oh-my-claudecode-multi-agent-orchestration
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
Chineseoh-my-claudecode Multi-Agent Orchestration
oh-my-claudecode 多Agent编排框架
Skill by ara.so — Claude Code Skills collection.
oh-my-claudecode (OMC) is a teams-first multi-agent orchestration framework for Claude Code. It enables parallel execution across specialized agents, autonomous development workflows, and intelligent requirements gathering with zero configuration required.
由ara.so开发的Skill — Claude Code Skills合集。
oh-my-claudecode(OMC)是一款面向团队的Claude Code多Agent编排框架。它无需任何配置即可实现跨专业Agent的并行执行、自主开发工作流以及智能需求收集。
Installation
安装方式
Plugin Installation (Recommended)
插件安装(推荐)
Install via Claude Code marketplace (one command at a time):
bash
/plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode
/plugin install oh-my-claudecodeThen run setup:
bash
/setup通过Claude Code应用商店安装(每次执行一条命令):
bash
/plugin marketplace add https://github.com/Yeachan-Heo/oh-my-claudecode
/plugin install oh-my-claudecode然后运行初始化:
bash
/setupNPM Installation (CLI Runtime)
NPM安装(CLI运行时)
bash
npm i -g oh-my-claude-sisyphus@latestThen run setup:
bash
omc setupPlugin Directory Mode: If using , add to setup:
--plugin-dir--plugin-dir-modebash
omc setup --plugin-dir-modebash
npm i -g oh-my-claude-sisyphus@latest然后运行初始化:
bash
omc setup插件目录模式:如果使用参数,初始化时需添加:
--plugin-dir--plugin-dir-modebash
omc setup --plugin-dir-modeor
或
export OMC_PLUGIN_ROOT=/path/to/plugins
omc setup
undefinedexport OMC_PLUGIN_ROOT=/path/to/plugins
omc setup
undefinedCore Concepts
核心概念
Execution Surfaces
执行场景
OMC operates on two surfaces:
- In-session skills () - Run inside Claude Code after plugin install
/... - Terminal CLI () - Run from shell after npm install
omc ...
OMC支持两种执行场景:
- 会话内技能()- 安装插件后在Claude Code内运行
/... - 终端CLI()- 安装npm包后在Shell中运行
omc ...
Enable Native Teams
启用原生团队功能
Add to :
~/.claude/settings.jsonjson
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
}
}在中添加配置:
~/.claude/settings.jsonjson
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
}
}Team Mode (Canonical Orchestration)
团队模式(标准编排)
Team is the primary multi-agent surface. It runs as a staged pipeline:
team-plan → team-prd → team-exec → team-verify → team-fix (loop)团队模式是主要的多Agent协作场景,它以阶段式流水线运行:
team-plan → team-prd → team-exec → team-verify → team-fix(循环)In-Session Team Usage
会话内团队使用方式
bash
undefinedbash
undefinedFix TypeScript errors with 3 executor agents
使用3个执行Agent修复TypeScript错误
/team 3:executor "fix all TypeScript errors"
/team 3:executor "fix all TypeScript errors"
Review security with 2 reviewer agents
使用2个评审Agent进行安全审查
/team 2:reviewer "audit authentication flow"
/team 2:reviewer "audit authentication flow"
General development task
通用开发任务
/team 4:developer "implement user profile feature"
undefined/team 4:developer "implement user profile feature"
undefinedCLI Team Workers (tmux)
CLI团队工作器(tmux)
Spawn real CLI workers in tmux panes:
bash
undefined在tmux面板中启动真实的CLI工作器:
bash
undefined2 Codex CLI workers for code review
2个Codex CLI工作器用于代码审查
omc team 2:codex "review auth module for security issues"
omc team 2:codex "review auth module for security issues"
2 Gemini CLI workers for UI design
2个Gemini CLI工作器用于UI设计
omc team 2:gemini "redesign UI components for accessibility"
omc team 2:gemini "redesign UI components for accessibility"
1 Claude CLI worker
1个Claude CLI工作器
omc team 1:claude "implement the payment flow"
omc team 1:claude "implement the payment flow"
Check worker status
查看工作器状态
omc team status auth-review
omc team status auth-review
Shutdown workers
关闭工作器
omc team shutdown auth-review
**Requirements**: Requires `codex`/`gemini` CLIs installed and active tmux session.omc team shutdown auth-review
**要求**:需要安装`codex`/`gemini` CLI工具,且存在活跃的tmux会话。Deep Interview (Requirements Gathering)
深度调研(需求收集)
Use Socratic questioning to clarify requirements before building:
bash
/deep-interview "I want to build a task management app"The interview:
- Uses weighted dimensions to measure clarity
- Exposes hidden assumptions
- Generates clear requirements document
- Prevents premature execution
使用苏格拉底式提问在开发前明确需求:
bash
/deep-interview "I want to build a task management app"深度调研特性:
- 使用加权维度衡量需求清晰度
- 挖掘隐藏假设
- 生成清晰的需求文档
- 避免过早执行开发
Deep Interview with Autoresearch
带自动调研的深度调研
bash
/deep-interview --autoresearch "build an e-commerce platform"Combines requirements gathering with automated research.
