convez.ai
SkillDesenvolvimento WebAvançado
Faheem2094

AgentCraft: Framework de Agente Estruturado para Geração de Código

Faheem2094/coding-agent-toolkit

Transforma LLMs em engenheiros de software estruturados e confiáveis

Ver no GitHub
InicialPluginScore · de 0 a 100

Framework universal que transforma modelos de linguagem (Claude, GPT-4) em agentes de software disciplinados e orientados a contexto. Define papéis estruturados (arquiteto, desenvolvedor, revisor, testador), janelas de contexto e contratos de saída para gerar código production-ready com qualidade previsível.

Instalação

$npx convez-ai Faheem2094/coding-agent-toolkit

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
28%
Último commit
há 6 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Geração automática de código full-stack com arquitetura definida
  • Orquestração de múltiplos agentes para diferentes fases do desenvolvimento
  • Padronização de saída de código com convenções e estruturas de projeto
  • Pair programming assistido com documentação e testes automáticos

Veredito

Framework ambicioso para estruturar agentes de código com gramática comportamental clara. Promete eliminar inconsistências em geração com LLMs através de orquestração de papéis e validação em pipeline, mas carece de evidências de implementação e maturidade em produção no repositório atual.

README

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

AgentCraft: The Universal Structured Coding Agent Framework for Multi-Model AI Development

Download

Welcome to the Future of Agentic Code Generation

In 2026, the landscape of software development has shifted from writing code line-by-line to orchestrating intelligent agents that understand intent, structure, and context. AgentCraft is not simply another coding assistant—it is a structured behavior framework that transforms any large language model (from OpenAI's GPT-4o to Anthropic's Claude 3.5 Sonnet) into a disciplined, context-aware software engineer. Think of it as the architectural scaffolding that gives an AI the ability to think in layers, execute in sequences, and deliver production-ready code with predictable quality.


Why AgentCraft Exists

Traditional coding agents operate like brilliant but chaotic artists—they can paint a masterpiece, but you never know if they'll use oil, watercolor, or finger paint. AgentCraft imposes a structural grammar on the creative process. It defines:

  • Roles (architect, coder, reviewer, tester)
  • Context windows (project scope, constraints, dependencies)
  • Output contracts (file structure, naming conventions, documentation standards)

This means your AI coding agent stops guessing and starts engineering.


Mermaid Diagram: The Agent Orchestration Flow

Below is the architectural blueprint of how AgentCraft orchestrates a typical code generation session:

graph TD
    A[User Intent Input] --> B{Intent Parser}
    B -->|Structured| C[Plan Generator]
    B -->|Ambiguous| D[Clarification Loop]
    D --> B
    
    C --> E[Role Selector]
    E --> F[Architect Agent]
    E --> G[Coder Agent]
    E --> H[Reviewer Agent]
    E --> I[Tester Agent]
    
    F --> J[Blueprint Output]
    G --> K[Code Blocks]
    H --> L[Quality Report]
    I --> M[Test Suite]
    
    J --> N[Assembly Pipeline]
    K --> N
    L --> N
    M --> N
    
    N --> O[Structured Repository Output]
    O --> P{Validation Gate}
    P -->|Pass| Q[Final Deliverable]
    P -->|Fail| R[Iteration Loop]
    R --> E

Example Profile Configuration

AgentCraft operates on profiles—JSON configurations that define the personality, constraints, and capabilities of your coding agent. Here is a sample profile for a full-stack web development agent:

agent_profile:
  name: "Full-stack Architect Pro"
  model_preference:
    primary: "claude-3-opus-20240229"
    fallback: "gpt-4-turbo"
  structural_rules:
    output_format: "modular_monorepo"
    file_nesting: "flat_per_feature"
    naming_convention: "kebab-case"
  behavior_constraints:
    max_iterations_per_task: 3
    require_tests: true
    documentation_threshold: "high"
  knowledge_base:
    - "react_v18"
    - "node_22"
    - "postgresql_16"
    - "tailwind_css_v4"
  response_style:
    verbosity: "explanatory_with_comments"
    comment_language: "english"
    tone: "professional_but_approachable"

This configuration tells AgentCraft to use Claude when available, fall back to GPT-4, enforce a monorepo structure with flat file nesting, and generate code that is both tested and well-documented. The agent will explain its reasoning as it outputs code, making it ideal for pair programming.


Example Console Invocation

Here is how you would launch AgentCraft from the command line to generate a complete REST API:

agentcraft --profile fullstack-pro.yaml \
           --output ./generated-api \
           --task "Build a RESTful API for a task management system with user authentication, CRUD operations for tasks, real-time notifications via WebSocket, and a PostgreSQL database. Include integration tests and Swagger documentation." \
           --model claude-3-sonnet-20240229 \
           --verbose

After execution, AgentCraft will produce a complete, structured repository at ./generated-api with:

  • A src/ directory containing modular code
  • A tests/ directory with Jest and Supertest integration tests
  • A docs/ folder with Swagger YAML and OpenAPI spec
  • A docker-compose.yml for local development

Emoji OS Compatibility Table

AgentCraft runs seamlessly across all major operating systems. Here is the compatibility matrix:

