convez.ai
AgenteDesenvolvimento WebIntermediário
Atis007

Claude Code Kit

Atis007/claude-code-kit

Configuração comportamental modular para agentes de código com advisor e build modes

Ver no GitHub
InicialPluginScore · de 0 a 100

Agent modular para configuração de comportamento de IA em coding, com dois modos: advisor (leitura e consulta) e build (implementação ativa). Inclui padrões stack-específicos para 12+ linguagens (Python, JavaScript, TypeScript, Go, Rust, Java, C, C#, PHP, React, React Native, Express) e é extensível para qualquer stack.

Instalação

$npx convez-ai Atis007/claude-code-kit

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
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Configurar agentes de IA (Claude Code, Cursor, Codex, OpenCode) com instruções consistentes
  • Implementar padrões de desenvolvimento stack-específicos automaticamente
  • Alternar entre modo consultor e modo builder usando estrutura de pastas

Veredito

Kit bem estruturado para padronizar comportamento de agentes de código across múltiplas ferramentas. Elegante na simplicidade (a presença de `context/` define modo), mas requer compreensão de cada ferramenta para instalação correta e integração em fluxo existente.

README

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

claude-code-kit

Modular configuration files for AI coding agents — Claude Code, OpenAI Codex, Cursor, and OpenCode. A single behavior file (CLAUDE.md) defaults to advisor mode. Drop a context/ folder into any project to activate build mode. Stack-specific standards and project scaffolding templates.

CLAUDE.md is the canonical behavior file — the exact same content drives every tool; only its filename and location differ per tool (see Using with Codex, Cursor, and OpenCode).

Ships with C, C#, Express, Go, Java, JavaScript, PHP, Python, React, React Native, Rust, and TypeScript. Designed to grow — add any stack by dropping in two files.

Works alongside the Karpathy behavioral guidelines plugin.

How It Works

A single behavior file installed globally for your tool of choice applies to every project:

  • No context/ directory in the project → Advisor Mode: the agent reads code and answers questions. No file modifications unless explicitly asked.
  • context/ directory exists in the project → Build Mode: the agent reads build-rules.md and all spec files, then implements against them. You lead, the agent builds.

No per-project behavior files. No switching. The presence of context/ is the switch. This works identically across Claude Code, Codex, Cursor, and OpenCode — only the global file's name/location differs per tool.

One-Time Setup

  1. Install the behavior file at your tool's global path — see Using with Codex, Cursor, and OpenCode for the exact name/location. For Claude Code: place CLAUDE.md at ~/.claude/CLAUDE.md.
  2. Install the Karpathy guidelines plugin (recommended, Claude Code only)

That's it. Every project now runs in advisor mode by default.

Using with Codex, Cursor, and OpenCode

CLAUDE.md is the single source of truth for agent behavior. The content is identical for every tool — you just install it under the name and location that tool expects. The context/ folder and all spec files are copied unchanged regardless of tool.

ToolWhere the behavior file goesHow
Claude Code~/.claude/CLAUDE.mdcopy as-is
OpenAI Codex~/.codex/AGENTS.mdrename CLAUDE.md → AGENTS.md
OpenCode~/.config/opencode/AGENTS.mdrename CLAUDE.md → AGENTS.md
Cursorproject AGENTS.md, User Rules, or .cursor/rules/*.mdcsee below

Codex and OpenCode read AGENTS.md as their instructions file (Codex does not read CLAUDE.md at all, so the rename is required). Both also support a project-level AGENTS.md if you prefer per-repo behavior over a global file.

Cursor has three options, pick one:

  • Project file — drop the renamed AGENTS.md in the project root. Cursor reads it as plain-markdown instructions.
  • User Rules — paste the CLAUDE.md contents into Settings → Rules → User Rules to apply it globally across all projects (closest to the other tools' global file).
  • Native project rule — create .cursor/rules/code-advisor.mdc and prepend this YAML frontmatter above the CLAUDE.md body:
    ---
    description: Code advisor / build-mode behavior
    alwaysApply: true
    ---
    

The advisor-vs-build-mode switch (presence of context/) works the same in every case — it is behavior described inside the file, not tool-specific config.