bash
/deep-interview --autoresearch "build an e-commerce platform"结合需求收集与自动化调研功能。
Autopilot (Autonomous Execution)
自主执行模式(Autopilot)
Single-agent autonomous development:
bash
undefined单Agent自主开发:
bash
undefinedIn-session
会话内使用
/autopilot "build a REST API for managing tasks"
/autopilot "build a REST API for managing tasks"
Natural language shortcut
自然语言快捷指令
autopilot: build a REST API for managing tasks
Autopilot characteristics:
- Single lead agent (not multi-agent like Team)
- Autonomous decision-making
- Persistent execution until verified complete
- Automatic error recoveryautopilot: build a REST API for managing tasks
自主执行特性:
- 单个主导Agent(不同于团队模式的多Agent)
- 自主决策
- 持续执行直到验证完成
- 自动错误恢复Multi-Model Advisory (CCG)
多模型咨询(CCG)
Combine Codex and Gemini expertise with Claude synthesis:
bash
/ccg Review this PR — architecture (Codex) and UI components (Gemini)This routes through and , then Claude synthesizes the combined feedback.
/ask codex/ask gemini结合Codex和Gemini的专业能力,由Claude进行综合分析:
bash
/ccg Review this PR — architecture (Codex) and UI components (Gemini)该指令会通过和获取反馈,再由Claude综合生成结果。
/ask codex/ask geminiDirect Provider Queries
直接调用服务商查询
bash
undefinedbash
undefinedAsk Codex specifically
专门咨询Codex
/ask codex "review this authentication implementation"
/ask codex "review this authentication implementation"
Ask Gemini specifically
专门咨询Gemini
/ask gemini "suggest improvements to this component design"
undefined/ask gemini "suggest improvements to this component design"
undefinedConfiguration
配置说明
Settings Location
设置文件位置
- Global:
~/.claude/settings.json - Project:
.claude/settings.json
- 全局配置:
~/.claude/settings.json - 项目配置:
.claude/settings.json
Key Environment Variables
关键环境变量
json
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1",
"OMC_PLUGIN_ROOT": "/path/to/plugins",
"ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY}",
"OPENAI_API_KEY": "${OPENAI_API_KEY}",
"GEMINI_API_KEY": "${GEMINI_API_KEY}"
}
}json
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1",
"OMC_PLUGIN_ROOT": "/path/to/plugins",
"ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY}",
"OPENAI_API_KEY": "${OPENAI_API_KEY}",
"GEMINI_API_KEY": "${GEMINI_API_KEY}"
}
}TypeScript/JavaScript Usage Patterns
TypeScript/JavaScript使用模式
Custom Agent Configuration
自定义Agent配置
typescript
// agents/custom-reviewer.ts
export interface AgentConfig {
role: string;
capabilities: string[];
model: string;
}
export const customReviewer: AgentConfig = {
role: "security-reviewer",
capabilities: [
"security-audit",
"vulnerability-detection",
"code-review"
],
model: "claude-sonnet-4"
};typescript
// agents/custom-reviewer.ts
export interface AgentConfig {
role: string;
capabilities: string[];
model: string;
}
export const customReviewer: AgentConfig = {
role: "security-reviewer",
capabilities: [
"security-audit",
"vulnerability-detection",
"code-review"
],
model: "claude-sonnet-4"
};Team Workflow Integration
团队工作流集成
typescript
// workflows/deployment-pipeline.ts
import { TeamOrchestrator } from 'oh-my-claude-sisyphus';
async function runDeploymentPipeline() {
const orchestrator = new TeamOrchestrator({
agents: 3,
role: 'executor',
task: 'run full deployment pipeline'
});
await orchestrator.execute();
const results = await orchestrator.verify();
if (!results.allPassed) {
await orchestrator.fix(results.failures);
}
}typescript
// workflows/deployment-pipeline.ts
import { TeamOrchestrator } from 'oh-my-claude-sisyphus';
async function runDeploymentPipeline() {
const orchestrator = new TeamOrchestrator({
agents: 3,
role: 'executor',
task: 'run full deployment pipeline'
});
await orchestrator.execute();
const results = await orchestrator.verify();
if (!results.allPassed) {
await orchestrator.fix(results.failures);
}
}Skill Extension Pattern
Skill扩展模式
typescript
// skills/custom-workflow.ts
export const customWorkflowSkill = {
name: 'custom-workflow',
description: 'Custom development workflow',
async execute(context: SkillContext) {
