Skip to main content

oh-my-claudecode-multi-agent-orchestration

Multi-agent orchestration for Claude Code with Team mode, autopilot, deep interview, and CLI workers

Zur Installation springen

Quellinformationen

Repository
reason-machines/claude-code-skills
Letzte Quellaktivität
16. Mai 2026 um 14:53
Erkannte Sprache von SKILL.md
Englisch
Sterne
4
Forks
1

Installationsoptionen

Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.

Quelldateien prüfen

Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.

SKILL.md wird angezeigt

SKILL.md
Quellanweisungen · Schreibgeschützte Vorschau
name
oh-my-claudecode-multi-agent-orchestration
description
Multi-agent orchestration for Claude Code with Team mode, autopilot, deep interview, and CLI workers
triggers
["set up oh-my-claudecode multi-agent system","orchestrate multiple Claude agents using omc","run team mode with specialized agents","use autopilot for autonomous development","start deep interview for requirements gathering","spawn codex or gemini CLI workers"]
# oh-my-claudecode Multi-Agent Orchestration > Skill by [ara.so](https://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. ## Installation ### Plugin Installation (Recommended) 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-claudecode ``` Then run setup: ```bash /setup ``` ### NPM Installation (CLI Runtime) ```bash npm i -g oh-my-claude-sisyphus@latest ``` Then run setup: ```bash omc setup ``` **Plugin Directory Mode**: If using `--plugin-dir`, add `--plugin-dir-mode` to setup: ```bash omc setup --plugin-dir-mode # or export OMC_PLUGIN_ROOT=/path/to/plugins omc setup ``` ## Core Concepts ### Execution Surfaces OMC operates on two surfaces: 1. **In-session skills** (`/...`) - Run inside Claude Code after plugin install 2. **Terminal CLI** (`omc ...`) - Run from shell after npm install ### Enable Native Teams Add to `~/.claude/settings.json`: ```json { "env": { "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1" } } ``` ## Team Mode (Canonical Orchestration) Team is the primary multi-agent surface. It runs as a staged pipeline: ``` team-plan → team-prd → team-exec → team-verify → team-fix (loop) ``` ### In-Session Team Usage ```bash # Fix TypeScript errors with 3 executor agents /team 3:executor "fix all TypeScript errors" # Review security with 2 reviewer agents /team 2:reviewer "audit authentication flow" # General development task /team 4:developer "implement user profile feature" ``` ### CLI Team Workers (tmux) Spawn real CLI workers in tmux panes: ```bash # 2 Codex CLI workers for code review omc team 2:codex "review auth module for security issues" # 2 Gemini CLI workers for UI design omc team 2:gemini "redesign UI components for accessibility" # 1 Claude CLI worker omc team 1:claude "implement the payment flow" # Check worker status omc team status auth-review # Shutdown workers omc team shutdown auth-review ``` **Requirements**: Requires `codex`/`gemini` CLIs installed and active tmux session. ## Deep Interview (Requirements Gathering) 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 ### Deep Interview with Autoresearch ```bash /deep-interview --autoresearch "build an e-commerce platform" ``` Combines requirements gathering with automated research. ## Autopilot (Autonomous Execution) Single-agent autonomous development: ```bash # In-session /autopilot "build a REST API for managing tasks" # Natural language shortcut 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 recovery ## Multi-Model Advisory (CCG) Combine Codex and Gemini expertise with Claude synthesis: ```bash /ccg Review this PR — architecture (Codex) and UI components (Gemini) ``` This routes through `/ask codex` and `/ask gemini`, then Claude synthesizes the combined feedback. ### Direct Provider Queries ```bash # Ask Codex specifically /ask codex "review this authentication implementation" # Ask Gemini specifically /ask gemini "suggest improvements to this component design" ``` ## Configuration ### Settings Location - Global: `~/.claude/settings.json` - Project: `.claude/settings.json` ### Key Environment Variables ```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 ### Custom Agent Configuration ```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 ```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 ```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 }); } }; ``` ## Common Workflows ### Full Feature Development ```bash # 1. Clarify requirements /deep-interview "add user authentication" # 2. Execute with team /team 4:developer "implement OAuth2 authentication based on requirements doc" # 3. Review with specialized agents /team 2:reviewer "security audit authentication implementation" ``` ### Code Review Pipeline ```bash # Use CLI workers for multi-model review 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 /ccg Complete code review covering architecture, security, and UX ``` ### Refactoring Large Codebase ```bash # Deep interview for strategy /deep-interview "refactor monolith to microservices" # Parallel execution with team /team 5:executor "implement microservices architecture per plan" # Continuous verification /team 2:validator "verify service boundaries and API contracts" ``` ### Research-Driven Development ```bash # Start with autoresearch /deep-interview --autoresearch "build real-time collaborative editor" # Execute with findings /autopilot "implement collaborative editor using CRDTs based on research" ``` ## CLI Reference ### Setup Commands ```bash # Basic setup omc setup # Setup with plugin directory mode omc setup --plugin-dir-mode # Force reinstall omc setup --force ``` ### Team Commands ```bash # Spawn team workers omc team <N>:<role> "<task>" # Worker status omc team status <task-id> # Shutdown workers omc team shutdown <task-id> # List all active teams omc team list ``` ### Diagnostic Commands ```bash # Run diagnostics /omc-doctor # Check plugin status /plugin list # Update marketplace plugin /plugin marketplace update omc ``` ## Troubleshooting ### Teams Not Working **Issue**: Team commands fail or warn about teams being disabled. **Solution**: Enable experimental teams in settings: ```json { "env": { "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1" } } ``` ### CLI Workers Not Spawning **Issue**: `omc team N:codex` fails. **Solutions**: 1. Verify tmux is running: `tmux ls` 2. Check CLI tools installed: `which codex` / `which gemini` 3. Ensure API keys configured in environment ### Plugin Not Found After Install **Issue**: `/setup` or skills not available. **Solutions**: ```bash # Clear plugin cache /omc-doctor # Reinstall plugin /plugin marketplace update omc /setup ``` ### Duplicate Skills/Agents **Issue**: Skills appear twice or setup duplicates resources. **Solution**: Use `--plugin-dir-mode` when running from custom plugin directory: ```bash export OMC_PLUGIN_ROOT=/path/to/plugins omc setup --plugin-dir-mode ``` ### npm Warning: deprecated prebuild-install **Issue**: Warning during `npm i -g oh-my-claude-sisyphus`. **Status**: Known upstream dependency warning from `better-sqlite3 → prebuild-install@7.1.3`. Tracked in issue #2913. Does not indicate install failure. ### Package Name Confusion **Issue**: Repository is `oh-my-claudecode` but npm package is different. **Clarification**: - Repository/plugin: `oh-my-claudecode` - npm package: `oh-my-claude-sisyphus` Always use `oh-my-claude-sisyphus` for npm commands: ```bash npm i -g oh-my-claude-sisyphus@latest ``` ## Advanced Patterns ### Combining Multiple Orchestration Modes ```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 }); } ``` ### Custom Provider Integration ```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"] ); ``` ### Persistent Task Management ```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; } } ``` ## Best Practices 1. **Start with Deep Interview** for unclear requirements 2. **Use Team mode** for coordinated multi-agent work 3. **Use CLI workers** (`omc team N:codex/gemini`) for provider-specific tasks 4. **Enable native teams** in settings before using `/team` 5. **Use CCG** for mixed backend/frontend review work 6. **Keep plugin updated** with `/plugin marketplace update omc` 7. **Use `--plugin-dir-mode`** if running with custom plugin directory 8. **Set environment variables** properly for API access 9. **Run `/omc-doctor`** when troubleshooting 10. **Check tmux session** before spawning CLI workers ## Resources - 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
Auf GitHub ansehen