convez.ai
SkillAutomaçãoIntermediário
mhenke

Claude Code Unplugged

mhenke/claude-code-unplugged

Skills portáveis do Claude Code para assistentes IA sem lock-in

Ver no GitHub
InicialPluginScore · de 0 a 100

Coleção de skills reutilizáveis extraídas do Claude Code, contendo workflows estruturados, prompts battle-tested e scripts de automação adaptáveis para qualquer assistente de IA (Cursor, Copilot, Gemini). Inclui padrões para desenvolvimento de features em 7 fases, revisão automática de código com personas paralelas, design frontend otimizado e integração com Git.

Instalação

$npx convez-ai mhenke/claude-code-unplugged

1 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

Estrelas
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 3 semanas
Versão
—
Issues abertas
16
Contribuidores
1

Casos de uso

  • Automação de fluxos de desenvolvimento multi-fase (discovery → design → implementação → review)
  • Revisão paralela de PRs com personas especializadas (bug hunters, simplificadores, analisadores)
  • Geração inteligente de commits e scripts Git
  • Orientação estruturada para design frontend com cores HSL e micro-animações

Veredito

Extração bem executada de lógica battle-tested do Claude Code em formato agnóstico. Oferece workflows robustos e reutilizáveis, mas demanda integração manual em cada assistente; valor máximo para quem trabalha com múltiplas ferramentas de IA.

README

A tradução para o português deste README ainda está na fila.

Claude Code Unplugged 🔌

Portable agent skills, workflows, commands, and automation scripts extracted from Claude Code, ready to use in any AI coding assistant (Cursor, Gemini CLI, Copilot, VS Code, etc.).

💡 Why this exists

Claude Code is fantastic, but some of its best features—like the structured multi-phase feature development loop, automated PR reviews, and smart commit generation—are tightly bound to its CLI.

We wanted to use those same battle-tested workflows, prompts, and helper scripts in Cursor, Gemini, Qwen, or custom local agents.

This repository extracts those prompt patterns and translates them into the open Agent Skills format (SKILL.md). Install the collection with:

npx skills add mhenke/claude-code-unplugged

No platform lock-in. No proprietary runtimes. Just great agent prompts and scripts that work anywhere.


📦 What's inside?

We've organized the extractions into a flat skills/ directory:

🚀 Workflows & Prompt Skills

Modular agent instructions that teach your AI assistant how to perform complex, multi-step engineering tasks:

  • frontend-design — Guides agents to build gorgeous, responsive UIs with curated HSL colors, Google Fonts, and smooth micro-animations instead of basic browser defaults.
  • commit-commands — Ported versions of /commit, /commit-push-pr, and /clean_gone translated into clean, step-by-step instructions.
  • code-review — The parallel agent PR review logic. Instructs your agent to spawn parallel review personas (project guideline checkers, bug hunters, and history analyzers) to score and comment on code.
  • feature-dev — A 7-phase software engineering cycle: Discovery ➡️ Codebase Exploration ➡️ Clarifying Questions ➡️ Architecture Design ➡️ Implementation ➡️ Quality Review ➡️ Summary.

🎭 Agent Personas & Output Styles

Instruct your coding assistant to act as a specific sub-agent or write in a different mode:

  • pr-review-toolkit — Specialised reviewer personas: Silent Failure Hunter, Code Simplifier, PR Test Analyzer, and Type Design Analyzer.
  • explanatory-output-style — Puts the model into an educational mode, requiring it to explain implementation choices with formatted ★ Insight boxes before and after code edits.
  • learning-output-style — A hybrid learning mode. Instead of doing all the work, the agent is instructed to write boilerplate, identify key design files, and request 5-10 line code contributions from you.
  • ralph-wiggum — Iterative self-enforcing feedback loop. The agent tracks its own promise-completion state and loops until done.

🛠️ Automation & Helper Scripts

Executable scripts bundled inside respective skill folders that your agent can run locally using its terminal/bash tool:

  • security-guidance — Ported security guidance and a regex-based warning script (python3 scripts/security_reminder_hook.py) for spotting secrets, raw innerHTML, and unsafe serialization.
  • github-management — Automation scripts copied directly from the Claude Code repository to auto-detect duplicate issues, close stale issues, manage labels, and run sweeps.

🔌 Plugin & Agent Development

Skills for building new plugins, agents, hooks, and skills for AI coding assistants:

  • plugin-dev — Comprehensive guide for developing new CLI plugins from scratch.
  • plugin-structure — Plugin directory layout, manifest config, component organization, and naming conventions.
  • plugin-settings — Patterns for user-configurable plugin settings via .agent/plugin-name.local.md files with YAML frontmatter.
  • agent-development — Agent structure, system prompts, triggering conditions, and development best practices.
  • hook-development — Creating PreToolUse/PostToolUse/Stop hooks with prompt-based enforcement patterns.
  • hookify — Extensible user-configured prompt hooks with a full rule engine and examples.
  • command-development — Slash command structure, YAML frontmatter fields, dynamic arguments, and user interaction patterns.
  • skill-development — Skill structure, progressive disclosure patterns, and best practices for writing new skills.
  • mcp-integration — Integrating Model Context Protocol (MCP) servers into plugins for external tool and service integration.
  • writing-rules — Hookify rule syntax and patterns for building agent governance rules.

🔄 Migration & SDK


⚙️ Installation

To add this collection to your local workspace, run:

npx skills add mhenke/claude-code-unplugged

After installation, the code-review skill is available in your workspace.


🕊️ Philosophy

  • CLI & Model Agnostic: Translated from proprietary syntax (! command interpolations or Claude-only hooks) into standard English directives that any modern LLM can interpret.
  • Pure Markdown: Every skill uses the standard SKILL.md format. Repo tooling uses Node.js, but the skills themselves are plain Markdown.
  • Keep it Simple: No complex compilation pipelines. Just copy, refactor, and distribute.

🤝 Contributing

Find a cool command, agent, or workflow in the Claude Code source tree that isn't here? Open a PR! You can run our built-in extraction utility to sync and validate:

node scripts/extract.js --source /path/to/claude-code --target .
node scripts/validate.js

Disclaimer: This repository is a community-driven project to make prompts portable. It is not affiliated with or endorsed by Anthropic.

Ver o original no GitHub

Bom para

Relacionados