Repo Structure

├── CLAUDE.md                              ← Behavior file; install per tool (once) — see "Using with Codex, Cursor, and OpenCode"
└── context/
    ├── build-rules.md                     ← Build mode instructions
    ├── templates/                         ← Universal, fill per project
    │   ├── ai-workflow-rules.md
    │   ├── progress-tracker.md
    │   ├── project-overview.md
    │   └── ui-context.md
    ├── standards/                         ← Stack-specific, reusable as-is
    │   ├── code-standards-c.md
    │   ├── code-standards-csharp.md
    │   ├── code-standards-express.md
    │   ├── code-standards-go.md
    │   ├── code-standards-java.md
    │   ├── code-standards-js.md
    │   ├── code-standards-php.md
    │   ├── code-standards-python.md
    │   ├── code-standards-react.md
    │   ├── code-standards-rn.md
    │   ├── code-standards-rust.md
    │   └── code-standards-typescript.md
    └── architecture/                      ← Stack starters, pick + customize
        ├── architecture-c.md
        ├── architecture-csharp.md
        ├── architecture-express.md
        ├── architecture-go.md
        ├── architecture-java.md
        ├── architecture-js.md
        ├── architecture-php.md
        ├── architecture-python.md
        ├── architecture-react.md
        ├── architecture-rn.md
        └── architecture-rust.md

Templates are language-agnostic — same files for every project, filled in with your specifics.

Standards and architecture are stack-specific — pick the ones matching your project, or create new ones for any stack.

Activating Build Mode for a Project

  1. Create a context/ folder in your project root
  2. Copy build-rules.md into context/
  3. Copy the templates you need from templates/ into context/ — flat, as siblings to build-rules.md, not inside a subfolder
  4. Pick your architecture starter from architecture/, copy into context/ and rename to architecture.md
  5. Pick your code standards from standards/, copy into context/
  6. Fill in the template placeholders

Important: In your project, all files must be flat inside context/. The subdirectories (templates/, standards/, architecture/) only exist in this repo for organization. Your project should look like this:

my-project/
├── context/
│   ├── build-rules.md
│   ├── project-overview.md
│   ├── architecture.md              ← renamed from architecture-{stack}.md
│   ├── ui-context.md
│   ├── code-standards-typescript.md
│   ├── code-standards-rn.md
│   ├── ai-workflow-rules.md
│   └── progress-tracker.md
├── src/
└── ...

Example Stack Combos

Files to copy flat into your project's context/ (alongside build-rules.md):

C (systems)

build-rules.md
project-overview.md
architecture.md              ← from architecture/architecture-c.md
code-standards-c.md
ai-workflow-rules.md
progress-tracker.md

C# / .NET (backend)

build-rules.md
project-overview.md
architecture.md              ← from architecture/architecture-csharp.md
code-standards-csharp.md
ai-workflow-rules.md
progress-tracker.md

Express.js + TypeScript (backend)

build-rules.md
project-overview.md
architecture.md              ← from architecture/architecture-express.md
code-standards-typescript.md
code-standards-express.md
ai-workflow-rules.md
progress-tracker.md

Go (backend / systems)

build-rules.md
project-overview.md
architecture.md              ← from architecture/architecture-go.md
code-standards-go.md
ai-workflow-rules.md
progress-tracker.md

Java (backend)

build-rules.md
project-overview.md
architecture.md              ← from architecture/architecture-java.md
code-standards-java.md
ai-workflow-rules.md
progress-tracker.md

JavaScript (vanilla / Node.js)

build-rules.md
project-overview.md
architecture.md              ← from architecture/architecture-js.md
code-standards-js.md
ai-workflow-rules.md
progress-tracker.md

PHP REST API (backend, no framework)

build-rules.md
project-overview.md
architecture.md              ← from architecture/architecture-php.md
code-standards-php.md
ai-workflow-rules.md
progress-tracker.md

Python (backend / scripting)

build-rules.md
project-overview.md
architecture.md              ← from architecture/architecture-python.md
code-standards-python.md
ai-workflow-rules.md
progress-tracker.md

React + Tailwind (web frontend)

build-rules.md
project-overview.md
architecture.md              ← from architecture/architecture-react.md
ui-context.md
code-standards-typescript.md
code-standards-react.md
ai-workflow-rules.md
progress-tracker.md

React Native + Expo (mobile)

build-rules.md
project-overview.md
architecture.md              ← from architecture/architecture-rn.md
ui-context.md
code-standards-typescript.md
code-standards-rn.md
ai-workflow-rules.md
progress-tracker.md

Rust (backend / systems)

build-rules.md
project-overview.md
architecture.md              ← from architecture/architecture-rust.md
code-standards-rust.md
ai-workflow-rules.md
progress-tracker.md

Full-stack: Combine frontend and backend files. Use one architecture.md covering both layers, or split into architecture-frontend.md and architecture-backend.md.

Adding a New Stack

The kit grows with you. To add Kotlin, Swift, Zig, or anything else:

  1. Create context/standards/code-standards-{stack}.md — your coding conventions for that stack
  2. Create context/architecture/architecture-{stack}.md — typical system structure and boundaries
  3. Add the combo to this README

That's it. The templates, workflow rules, and CLAUDE.md work with any stack unchanged.

Prerequisites

License

MIT

Ver o original no GitHub

Bom para

Relacionados