Operating SystemStatusNotes
Windows 11 Pro✅Full support via PowerShell and WSL2
macOS Sonoma✅Native ARM and Intel support
Ubuntu 24.04 LTS✅Optimal performance for server deployments
Debian 12✅Fully tested with Docker and Podman
Fedora 40✅Community-verified stability
Arch Linux⚠️Works but requires manual dependency resolution

Feature List

AgentCraft is packed with capabilities designed to make AI-assisted development structured, scalable, and predictable:

  • Multi-Model Orchestration — Seamlessly switch between OpenAI GPT-4o, GPT-4 Turbo, and Claude 3 Opus/Sonnet/Haiku based on task complexity and cost requirements
  • Structured Output Contracts — Define file structure, naming conventions, and code architecture before generation begins
  • Iterative Refinement — Built-in review and test loops that catch errors before they reach production
  • Context Window Optimization — Intelligently truncates and prioritizes context to stay within model token limits
  • Plugin Architecture — Extend AgentCraft with custom validators, formatters, and post-processors
  • Multi-Language Support — Output code in Python, JavaScript, TypeScript, Rust, Go, Java, C#, PHP, Ruby, and Swift
  • Responsive CLI — Real-time progress indicators, estimated time to completion, and detailed logs
  • 24/7 Background Mode — Run generation tasks as daemon processes with webhook callbacks
  • Cost Tracking Dashboard — Monitor API usage and cost per generation session
  • Version Control Integration — Auto-commit generated code to Git repositories with structured commit messages

SEO-Friendly Keywords

This project is optimized for developers searching for:

  • Structured AI coding agent
  • Multi-model code generation framework
  • Claude API code generator
  • OpenAI GPT-4 structured output tool
  • AI software engineering pipeline
  • Reproducible code generation from LLMs
  • Production-ready AI development assistant
  • Agentic coding with context management
  • Scalable AI pair programming framework

OpenAI API and Claude API Integration

AgentCraft provides first-class integration with both major API providers:

OpenAI Integration

  • Models Supported: GPT-4 Turbo (gpt-4-turbo-2024-04-09), GPT-4o (gpt-4o-2024-08-06), GPT-3.5 Turbo
  • Features: Function calling for structured outputs, streaming responses, and token-level cost control
  • Configuration: Set your OPENAI_API_KEY environment variable or pass it via --api-key
  • Advanced: Custom fine-tuned model endpoints for domain-specific code generation

Claude API Integration

  • Models Supported: Claude 3 Opus, Claude 3 Sonnet, Claude 3.5 Sonnet, Claude 3 Haiku
  • Features: Extended context windows (200K tokens), XML-output parsing, and tool-use integration
  • Configuration: Set your ANTHROPIC_API_KEY environment variable
  • Advanced: Prompt caching for repetitive structural contexts (reduces cost by up to 40%)

Responsive UI and Multilingual Support

While AgentCraft is primarily a CLI tool, we have developed a web companion interface that provides:

  • A responsive dashboard for monitoring generation jobs
  • Drag-and-drop profile configuration
  • Visual diff viewer for code output iterations
  • Dark mode and accessibility compliance (WCAG 2.1)

Multilingual Support: AgentCraft can generate code comments, documentation, and user-facing strings in:

  • English (default)
  • Spanish
  • Mandarin Chinese
  • Japanese
  • German
  • French
  • Arabic (with RTL layout support)
  • Hindi

The AI agent automatically detects the project's target language and adjusts its output accordingly.


24/7 Customer Support

We understand that building with AI can present unique challenges. AgentCraft offers:

  • 24/7 Priority Support for enterprise license holders via dedicated Slack channel
  • Community Forum with response times under 2 hours during business days
  • Knowledge Base with 50+ tutorials, common troubleshooting guides, and video walkthroughs
  • Email Support with guaranteed 4-hour first response

Getting Started

  1. Download AgentCraft using the button below
  2. Extract the archive to your preferred directory
  3. Configure your API keys in the .agentcraft/config.yaml file
  4. Run your first generation with the example profile above
  5. Iterate and refine your agent's behavior

Download


Disclaimer

Important: AgentCraft is a tool designed to assist developers in generating code structures and scaffolding. The generated code should always be reviewed, tested, and audited by a qualified human developer before deployment to production environments. The creators of AgentCraft assume no liability for code quality, security vulnerabilities, or licensing compliance of AI-generated outputs. Use at your own risk. Always maintain human oversight in the software development lifecycle.

AgentCraft does not store or transmit your API keys. All API calls are made directly from your local environment to the respective model providers (OpenAI, Anthropic). We recommend reviewing each provider's data usage policy before use.


License

This project is licensed under the MIT License. You are free to use, modify, and distribute AgentCraft for personal, educational, or commercial purposes. See the LICENSE file for full terms.


Contributing

We welcome contributions from the community! Please see our CONTRIBUTING.md file for guidelines on:

  • Bug reports and feature requests
  • Code of conduct
  • Pull request workflow
  • Development environment setup

Final Download Link

Download

AgentCraft — Because great code deserves great structure.

Ver o original no GitHub

Bom para

Relacionados