// 1. Requirements gathering
await context.runSkill('/deep-interview', {
prompt: context.userInput
});
// 2. Team execution
await context.runSkill('/team', {
agents: 3,
role: 'executor',
task: context.clarifiedRequirements
});
// 3. Verification
await context.runSkill('/verify', {
criteria: context.acceptanceCriteria
});
}
};typescript
// skills/custom-workflow.ts
export const customWorkflowSkill = {
name: 'custom-workflow',
description: 'Custom development workflow',
async execute(context: SkillContext) {
// 1. 需求收集
await context.runSkill('/deep-interview', {
prompt: context.userInput
});
// 2. 团队执行
await context.runSkill('/team', {
agents: 3,
role: 'executor',
task: context.clarifiedRequirements
});
// 3. 验证
await context.runSkill('/verify', {
criteria: context.acceptanceCriteria
});
}
};Common Workflows
常见工作流
Full Feature Development
完整功能开发
bash
undefinedbash
undefined1. Clarify requirements
1. 明确需求
/deep-interview "add user authentication"
/deep-interview "add user authentication"
2. Execute with team
2. 团队执行开发
/team 4:developer "implement OAuth2 authentication based on requirements doc"
/team 4:developer "implement OAuth2 authentication based on requirements doc"
3. Review with specialized agents
3. 专业Agent评审
/team 2:reviewer "security audit authentication implementation"
undefined/team 2:reviewer "security audit authentication implementation"
undefinedCode Review Pipeline
代码审查流水线
bash
undefinedbash
undefinedUse CLI workers for multi-model review
使用CLI工作器进行多模型评审
omc team 1:codex "review code architecture and design patterns"
omc team 1:gemini "review user experience and component structure"
omc team 1:codex "review code architecture and design patterns"
omc team 1:gemini "review user experience and component structure"
Or use tri-model advisory
或使用三模型咨询
/ccg Complete code review covering architecture, security, and UX
undefined/ccg Complete code review covering architecture, security, and UX
undefinedRefactoring Large Codebase
大型代码库重构
bash
undefinedbash
undefinedDeep interview for strategy
深度调研确定策略
/deep-interview "refactor monolith to microservices"
/deep-interview "refactor monolith to microservices"
Parallel execution with team
团队并行执行
/team 5:executor "implement microservices architecture per plan"
/team 5:executor "implement microservices architecture per plan"
Continuous verification
持续验证
/team 2:validator "verify service boundaries and API contracts"
undefined/team 2:validator "verify service boundaries and API contracts"
undefinedResearch-Driven Development
调研驱动开发
bash
undefinedbash
undefinedStart with autoresearch
先进行自动调研
/deep-interview --autoresearch "build real-time collaborative editor"
/deep-interview --autoresearch "build real-time collaborative editor"
Execute with findings
根据调研结果执行开发
/autopilot "implement collaborative editor using CRDTs based on research"
undefined/autopilot "implement collaborative editor using CRDTs based on research"
undefinedCLI Reference
CLI参考
Setup Commands
初始化命令
bash
undefinedbash
undefinedBasic setup
基础初始化
omc setup
omc setup
Setup with plugin directory mode
插件目录模式初始化
omc setup --plugin-dir-mode
omc setup --plugin-dir-mode
Force reinstall
强制重新安装
omc setup --force
undefinedomc setup --force
undefinedTeam Commands
团队命令
bash
undefinedbash
undefinedSpawn team workers
启动团队工作器
omc team <N>:<role> "<task>"
omc team <N>:<role> "<task>"
Worker status
查看工作器状态
omc team status <task-id>
omc team status <task-id>
Shutdown workers
关闭工作器
omc team shutdown <task-id>
omc team shutdown <task-id>
List all active teams
列出所有活跃团队
omc team list
undefinedomc team list
undefinedDiagnostic Commands
诊断命令
bash
undefinedbash
undefinedRun diagnostics
运行诊断
/omc-doctor
/omc-doctor
Check plugin status
检查插件状态
/plugin list
/plugin list
Update marketplace plugin
更新应用商店插件
/plugin marketplace update omc
undefined/plugin marketplace update omc
undefinedTroubleshooting
故障排除
Teams Not Working
团队功能无法使用
Issue: Team commands fail or warn about teams being disabled.
