🤖 OMX Multi-Model Scheduling Guide: Claude vs Codex

Version: v1.0 | Updated: 2026-04-30

Detailed explanation of the selection and scheduling strategies for Claude and Codex dual engines in OMX

📐 1. Architecture Overview

1.1 Dual Engine Architecture Diagram

┌──────────────────────────────────────────────────────────────┐
│                    🎯 OMX 编排层                              │
│              任务调度 + 状态管理                               │
└──────────────────────────────────────────────────────────────┘
                              ↓
         ┌─────────────────────────────────────────┐
         │                                         │
         ▼                                         ▼
┌─────────────────────┐         ┌─────────────────────┐
│   🟢 Codex 引擎      │   ◄►    │   🟠 Claude 引擎     │
│                     │  协作   │                     │
│   主运行环境         │         │   外部调用           │
│   gpt-5.5           │         │   Claude CLI        │
│                     │         │                     │
│   ✅ 可修改文件       │         │   ❌ 仅只读分析       │
│   ✅ 运行测试         │         │   ✅ 外部知识库       │
│   ✅ Team 并行       │         │   ✅ 最佳实践         │
│                     │         │                     │
│   omx               │         │   omx ask claude    │
│   omx team N:agent  │         │   --agent-prompt    │
└─────────────────────┘         └─────────────────────┘
                              ↓
         ┌─────────────────────────────────────────┐
         │        🤖 Agents 层                      │
         │  executor | debugger | architect        │
         │  explore | designer | test-engineer     │
         │  (每个 Agent 可配置不同模型和推理深度)    │
         └─────────────────────────────────────────┘

⚖️ 2. Core Differences Comparison

🎯 3. Task Routing Decision Flow

                    用户任务输入
                         ↓
                ┌────────────────┐
                │ 🤔 是否需要修改代码？│
                └────────────────┘
                         ↓
        ┌────────────────┴────────────────┐
        │                                 │
        ▼                                 ▼
   ┌────────┐                       ┌────────┐
   │是：需要 │                       │否：只读 │
   │  修改   │                       │  分析   │
   └────────┘                       └────────┘
        │                                 │
        ▼                                 ▼
   🟢 Codex 引擎                    🟠 Claude 引擎
        │                                 │
   ┌────┴────┐                      ┌────┴────┐
   │         │                      │         │
   ▼         ▼                      ▼         ▼
单文件    多文件并行              架构分析   代码审查
executor  omx team               architect  code-reviewer

📊 Scenario Decision Table

🔧 4. Codex Engine Detailed Explanation

4.1 Startup Methods

4.2 Agent Types and Complexity

4.3 Team Mode Detailed Explanation

用户任务
    ↓
omx team N:agent "description"
    ↓
┌─────────────────────────────────────┐
│  🎯 Leader (tmux 主 pane)           │
│  - 任务分发                         │
│  - 结果整合                         │
│  - 最终验证                         │
└─────────────────────────────────────┘
    ↓ 分发
┌──────────┬──────────┬──────────┬──────────┐
│ Worker 1 │ Worker 2 │ Worker 3 │ Worker N │
│(独立     │(独立     │(独立     │(独立     │
│worktree) │worktree) │worktree) │worktree) │
│ executor │ debugger │ designer │ test-eng │
└──────────┴──────────┴──────────┴──────────┘
    ↓ 各自完成
    ↓
Leader 整合 + 验证
    ↓
omx team await / shutdown

Team Command Quick Reference

🟠 5. Claude Engine Detailed Explanation

5.1 omx ask Command

5.2 Claude Agent Roles

🚀 6. Complete Workflow Examples

6.1 New Feature Development Full Process

🔴【阶段 1】需求分析 - Claude
omx ask claude --agent-prompt=analyst "分析用户认证模块需求"

🔴【阶段 2】架构设计 - Claude
omx ask claude --agent-prompt=architect "设计认证模块架构方案"

🔴【阶段 3】方案评审 - Claude
omx ask claude --agent-prompt=critic "评审架构方案的问题"

🟢【阶段 4】并行实现 - Codex Team
omx team 3:executor "实现 auth API" "实现前端组件" "编写单元测试"

🟢【阶段 5】调试修复 - Codex
omx --agent debugger "修复测试失败"

🔴【阶段 6】代码审查 - Claude
omx ask claude --agent-prompt=code-reviewer "审查认证模块代码"

🔴【阶段 7】安全审查 - Claude
omx ask claude --agent-prompt=security-reviewer "安全审查认证流程"

🟢【阶段 8】最终验证 - Codex
omx --agent verifier "验证功能完整性"

6.2 Bug Fix Flow

⚙️ 7. Configure Custom Agent Models

7.1 Edit Agent Configuration

# 查看当前 agent 配置
cat ~/.codex/agents/executor.toml | grep model

# 编辑特定 agent
omx agents edit executor

# 添加新 agent
omx agents add my-custom-agent --scope user

7.2 Agent TOML Configuration Example

# ~/.codex/agents/executor.toml

name = "executor"
description = "Code implementation, refactoring, feature work"

# 模型配置
model = "gpt-5.4"               # 可改为 gpt-5.5, gpt-5-spark 等
model_reasoning_effort = "medium"  # low | medium | high | xhigh

# 角色 posture
posture = "deep-worker"         # 执行优先姿态
model_class = "standard"        # standard | frontier

7.3 Model Suggestions for Different Complexity

✅ 8. Current Configuration Status

❓ 9. Frequently Asked Questions

A: omx ask claude is an external call mode; Claude cannot access Codex file operation tools. You must use Codex when code modification is needed.

A: Edit each Agent's TOML configuration file, setting different model and model_reasoning_effort values.

A: Use normal session for single-file tasks; use Team for multi-module parallel tasks; use Claude for external perspective analysis.

A: Claude has a broader external knowledge base, suitable for architecture reviews, best practice references, documentation lookups, and other scenarios requiring external perspectives.

🎯 Quick Decision Tree

需要修改代码？
├── 是 → Codex
│   ├── 单文件 → omx --agent executor
│   ├── 多文件 → omx team N:executor
│   └── 调试 → omx --agent debugger
│
└── 否 → Claude
    ├── 架构设计 → omx ask claude --agent-prompt=architect
    ├── 代码审查 → omx ask claude --agent-prompt=code-reviewer
    ├── 外部文档 → omx ask claude --agent-prompt=researcher
    └── 设计评审 → omx ask claude --agent-prompt=critic

💡 Tip: Choose the appropriate engine based on task characteristics to fully leverage the dual-engine advantages!