coding agent. Each skill is a self-contained SKILL.md workflow; together they form a pipeline that:
- Onboards you once — a resume / LinkedIn PDF (links and all) becomes the canonical master profile every other skill reads
- Finds and scores jobs — 13 pluggable adapters (ATS-direct, SmartRecruiters, YC, HN…), deduped and freshness-verified, ranked against your profile
- Builds the materials — ATS-optimized tailored resume PDFs, cover letters, and application-form answers
- Runs the recruiter loop — classifies inbox emails, infers application status, drafts tone-matched replies
- Preps the human moments — interview question banks + mock interviews, and offer-negotiation coaching
This is a filter, not a spray-and-pray tool. The skills never fabricate, never echo your private goals into anything a recruiter sees, and never submit anything for you — every artifact stops before send so you have the final call. See
skills/_shared/RULES.md.
Bring your own keys — but most skills need none, because the model itself is the engine. ~14 of the 20 skills are pure reasoning (no external API); only job search and company research use optional data-API keys.
These are the reusable brain behind JobClaw — whatNxtAI's autonomous job-hunt agent — but they stand alone: run them yourself in any supported runtime, no agent required.
Features
| Feature | Description |
|---|---|
| One master profile | Onboard once from a resume/LinkedIn PDF (mines embedded links, can import ChatGPT/Claude/Gemini career memories); every skill reads it, nothing is re-asked. |
| Multi-source search | 13 adapters across Greenhouse/Lever/Ashby/Workday/SmartRecruiters/Teamtailor/YC/Cutshort/HN + Adzuna/SerpApi/Firecrawl, deduped across runs and verified live for freshness. |
| Fit scoring | ATS-style match % + matched/missing keywords + gaps + an apply-or-skip call, work-authorization aware (judged against the job's region). |
| ATS resume PDFs | Tailored, keyword-aligned resumes rendered via rendercv (9 themes, clickable links, ATS-safe default) with a visual-QA auto-fix loop. |
| Recruiter inbox loop | Classify → infer status → draft reply, driven by a shared email-class × application-status taxonomy. |
| Interview + negotiation | Question banks with STAR talking points, a text mock-interview loop with a scored report, and leverage-based counter-offer drafts. |
| Region intelligence | Per-region (US, India) sources, sponsorship logic, resume/comp conventions — add a country by copying one schema file. |
| Human-in-the-loop | The model evaluates and drafts; you decide and send. No auto-submit, ever. |
| Never fabricates | Selects, reorders, and reframes real experience; flags genuine gaps instead of inventing them. |
Quick start
# 1. Clone
git clone https://github.com/jain777/jobclaw-skills && cd jobclaw-skills
# 2. Wire it into your agent (symlinks skills/ + generates Cursor rules; idempotent)
./scripts/install.sh # all runtimes — or: codex | cursor | hermes | openclaw | claude
# 3. (Optional) install rendercv for resume PDFs, then check your setup
python3 -m venv skills/render-resume/.venv
skills/render-resume/.venv/bin/pip install "rendercv[full]"
python3 scripts/doctor.py # reports rendercv + which optional API keys are set
# 4. Open your agent in this directory and onboard
# e.g. /build-profile (share a resume/LinkedIn PDF)
Use with your runtime
These follow the open agent-skills standard (SKILL.md), so the same skills/ directory drives every runtime. install.sh (a stdlib-only wrapper over scripts/setup_runtime.py) links skills/ into each runtime's discovery path; pass --global for user-level dirs, or --copy on Windows / restricted filesystems.
| Runtime | Wire it | Then invoke |
|---|---|---|
| Claude Code | plugin (below), or ./scripts/install.sh claude | /skill-name |
| Codex | ./scripts/install.sh codex (links .agents/skills) | /skills, $mention, or auto |
| Cursor | ./scripts/install.sh cursor (generates .cursor/rules/*.mdc) | auto-attach, or @rule-name |
| Hermes | native (./skills/); --global links ~/.hermes/skills/ | auto (description-gated) |
| OpenClaw | native — run with this repo as the workspace | openclaw skills … or auto |
Every runtime also reads AGENTS.md, the runtime-neutral guide. Full matrix, invocation details, and tool-namespacing notes: docs/RUNTIMES.md.