Solution: Enable experimental teams in settings:
json
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
}
}问题:团队命令执行失败或提示团队功能已禁用。
解决方案:在设置中启用实验性团队功能:
json
{
"env": {
"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
}
}CLI Workers Not Spawning
CLI工作器无法启动
Issue: fails.
omc team N:codexSolutions:
- Verify tmux is running:
tmux ls - Check CLI tools installed: /
which codexwhich gemini - Ensure API keys configured in environment
问题:执行失败。
omc team N:codex解决方案:
- 验证tmux是否运行:
tmux ls - 检查CLI工具是否安装:/
which codexwhich gemini - 确保环境中已配置API密钥
Plugin Not Found After Install
安装后插件找不到
Issue: or skills not available.
/setupSolutions:
bash
undefined问题:或技能无法使用。
/setup解决方案:
bash
undefinedClear plugin cache
清除插件缓存
/omc-doctor
/omc-doctor
Reinstall plugin
重新安装插件
/plugin marketplace update omc
/setup
undefined/plugin marketplace update omc
/setup
undefinedDuplicate Skills/Agents
重复的技能/Agent
Issue: Skills appear twice or setup duplicates resources.
Solution: Use when running from custom plugin directory:
--plugin-dir-modebash
export OMC_PLUGIN_ROOT=/path/to/plugins
omc setup --plugin-dir-mode问题:技能重复出现或初始化时重复创建资源。
解决方案:从自定义插件目录运行时使用:
--plugin-dir-modebash
export OMC_PLUGIN_ROOT=/path/to/plugins
omc setup --plugin-dir-modenpm Warning: deprecated prebuild-install
npm警告:deprecated prebuild-install
Issue: Warning during .
npm i -g oh-my-claude-sisyphusStatus: Known upstream dependency warning from . Tracked in issue #2913. Does not indicate install failure.
better-sqlite3 → prebuild-install@7.1.3问题:执行时出现警告。
npm i -g oh-my-claude-sisyphus状态:这是的上游依赖警告,已在Issue #2913中跟踪,不代表安装失败。
better-sqlite3 → prebuild-install@7.1.3Package Name Confusion
包名混淆
Issue: Repository is but npm package is different.
