Parallel Agents Orchestration

When to Use Parallel Agents

Use this pattern when you have multiple independent problems that:

• Don't share state or dependencies
• Can be understood without context from others
• Won't interfere with each other
• Would benefit from parallel investigation

Good candidates:

• Multiple test failures in different files
• Unrelated subsystems broken independently
• Research tasks across different domains
• Code reviews of separate modules

NOT good candidates:

• Failures that might be related (fix one, fix others)
• Tasks requiring sequential context
• Shared state modifications

The Parallel Dispatch Pattern

Step 1: Analyze and Group

Problem: 6 failing tests across 3 domains

Analysis:
├── Authentication (auth.test.ts)
│   ├── Login failure
│   └── Session expiry
├── Products API (products.test.ts)
│   ├── List filtering
│   └── Price calculation
└── UI Components (components.test.ts)
    ├── Modal rendering
    └── Form validation

These are INDEPENDENT - parallel dispatch appropriate.

Step 2: Create Focused Agent Tasks

Each agent gets:

1. Specific scope - One domain only
2. Full context - All info needed
3. Clear constraints - What NOT to change
4. Expected output - What to deliver

// Agent 1: Authentication
const authTask = {
  domain: "Authentication",
  files: ["src/auth/", "tests/auth.test.ts"],
  failures: ["Login failure", "Session expiry"],
  constraints: "Don't modify product or UI code",
  deliverable: "Fixed tests with explanation"
};

// Agent 2: Products API
const productsTask = {
  domain: "Products API",
  files: ["src/api/products/", "tests/products.test.ts"],
  failures: ["List filtering", "Price calculation"],
  constraints: "Don't modify auth or UI code",
  deliverable: "Fixed tests with explanation"
};

// Agent 3: UI Components
const uiTask = {
  domain: "UI Components",
  files: ["src/components/", "tests/components.test.ts"],
  failures: ["Modal rendering", "Form validation"],
  constraints: "Don't modify auth or API code",
  deliverable: "Fixed tests with explanation"
};

Step 3: Dispatch Concurrently

Using Claude Code's Task tool:

// In Claude Code, dispatch multiple agents in ONE message
// This runs them concurrently

<Task: "Fix auth tests" agent="code-fixer">
  Investigate and fix authentication test failures in tests/auth.test.ts.
  Focus ONLY on auth domain. Do not modify product or UI code.
  Return: Summary of root cause and fix applied.
</Task>

<Task: "Fix products tests" agent="code-fixer">
  Investigate and fix products API test failures in tests/products.test.ts.
  Focus ONLY on products domain. Do not modify auth or UI code.
  Return: Summary of root cause and fix applied.
</Task>

<Task: "Fix UI tests" agent="code-fixer">
  Investigate and fix UI component test failures in tests/components.test.ts.
  Focus ONLY on UI domain. Do not modify auth or API code.
  Return: Summary of root cause and fix applied.
</Task>

Step 4: Synthesize Results

After all agents complete:

1. Review summaries - Understand each fix
2. Check for conflicts - Any overlapping changes?
3. Run full test suite - Verify all fixes work together
4. Commit atomically - One commit per domain or all together

FrankX Orchestration Patterns

Content Pipeline Parallel Dispatch

Task: Publish new blog post

Parallel Agents:
├── SEO Agent
│   ├── Keyword optimization
│   ├── Meta description
│   └── Schema markup
├── Social Agent
│   ├── Twitter thread
│   ├── LinkedIn post
│   └── Instagram caption
└── Email Agent
    ├── Newsletter segment
    └── Welcome sequence update

Synthesis: Coordinate publish timing

Multi-Domain Research

Task: Research AI agent frameworks

Parallel Agents:
├── LangGraph Research
│   └── Patterns, examples, best practices
├── Claude SDK Research
│   └── Patterns, examples, best practices
├── OpenAI AgentKit Research
│   └── Patterns, examples, best practices
└── Oracle ADK Research
    └── Patterns, examples, best practices

Synthesis: Comparison matrix, recommendations

Codebase Analysis

Task: Comprehensive code review

Parallel Agents:
├── Security Analyst
│   └── Vulnerabilities, OWASP
├── Performance Analyst
│   └── Bottlenecks, optimizations
├── Accessibility Analyst
│   └── WCAG compliance
└── Code Quality Analyst
    └── Patterns, DRY, complexity

Synthesis: Prioritized action items

Agent Prompt Template

## Task: [Specific Domain Problem]

### Context
[All information the agent needs]

### Scope
- Files to examine: [list]
- Files to modify: [list]
- Domain boundaries: [what's in/out]

### Constraints
- DO NOT modify: [list of off-limits areas]
- DO NOT introduce: [breaking changes, new dependencies, etc.]
- MUST follow: [coding standards, patterns]

### Expected Output
1. Root cause analysis (2-3 sentences)
2. Fix applied (file paths and summary)
3. Verification steps taken
4. Any concerns or recommendations

### Success Criteria
- [Specific test passes]
- [No regressions]
- [Clean implementation]

Anti-Patterns

Bad Practice	Why It Fails
Vague scope	Agent changes wrong files
Missing context	Agent asks questions or guesses
No constraints	Agents conflict with each other
Coupled tasks	Agents duplicate work or conflict
No synthesis	Fixes don't integrate properly

Real Example

Input: 6 test failures across 3 files

Sequential approach: ~45 minutes
- Investigate each failure one by one
- Context switching between domains
- Waiting for each fix before next

Parallel approach: ~15 minutes
- 3 agents work simultaneously
- Each focuses on one domain
- Synthesis adds 5 minutes
- 3x faster overall

Result: 6 failures fixed, 0 conflicts, 3x speedup

When to Use This Skill

• Multiple independent failures to fix
• Research across several domains
• Code reviews of separate modules
• Content creation across platforms
• Any task naturally decomposable into parallel work

Related Skills

• implementation-planning - Plan parallel execution
• systematic-debugging - Each agent uses this
• test-driven-development - Verify fixes with tests