Claude Code plugin (one command):
/plugin marketplace add jain777/jobclaw-skills
/plugin install jobclaw-skills
(Replace jain777/jobclaw-skills with your fork/repo. The plugin auto-loads every skill in skills/.)
Usage
Invoke a skill by name (slash command in Claude Code / Cursor / Codex; description-matched auto-invoke in Hermes / OpenClaw):
/build-profile → onboard: resume/LinkedIn PDF → master profile
/find-jobs → search + rank jobs against your profile
/score-fit → ATS-style fit score + apply/skip call for one job
/apply-to-job <url> → full chain: score → resume PDF → cover letter → answers (stops before submit)
/tailor-resume → job-specific ATS resume, rendered to PDF + visually QA'd
/write-cover-letter → one-page, region-aware cover letter
/answer-application-questions → format-obeying answers to a form's questions
/triage-inbox → classify a recruiter email + extract dates/links/asks
/infer-status → canonical application status + next action
/draft-reply → tone-matched reply for a chosen intent
/research-company → comp, day-to-day, growth, sentiment, red flags
/prep-interview → likely questions + STAR talking points
/mock-interview → text interview loop + scored report
/coach-negotiation → leverage analysis + ranges + a draft counter (you send it)
/career-coach → open Q&A loaded with your profile
/write-outreach → cold recruiter/referral message with one CTA
/map-career-path → transition examples + a gap roadmap
Or just paste a job URL or description — apply-to-job / score-fit auto-detect it and capture the job once (jobs/current.json), so you never retype the JD.
How it works
/build-profile profile/master-profile.md ── the shared memory ──┐
│ │
▼ │
/find-jobs ──► /score-fit ──► apply or skip │
│ │ (>= threshold) │
│ ▼ │
│ /apply-to-job ──► /tailor-resume (PDF) │
│ /write-cover-letter reads ─┤
│ /answer-application-questions │
│ │ │
│ stops before submit — you send │
▼ │
recruiter loop: /triage-inbox ──► /infer-status ──► /draft-reply reads ─┤
│
interview: /research-company ──► /prep-interview ──► /mock-interview │
│
offer: /coach-negotiation (draft counter — you send) reads ─┘
Skills compose through JSON sidecars (jobs/current.json, scores/, resumes/,
companies/, inbox/), not framework calls — so the chain runs on any runtime.
Full stage map: knowledge/pipeline.md.
Skill catalog (20)
Foundation & orchestration
| Skill | Does | Keys |
|---|---|---|
build-profile | Turns resume / LinkedIn PDF / AI-memory + enriched links into the canonical master profile. | none |
apply-to-job | Orchestrates the apply chain for one job (score → tailored resume PDF → cover letter → answers) in auto-pilot or review mode; stops before submit. | none |
Discovery
| Skill | Does | Keys |
|---|---|---|
find-jobs | Searches jobs across 13 pluggable adapters (ATS-direct, SmartRecruiters, Teamtailor, YC, Cutshort, Adzuna, SerpApi, Firecrawl, HN…); dedupes + verifies freshness. See docs/JOB_SEARCH.md. | optional |
score-fit | ATS-style fit score + matched/missing keywords + gaps + apply/skip call; work-auth aware. | none |
research-company | Comp, day-to-day, growth, news, sentiment, red flags. | optional (web) |
map-career-path | Real transition examples + a skill/experience gap roadmap. | optional (CSE) |
Materials
| Skill | Does | Keys |
|---|---|---|
tailor-resume | Job-specific, ATS-optimized resume — renders the PDF by default (+ visual QA); never fabricates. | rendercv |
render-resume | Typesets a resume to a polished PDF via rendercv (9 themes, clickable links, ATS-safe default). | rendercv |
review-render | Visual-QA auto-fix loop for the rendered PDF (widows/overflow/fill). | rendercv |
write-cover-letter | Tone-controlled, region-aware, one-page letter. | none |
answer-application-questions | Correct, format-obeying answers to application form questions. | none |
Comms & status
| Skill | Does | Keys |
|---|---|---|
write-outreach | Cold recruiter/referral message with one clear CTA. | none |
triage-inbox | Classifies a recruiter email + extracts dates/links/asks + suggested action. | none |
