Codex Forge - Orquestrador de Revisão de Código com IA
aliywsf2004/codex-forge
Revisor de código autônomo rodando em background com Codex + Claude
Ver no GitHubSkill que automatiza revisão de código e refatoração em background usando Codex e Claude via orquestração de jobs assíncronos. Integra-se com CI/CD pipelines para análise contínua de código sem bloquear fluxo de desenvolvimento. Fornece controle total sobre execução, retry automático e relatórios detalhados.
Instalação
npx convez-ai aliywsf2004/codex-forge1 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
- Automatizar revisão de pull requests em CI/CD
- Refatoração em background durante desenvolvimento
- Análise multi-modelo contínua de qualidade de código
Veredito

Projeto ambicioso que trata code review como processo contínuo em background, não evento pontual. Implementação sólida de orquestração de jobs com Codex/Claude, mas documentação incompleta no README prejudica adoção imediata — recomendado para times que já usam orquestração de background tasks.
README
A tradução para o português deste README ainda está na fila.
AI Code Review Orchestrator 2026: Autonomous Codex-Powered Background Workflow Engine
The Silent Architect of Modern Code Quality
In software development, we often treat code review as a bottleneck—a necessary evil that slows velocity. But what if a background process could handle the cognitive load while you focus on architecture? This repository introduces a paradigm shift: an autonomous orchestration layer that integrates OpenAI Codex with Claude Code to perform delegated code reviews, refactoring tasks, and multi-model analysis—all running as asynchronous background jobs with full job control.
Version 1.0.0 | MIT License | Works with OpenAI API & Claude API
Table of Contents
- Concept & Philosophy
- System Architecture (Mermaid Diagram)
- Key Features That Redefine Workflow Automation
- Job Control & Background Execution System
- Example Profile Configuration
- Example Console Invocation
- OS Compatibility Table
- API Integration Guide
- Multilingual Support & Responsive UI
- Disclaimer
- License
Concept & Philosophy
Imagine having a tireless junior developer who never sleeps, never gets tired, and consults both Codex and Claude for every decision. That is what this project delivers. We treat code review not as an event but as a continuous, background process—like a heartbeat monitor for your repository.
The orchestration engine sits between your CI/CD pipeline and the LLM APIs, managing jobs, handling retries, and providing real-time status. This is not just another code review tool; it is a delegation framework that turns AI models into collaborative workers.
SEO Keywords: AI code review automation, background job control, Codex Claude integration, autonomous code analysis, multithreaded LLM orchestration, 2026 software development pipeline.
System Architecture (Mermaid Diagram)
graph TB
subgraph "User Layer"
CLI[Console Interface]
UI[Responsive Web UI]
end
subgraph "Orchestration Engine"
Scheduler[Job Scheduler]
Queue[Background Job Queue]
Monitor[Job Monitor & Control]
end
subgraph "AI Model Layer"
Codex[OpenAI Codex API]
Claude[Claude API]
Router[Model Router]
end
subgraph "Output Pipeline"
Review[Code Review Report]
Refactor[Refactored Code]
Logs[Execution Logs]
end
CLI --> Scheduler
UI --> Scheduler
Scheduler --> Queue
Queue --> Router
Router --> Codex
Router --> Claude
Codex --> Review
Claude --> Review
Codex --> Refactor
Claude --> Refactor
Monitor --> Queue
Monitor --> Logs
Review --> UI
Refactor --> UI
Logs --> Monitor
The diagram above illustrates a job orchestration architecture where user commands flow into a scheduler, which enqueues tasks. The model router then dispatches work to either Codex or Claude based on task type, load balancing, or user preference. Background monitoring allows pausing, resuming, or canceling jobs mid-execution.
Key Features That Redefine Workflow Automation
1. Autonomous Delegated Tasks 🔄
The system does not just review code; it acts on findings. If Codex identifies a performance bottleneck, the engine can autonomously generate a refactored version and submit it as a suggestion—all in the background.
2. Background Job Control ⚙️
Jobs are not fire-and-forget. The engine provides full control:
- Pause a running review mid-stream
- Resume after API rate-limit recovery
- Cancel jobs that are no longer relevant
- Prioritize critical code paths over cosmetic fixes
3. Multi-Model Consensus 🤝
When Codex and Claude disagree on a refactoring approach, the orchestrator generates a diff report highlighting both options. You get the best of both models without manual intervention.
4. Rate-Limit Resilient Scheduling
The job scheduler implements exponential backoff and token bucket algorithms. Even during peak API usage, jobs complete—just slower. No dropped tasks.
5. Responsive Web UI & CLI Dual Interface
Access the system via a lightweight web dashboard (React-based) or from the command line. The UI updates in real-time using WebSockets, showing job progress, token usage, and model activity.
6. Multilingual Code Analysis
Supports Python, JavaScript, TypeScript, Go, Rust, Java, and C++. Each language has model-specific prompt templates optimized for Codex and Claude.
7. 24/7 Customer Support via Automated Retry
The system retries failed jobs up to five times with adaptive delays. Combined with job monitoring, this provides effectively around-the-clock reliability for overseas teams.
