Scripts2026年5月31日·1 分钟阅读

Oh My Codex — Multi-Agent Orchestration for AI Coding

Add hooks, agent teams, HUDs, and collaborative workflows to Codex and other AI coding tools with a lightweight orchestration layer.

Agent 就绪

Agent 可直接安装

这个资产可安装;Agent 先选择当前运行时、检查安装计划,再运行匹配命令。

Native · 98/100策略:允许
Agent 入口
任意 MCP/CLI Agent
类型
Skill
安装
Single
信任
信任等级:Established
入口
Oh My Codex
直接安装命令
npx -y tokrepo@latest install c1bb500b-5ca7-11f1-9bc6-00163e2b0d79 --target codex

先 dry-run 确认安装计划,再运行此命令。

Introduction

Oh My Codex (OmX) is an open-source orchestration layer that extends AI coding agents with team coordination, hooks, and HUD overlays. It lets developers run multiple agents in parallel on the same project while maintaining visibility into what each agent is doing and how resources are being consumed.

What Oh My Codex Does

  • Wraps existing coding agents (Codex, Claude Code, others) with coordination hooks
  • Runs multiple agents concurrently with task partitioning and conflict avoidance
  • Provides a terminal HUD showing agent status, token usage, and file locks
  • Supports pre/post hooks for validation, testing, and commit policies
  • Enables agent-to-agent delegation for complex multi-step tasks

Architecture Overview

OmX runs as a lightweight TypeScript process that sits between the developer and one or more agent instances. It manages a shared state file that tracks which files each agent is working on, enforces locking rules, and aggregates output streams into a unified terminal view. Hooks are shell scripts or Node modules triggered on agent lifecycle events.

Self-Hosting & Configuration

  • Install globally via npm or run from source with Node.js 20+
  • Initialize a project with omx init to create the .omx config directory
  • Define agent team compositions in omx.config.ts
  • Set hook scripts for pre-commit, post-task, and error-recovery events
  • Configure token budgets and concurrency limits per agent role

Key Features

  • Zero-config default that works out of the box with single-agent setups
  • Terminal HUD with real-time token counters, file diffs, and progress bars
  • Agent teams with named roles (architect, implementer, reviewer, tester)
  • Hook system for injecting custom logic at any point in the agent lifecycle
  • Compatible with multiple agent runtimes through a plugin interface

Comparison with Similar Tools

  • Claude Code subagents — built into Claude Code; OmX adds cross-runtime orchestration and HUDs
  • Multica — full platform with web UI; OmX is a lightweight CLI-first layer
  • Aider — single-agent pair programming; OmX coordinates multiple agents
  • CrewAI — Python-based agent framework; OmX wraps existing CLI agents directly

FAQ

Q: Does it work with Claude Code? A: Yes. OmX can wrap Claude Code sessions and coordinate them with other agents.

Q: How are file conflicts handled? A: OmX uses file-level locks. If two agents need the same file, tasks are serialized automatically.

Q: Can I use it with a single agent? A: Yes. Even with one agent, the HUD and hook system provide useful observability.

Q: Is there a performance overhead? A: Minimal. OmX adds a thin coordination layer and does not intercept LLM API calls.

Sources

讨论

登录后参与讨论。
还没有评论,来写第一条吧。

相关资产