OpenSkills
huaguihai/openskills
Skills open-source para potencializar agentes IA com automação e análise
Ver no GitHubColeção de 12 skills open-source para agentes IA (OpenClaw e Claude Code), incluindo automação, segurança, geração de conteúdo e análise. Oferece funcionalidades prontas como vetting de segurança, web scraping inteligente, geração de imagens com Grok e monitoramento de contexto.
Instalação
npx convez-ai huaguihai/openskills1 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
- Adicionar capacidades de segurança e auditoria a agentes IA
- Automatizar pipelines de escrita (blogs, READMEs, curação de conteúdo)
- Implementar web scraping inteligente e busca de APIs públicas
- Gerar e editar imagens com modelos Grok em fluxos de agentes
Veredito

Conjunto maduro e bem documentado de skills reutilizáveis para quem usa OpenClaw/Claude Code. Cobre casos de uso reais (segurança, conteúdo, web scraping), mas exige familiaridade com agentes IA e instalação manual via shell.
README
A tradução para o português deste README ainda está na fila.
openskills
Open-source AI agent skills for OpenClaw and Claude Code.
Good for: Developers using OpenClaw or Claude Code who want to add capabilities to their AI agent Not for: Users who don't use AI agents
Agent Quick Start
Copy this block to your AI assistant:
# Download the skill collection
git clone https://github.com/huaguihai/openskills.git
cd openskills
# Install a single skill (sentinel as example)
cp -r skills/sentinel ~/.openclaw/skills/
# Or for Claude Code users
cp -r skills/sentinel ~/.claude/skills/
# Verify installation
ls ~/.openclaw/skills/sentinel/SKILL.md # Should print the file path
# Install all skills
for skill in skills/*/; do cp -r "$skill" ~/.openclaw/skills/; done
Skills
| Skill | Description | Category |
|---|---|---|
| sentinel | Unified security — skill vetting, dependency interception (3-layer defense), project health check, system audit | Security |
| readme-writer | Generate, optimize, or QA-check READMEs that serve both humans and AI agents, with S/M/L project grading | Docs |
| blog-pipeline | End-to-end blog writing pipeline with style enforcement and independent review | Writing |
| public-apis | Find and recommend free public APIs across 51 categories | Dev |
| opportunity-radar | Indie dev opportunity discovery — 10 transformation strategies to find software business ideas | Business |
| smart-fetch | Smart web scraping router — auto-selects the right tool (Jina/WebFetch/curl) for any URL | Utility |
| grok-image | AI image generation and editing using Grok Imagine models | Creative |
| auto-approve | Analyze tool approval logs, discover high-frequency patterns, add to auto-approve list | Efficiency |
| context-monitor | Status-line meter for Claude Code context usage — color warnings before overflow, handles /compact correctly | Efficiency |
| curator-pipeline | Curation pipeline — turn YouTube transcripts, article URLs, or pasted text into sellable HTML-PPT products | Writing |
| handoff | Cross-agent work handoff protocol — factual HANDOFF.md + clean worktree on the way out, reconcile against real git state before picking up | Workflow |
| agnes-ai | Agnes AI image AND video generation — text/image-to-image, plus text/image/multi-image/keyframe video (async), results auto-downloaded locally | Creative |
Installation
# Install a single skill
cp -r skills/<skill-name> ~/.openclaw/skills/
# Or for Claude Code users
cp -r skills/<skill-name> ~/.claude/skills/
# Install all skills
for skill in skills/*/; do cp -r "$skill" ~/.openclaw/skills/; done
For sentinel's auto-interception (M2), you also need to configure Claude Code hooks — see sentinel/SKILL.md.
How It Works
graph LR
A[User request] --> B{Claude Code / OpenClaw}
B --> C[Match SKILL.md description]
C --> D[Load skill instructions]
D --> E[Execute skill workflow]
E --> F[Return result]
Each skill is a standalone directory. The core is SKILL.md:
- Trigger: Claude matches the user's request against the
descriptionfield in SKILL.md frontmatter - Load: On match, the SKILL.md body is loaded into context
- Execute: Follow instructions — read
references/, runscripts/, call tools - Output: Complete the task and return results
Skill directory structure
skills/<name>/
├── SKILL.md # Required: trigger description + instructions
├── references/ # Optional: reference docs (loaded on demand, saves tokens)
├── scripts/ # Optional: executable scripts for deterministic tasks
└── assets/ # Optional: templates, icons, static files
Troubleshooting
Skill not triggering after install?
→ Verify the file is in place: ls ~/.openclaw/skills/<name>/SKILL.md. Skill triggering is description-based — try phrasing your request closer to the skill's description
Want to install to Claude Code instead of OpenClaw?
→ Replace ~/.openclaw/skills/ with ~/.claude/skills/, everything else is the same
How to write your own skill? → Look at any existing skill's SKILL.md for the format. Core: YAML frontmatter (name + description) + Markdown instruction body