Shared Claude Skills – Vuln Hunter
genkist/shared-claude-skills
Scanner de vulnerabilidades paralelo com 15 agentes especializados e validação de ruído
Ver no GitHubColeção de skills (plugins) para Claude Code focados em segurança e workflows de desenvolvimento. O skill principal, vuln-hunter, é um scanner de vulnerabilidades multi-estágio que executa 15 agentes especializados em paralelo, cada um caçando uma classe de vulnerabilidade diferente, seguido de validação e rastreamento de impacto.
Instalação
npx convez-ai genkist/shared-claude-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
- Auditoria de segurança em código-fonte e configurações de infraestrutura
- Varredura de mudanças git (diff, staged) antes de commit/deploy
- Identificação de padrões de segurança específicos de IA e lógica de negócio
- Validação de reachability e impacto efetivo de findings críticos
Veredito

Ferramenta sólida e bem arquitetada para scanning de segurança em escala. A pipeline paralela com validação reduz falsos positivos, e a cobertura de 15 classes de vulnerabilidade (incluindo padrões de IA) é abrangente. Recomendada para times que precisam de scanning automatizado rigoroso em CI/CD ou code review.
README
A tradução para o português deste README ainda está na fila.
shared-claude-skills
A collection of Claude Code plugins (skills) for security and development workflows.
Skills
vuln-hunter
Multi-stage security scanner for application code and infrastructure configurations, inspired by Cloudflare's Project Glasswing.
Instead of a single generic scan, parallel specialized agents each hunt one vulnerability class, then a validation pass eliminates noise.
Pipeline:
RECON → HUNT (15 agents, parallel) → VALIDATE → TRACE → REPORT
| Phase | What it does |
|---|---|
| RECON | Maps entry points, trust boundaries, data flows, and infra topology |
| HUNT | 15 agents run simultaneously, each targeting one vulnerability class |
| VALIDATE | Eliminates false positives, deduplicates, normalizes severity |
| TRACE | Verifies reachability (code) or effective impact (infra) for Critical/High findings |
Vulnerability classes covered: Injection · Auth & Session · Secrets Exposure · Access Control · Crypto Failures · Input Validation · SSRF & Redirects · Supply Chain · Concurrency · Security Misconfiguration · AI-Specific Patterns · Business Logic · IAM & Privilege · Container Security · Network Exposure
Usage:
/vuln-hunter # full scan of current directory (code + infra)
/vuln-hunter diff # scan git diff (staged + unstaged)
/vuln-hunter staged # scan staged changes only
/vuln-hunter path/to/file # scan a specific file or directory
Installation
Clone the repository, then load the plugin with --plugin-dir:
git clone https://github.com/genkist/shared-claude-skills.git
Start Claude Code with a specific plugin:
claude --plugin-dir ./shared-claude-skills/vuln-hunter
Or add to your project's plugin config (.claude/settings.json):
{
"pluginDirs": ["/absolute/path/to/shared-claude-skills/vuln-hunter"]
}
The skill becomes available as a slash command once the plugin is loaded.
Keep skills up to date:
cd shared-claude-skills && git pull
Contributing
New skills are welcome. Each skill lives in its own top-level directory with the following structure:
my-skill/
├── .claude-plugin/
│ └── plugin.json # name, version, description, author
└── skills/
└── my-skill/
└── SKILL.md # skill definition and instructions
plugin.json minimum fields:
{
"name": "my-skill",
"version": "0.1.0",
"description": "One-line description of what the skill does.",
"author": {
"name": "your-github-username"
}
}
Open a pull request with your new skill directory.
License
MIT