Skip to main content

🎭 Multi-Agent Orchestration

Learn how to combine multiple agents into powerful workflows, like conducting an expert team working in concert.

Why Multi-Agent?​

A single agent is like a solo expert, while multi-agent orchestration is like assembling a professional team:

  • πŸ—οΈ Planner creates the plan
  • πŸ§ͺ TDD Guide executes development
  • πŸ” Code Reviewer reviews quality
  • 🎯 E2E Runner verifies functionality

Orchestration Patterns​

Pattern 1: Sequential Orchestration​

Use case: Standard feature development workflow

# Sequential execution
/plan Implement shopping cart    # Step 1
/tdd --feature=cart              # Step 2
/code-review                     # Step 3
/e2e Test shopping flow          # Step 4

Pattern 2: Parallel Orchestration​

Use case: Multi-dimensional code review

Pattern 3: Conditional Orchestration​

Use case: Choosing different paths based on conditions

Practical Examples​

Example 1: Full Feature Development​

# 1. Requirements analysis and planning
/plan Implement a user authentication system

# After confirmation...

# 2. Architecture design (for complex features)
/architect Design auth architecture

# 3. TDD development
/tdd --feature="auth-login"
/tdd --feature="auth-register"
/tdd --feature="auth-logout"

# 4. Code review
/code-review src/auth/

# 5. Security audit
/security src/auth/

# 6. End-to-end testing
/e2e Test the complete auth flow

# 7. Performance check
/perf --module=auth

Example 2: Bug Fix Workflow​

# 1. Debug and analyze
/debug TypeError: Cannot read property 'user' of undefined

# 2. TDD fix (write reproduction test first)
/tdd --feature="fix-null-user"

# 3. Code review
/code-review

# 4. Regression testing
/e2e Test the affected flows

Example 3: Large-Scale Refactoring​

# 1. Architecture assessment
/architect Assess current architecture issues

# 2. Create a refactoring plan
/plan Refactor the data access layer

# 3. Parallel review of current code
# (Multiple agents working simultaneously)
/code-review src/repositories/
/security src/repositories/
/perf src/repositories/

# 4. Incremental refactoring
/tdd --feature="refactor-user-repo"
/tdd --feature="refactor-order-repo"

# 5. Comprehensive testing
/e2e Regression test all features

Best Practices​

1. Plan First​

# βœ… Always plan complex features first
/plan Feature description

# ❌ Jump straight into coding
Start writing code directly...

2. Review Promptly​

# βœ… Review code immediately after writing
/code-review

# ❌ Accumulate lots of code before reviewing
Write 1000 lines then review...

3. Automated Testing​

# βœ… TDD ensures quality
/tdd --feature="feature"

# ❌ Write tests after completing the feature
Finish feature then add tests...

Advanced Tips​

Parallel Task Execution​

For independent tasks, run multiple agents in parallel:

Launch 3 agents simultaneously:
1. Agent 1: Security review of auth module
2. Agent 2: Performance analysis of cache system
3. Agent 3: Type checking of utilities

Multi-Perspective Analysis​

For complex problems, use multiple roles for analysis:

  • πŸ‘¨β€πŸ’» Senior Engineer perspective: Code quality
  • πŸ”’ Security Expert perspective: Vulnerabilities
  • ⚑ Performance Expert perspective: Bottlenecks
  • πŸ“ Documentation Expert perspective: Maintainability

πŸ’‘ Tip: The key to multi-agent orchestration is clearly defining each agent's responsibility boundaries to avoid duplicate work!