draft-reply | Tone-matched reply draft for a chosen intent. | none |
infer-status | Canonical application status + next action from signals. | none |
Interview, negotiation & coaching
| Skill | Does | Keys |
|---|---|---|
prep-interview | Likely questions + STAR talking points + a prep plan. | none |
mock-interview | Text interview loop + a scored report. | none |
coach-negotiation | Leverage analysis + ranges + a draft counter email (you send it). | none |
career-coach | Open Q&A loaded with your master profile. | none |
request-human-input | The reasoning half of human-in-the-loop; composes a mobile-first question + parses the reply (pairs with the optional agent). | none |
Knowledge packs
Shared intelligence every skill reads:
knowledge/pipeline.md— the stage map + each skill's next step (how the skills chain).knowledge/regions/— per-region (US, India) job sources, visa/sponsorship logic, resume & comp conventions. Add a country by copying_schema.md.knowledge/work-authorization.md— canonical work-auth enum + the cross-region rule (judge against the job's region).knowledge/companies/— a 3500+ company catalog (companies.csv, all industries + YC + frontier AI) + a cached ATS resolver (resolved.json).knowledge/ai-roles.md— AI title filter + role archetypes.knowledge/status/taxonomy.md— email-class × application-status decision table.
Project structure
jobclaw-skills/
├── AGENTS.md # Runtime-neutral entry point (Codex / Cursor / Hermes)
├── CLAUDE.md # Claude Code contributor notes
├── skills/ # 20 SKILL.md skills (+ _shared/RULES.md, the canon)
│ └── <verb-noun>/
│ ├── SKILL.md # one skill workflow
│ ├── reference/ # optional schemas / rubrics
│ └── scripts/ # optional Python (stdlib) helpers
├── knowledge/ # shared packs: regions, companies, ai-roles, status, pipeline
├── scripts/
│ ├── install.sh # wire skills into a runtime (wrapper)
│ ├── setup_runtime.py # per-runtime discovery wiring
│ ├── gen_cursor_rules.py # SKILL.md → .cursor/rules/*.mdc
│ └── doctor.py # check rendercv + API keys
├── .cursor/rules/ # generated Cursor rules (one per skill)
├── docs/ # RUNTIMES.md (compat matrix) + JOB_SEARCH.md
├── tests/ # smoke.py runner + fixtures
└── profile/ # master-profile.example.md (your real one is gitignored)
Keys (all optional)
Copy .env.example → .env. Most skills need nothing. Job search gets stronger as you add keys (Adzuna is free; SerpApi/Firecrawl freemium) — full breakdown in docs/JOB_SEARCH.md.
Design notes
- The master profile (
profile/master-profile.md) is the shared memory;build-profilecreates it, the rest consume it. Skills also compose via small JSON sidecars (e.g.scores/<id>.score.json→tailor-resume). - The model is the engine — no external LLM key for reasoning skills; runs on whichever model your runtime provides.
- Runtime-agnostic by design —
SKILL.md+ relative-path knowledge packs + JSON sidecars, no framework lock-in. Seedocs/RUNTIMES.md. - Scripts are Python 3 stdlib only (zero build), print JSON to stdout, read keys from env, and degrade gracefully when a key is absent.
- Never fabricates — every materials/comms skill is grounded in the profile; see
skills/_shared/RULES.md.
Roadmap
More region packs (Canada / EU / Singapore), Naukri & Internshala adapters (India consumer/FMCG recall), and direct enterprise-ATS adapters. (Relevance ranking shipped — find-jobs computes a real fit_rank and score-fit/build-profile share one matching engine; see knowledge/relevance.md.) Job-search strategy + caveats: docs/JOB_SEARCH.md.
About
JobClaw is built by whatNxtAI. These skills are the open-source brain; the full product is an autonomous job-hunt agent that runs the same pipeline for you end to end. Encouraging-coach by design: it does the legwork, you make the calls.
License & credits
MIT. "JobClaw" and "whatNxtAI" are brand names of the whatNxtAI project. The AI-role archetype taxonomy in knowledge/ai-roles.md is adapted from the open-source career-ops project. Contributions welcome — see CONTRIBUTING.md.