oh-my-claudecodeClarification:
- Repository/plugin:
oh-my-claudecode - npm package:
oh-my-claude-sisyphus
Always use for npm commands:
oh-my-claude-sisyphusbash
npm i -g oh-my-claude-sisyphus@latest问题:仓库名为但npm包名不同。
oh-my-claudecode说明:
- 仓库/插件:
oh-my-claudecode - npm包:
oh-my-claude-sisyphus
npm命令请始终使用:
oh-my-claude-sisyphusbash
npm i -g oh-my-claude-sisyphus@latestAdvanced Patterns
高级模式
Combining Multiple Orchestration Modes
组合多种编排模式
typescript
// Hybrid workflow: deep interview → team → autopilot cleanup
async function hybridWorkflow(task: string) {
// Phase 1: Requirements
const requirements = await runDeepInterview(task);
// Phase 2: Core implementation with team
await runTeam({
agents: 4,
role: 'executor',
task: requirements.implementation
});
// Phase 3: Polish with autopilot
await runAutopilot({
task: 'refine and optimize implementation',
context: requirements
});
}typescript
// 混合工作流:深度调研 → 团队开发 → 自主执行优化
async function hybridWorkflow(task: string) {
// 阶段1:需求收集
const requirements = await runDeepInterview(task);
// 阶段2:团队实现核心功能
await runTeam({
agents: 4,
role: 'executor',
task: requirements.implementation
});
// 阶段3:自主执行优化
await runAutopilot({
task: 'refine and optimize implementation',
context: requirements
});
}Custom Provider Integration
自定义服务商集成
typescript
// Add custom provider to advisory workflow
export async function consultProviders(
question: string,
providers: string[]
) {
const responses = await Promise.all(
providers.map(p => askProvider(p, question))
);
return synthesizeResponses(responses);
}
// Usage
const feedback = await consultProviders(
"Review this architecture",
["codex", "gemini", "claude"]
);typescript
// 在咨询工作流中添加自定义服务商
export async function consultProviders(
question: string,
providers: string[]
) {
const responses = await Promise.all(
providers.map(p => askProvider(p, question))
);
return synthesizeResponses(responses);
}
// 使用示例
const feedback = await consultProviders(
"Review this architecture",
["codex", "gemini", "claude"]
);Persistent Task Management
持久化任务管理
typescript
// Track long-running tasks across sessions
interface PersistentTask {
id: string;
status: 'planning' | 'executing' | 'verifying' | 'complete';
checkpoint: any;
}
export async function resumeTask(taskId: string) {
const task = await loadTask(taskId);
switch (task.status) {
case 'planning':
await runTeamPlan(task);
break;
case 'executing':
await runTeamExec(task);
break;
case 'verifying':
await runTeamVerify(task);
break;
}
}typescript
// 跨会话跟踪长期运行任务
interface PersistentTask {
id: string;
status: 'planning' | 'executing' | 'verifying' | 'complete';
checkpoint: any;
}
export async function resumeTask(taskId: string) {
const task = await loadTask(taskId);
switch (task.status) {
case 'planning':
await runTeamPlan(task);
break;
case 'executing':
await runTeamExec(task);
break;
case 'verifying':
await runTeamVerify(task);
break;
}
}Best Practices
最佳实践
- Start with Deep Interview for unclear requirements
- Use Team mode for coordinated multi-agent work
- Use CLI workers () for provider-specific tasks
omc team N:codex/gemini - Enable native teams in settings before using
/team - Use CCG for mixed backend/frontend review work
- Keep plugin updated with
/plugin marketplace update omc - Use if running with custom plugin directory
--plugin-dir-mode - Set environment variables properly for API access
- Run when troubleshooting
/omc-doctor - Check tmux session before spawning CLI workers
- 需求不明确时先使用深度调研
- 团队协作使用团队模式
- 服务商特定任务使用CLI工作器()
omc team N:codex/gemini - 使用前先启用原生团队功能
/team - 混合前后端评审使用CCG
- 定期更新插件()
/plugin marketplace update omc - 自定义插件目录运行时使用
--plugin-dir-mode - 正确设置环境变量以获取API访问权限
- 故障排除时运行
/omc-doctor - 启动CLI工作器前检查tmux会话
Resources
资源
- Homepage: https://oh-my-claudecode.dev
- Documentation: https://yeachan-heo.github.io/oh-my-claudecode-website
- Repository: https://github.com/Yeachan-Heo/oh-my-claudecode
- Discord: https://discord.gg/PUwSMR9XNk
- npm Package: https://www.npmjs.com/package/oh-my-claude-sisyphus
- Migration Guide: https://github.com/Yeachan-Heo/oh-my-claudecode/blob/main/docs/MIGRATION.md
- 主页:https://oh-my-claudecode.dev
- 文档:https://yeachan-heo.github.io/oh-my-claudecode-website
- 仓库:https://github.com/Yeachan-Heo/oh-my-claudecode
- Discord:https://discord.gg/PUwSMR9XNk
- npm包:https://www.npmjs.com/package/oh-my-claude-sisyphus
- 迁移指南:https://github.com/Yeachan-Heo/oh-my-claudecode/blob/main/docs/MIGRATION.md