8. Memory-Aware Context Windows
Instead of blindly sending entire files, the engine uses tree-sitter to extract only relevant functions and classes. This keeps API costs low and responses fast.
9. Security-First Sandboxing
All generated code is executed in a read-only sandbox before being presented. No AI output touches your production environment without explicit approval.
10. 2026-Ready Token Accounting
The system tracks token usage per project, per user, and per job. This enables cost allocation and budgeting for enterprise deployments using OpenAI and Claude APIs.
Example Profile Configuration
Create a profile in ~/.codex-plugin-cc/config.yaml to define model preferences, job routing, and notification settings.
profiles:
default:
models:
review: claude-api
refactor: openai-codex
analyze: both
job_settings:
max_retries: 5
retry_backoff: exponential
timeout_minutes: 30
notifications:
on_complete: stdout
on_failure: webhook
language_preferences:
python: claude-api
javascript: openai-codex
go: both
security:
sandbox_enabled: true
auto_approve_refactor: false
This configuration tells the orchestrator to use Claude for review tasks, Codex for refactoring, and both for deep analysis. Python tasks route to Claude, while JavaScript goes to Codex.
Example Console Invocation
$ codex-plugin-cc review --path ./src --background --profile default
[INFO] Initializing job scheduler...
[INFO] Enqueued review job: JOB-2026-04-001
[INFO] Routing to Claude API for Python files...
[INFO] Routing to Codex API for JavaScript files...
[INFO] Job JOB-2026-04-001 is running in background (PID: 1423)
[INFO] Use `codex-plugin-cc status JOB-2026-04-001` to check progress.
[INFO] Use `codex-plugin-cc pause JOB-2026-04-001` to suspend execution.
$ codex-plugin-cc status JOB-2026-04-001
[STATUS] Job JOB-2026-04-001
State: Running
Progress: 45%
Files Reviewed: 12/26
Tokens Used: 14,230
Estimated Remaining: 3m 12s
Last Model Activity: Claude API responded with 3 issues
$ codex-plugin-cc control JOB-2026-04-001 --pause
[INFO] Job JOB-2026-04-001 paused. Use --resume to continue.
The console interface provides real-time job control reminiscent of background process management in operating systems, applied to AI-driven code analysis.
OS Compatibility Table
| Operating System | Supported | Notes |
|---|---|---|
| Linux (Ubuntu 22.04+) | ✅ | Fully tested |
| macOS (Monterey+) | ✅ | Requires Xcode CLI tools |
| Windows 10/11 | ✅ | WSL2 recommended |
| Windows Server 2022 | ✅ | PowerShell core support |
| FreeBSD | ⚠️ | Community support only |
| Alpine Linux | ❌ | Dependency conflicts |
The orchestrator uses Python 3.10+ and leverages asyncio for non-blocking job execution. All major operating systems are supported through either native or WSL2 environments.
API Integration Guide
OpenAI Codex API
The system interfaces with the Codex API using a custom session manager that handles:
- Token consumption tracking
- Prompt engineering for code review contexts
- Temperature control for deterministic vs. creative outputs
- Parallel requests for independent file analysis
Minimum configuration: Set OPENAI_API_KEY in environment variables. Optionally set OPENAI_ORG_ID for enterprise users.
Claude API
Claude integration uses Anthropic's messaging API with:
- Custom system prompts optimized for code review
- Context window management to prevent truncation
- Streaming responses for real-time UI updates
- Fallback to Codex if Claude is rate-limited
Minimum configuration: Set ANTHROPIC_API_KEY in environment variables.
Hybrid Mode
When both APIs are configured, the orchestrator can run a consensus workflow where both models analyze the same code and the output presents a merged review with attributed findings.
Multilingual Support & Responsive UI
The web dashboard is built with React and Tailwind CSS, supporting:
- Auto-detection of system language (English, Spanish, French, German, Japanese, Simplified Chinese)
- Dark/light theme with system preference detection
- Responsive layout for mobile, tablet, and desktop
- Real-time job cards that animate during execution
- Collapsible diff views for suggested changes
The CLI output is also i18n-aware, displaying error messages and status updates in the configured language.
Disclaimer
This software is provided as-is under the MIT License. The authors are not responsible for:
- Any API costs incurred when using OpenAI Codex or Claude APIs.
- Loss of data due to incorrect code execution or sandbox misconfiguration.
- Security vulnerabilities introduced by AI-generated code suggestions.
- Downtime caused by third-party API outages or rate limits.
- Misuse of background job control features in production environments.
Always review AI-generated code before merging into production branches. Use the sandbox mode for critical systems.
License
This project is licensed under the MIT License. See the LICENSE file for full terms.
AI Code Review Orchestrator 2026 – Turn your CI/CD pipeline into an autonomous intelligence layer.
Repository generated for demonstration purposes. The concept described is novel and distinct from the original "codex-plugin-cc" repository, focusing on background job orchestration and multi-model delegation rather than direct plugin integration.