SKILL.md
$2b
// 2. Spawn specialized agents
mcp__claude-flow__agent_spawn({ type: "researcher", name: "Agent 1" })
// 3. Orchestrate tasks
mcp__claude-flow__task_orchestrate({ task: "...", strategy: "parallel" })
## Core Concepts
### Swarm Topologies
**Mesh Topology** - Peer-to-peer communication, best for research and analysis
- All agents communicate directly
- High flexibility and resilience
- Use for: Research, analysis, brainstorming
**Hierarchical Topology** - Coordinator with subordinates, best for development
- Clear command structure
- Sequential workflow support
- Use for: Development, structured workflows
**Star Topology** - Central coordinator, best for testing
- Centralized control and monitoring
- Parallel execution with coordination
- Use for: Testing, validation, quality assurance
**Ring Topology** - Sequential processing chain
- Step-by-step processing
- Pipeline workflows
- Use for: Multi-stage processing, data pipelines
### Agent Strategies
**Adaptive** - Dynamic adjustment based on task complexity
**Balanced** - Equal distribution of work across agents
**Specialized** - Task-specific agent assignment
**Parallel** - Maximum concurrent execution
## Pattern 1: Research Swarm
### Purpose
Deep research through parallel information gathering, analysis, and synthesis.
### Architecture
// Initialize research swarm
mcp__claude-flow__swarm_init({
"topology": "mesh",
"maxAgents": 6,
"strategy": "adaptive"
})
// Spawn research team
const researchAgents = [
{
type: "researcher",
name: "Web Researcher",
capabilities: ["web-search", "content-extraction", "source-validation"]
},
{
type: "researcher",
name: "Academic Researcher",
capabilities: ["paper-analysis", "citation-tracking", "literature-review"]
},
{
type: "analyst",
name: "Data Analyst",
capabilities: ["data-processing", "statistical-analysis", "visualization"]
},
{
type: "analyst",
name: "Pattern Analyzer",
capabilities: ["trend-detection", "correlation-analysis", "outlier-detection"]
},
{
type: "documenter",
name: "Report Writer",
capabilities: ["synthesis", "technical-writing", "formatting"]
}
]
// Spawn all agents
researchAgents.forEach(agent => {
mcp__claude-flow__agent_spawn({
type: agent.type,
name: agent.name,
capabilities: agent.capabilities
})
})
### Research Workflow
#### Phase 1: Information Gathering
// Parallel information collection
mcp__claude-flow__parallel_execute({
"tasks": [
{
"id": "web-search",
"command": "search recent publications and articles"
},
{
"id": "academic-search",
"command": "search academic databases and papers"
},
{
"id": "data-collection",
"command": "gather relevant datasets and statistics"
},
{
"id": "expert-search",
"command": "identify domain experts and thought leaders"
}
]
})
// Store research findings in memory
mcp__claude-flow__memory_usage({
"action": "store",
"key": "research-findings-" + Date.now(),
"value": JSON.stringify(findings),
"namespace": "research",
"ttl": 604800 // 7 days
})
#### Phase 2: Analysis and Validation
// Pattern recognition in findings
mcp__claude-flow__pattern_recognize({
"data": researchData,
"patterns": ["trend", "correlation", "outlier", "emerging-pattern"]
})
// Cognitive analysis
mcp__claude-flow__cognitive_analyze({
"behavior": "research-synthesis"
})
// Quality assessment
mcp__claude-flow__quality_assess({
"target": "research-sources",
"criteria": ["credibility", "relevance", "recency", "authority"]
})
// Cross-reference validation
mcp__claude-flow__neural_patterns({
"action": "analyze",
"operation": "fact-checking",
"metadata": { "sources": sourcesArray }
})
#### Phase 3: Knowledge Management
// Search existing knowledge base
mcp__claude-flow__memory_search({
"pattern": "topic X",
"namespace": "research",
"limit": 20
})
// Create knowledge graph connections
mcp__claude-flow__neural_patterns({
"action": "learn",
"operation": "knowledge-graph",
"metadata": {
"topic": "X",
"connections": relatedTopics,
"depth": 3
}
})
// Store connections for future use
mcp__claude-flow__memory_usage({
"action": "store",
"key": "knowledge-graph-X",
"value": JSON.stringify(knowledgeGraph),
"namespace": "research$graphs",
"ttl": 2592000 // 30 days
})
#### Phase 4: Report Generation
// Orchestrate report generation
mcp__claude-flow__task_orchestrate({
"task": "generate comprehensive research report",
"strategy": "sequential",
"priority": "high",
"dependencies": ["gather", "analyze", "validate", "synthesize"]
})
// Monitor research progress
mcp__claude-flow__swarm_status({
"swarmId": "research-swarm"
})
// Generate final report
mcp__claude-flow__workflow_execute({
"workflowId": "research-report-generation",
"params": {
"findings": findings,
"format": "comprehensive",
"sections": ["executive-summary", "methodology", "findings", "analysis", "conclusions", "references"]
}
})
### CLI Fallback
Quick research swarm
npx claude-flow swarm "research AI trends in 2025" \
--strategy research \
--mode distributed \
--max-agents 6 \
--parallel \
--output research-report.md
## Pattern 2: Development Swarm
### Purpose
Full-stack development through coordinated specialist agents.
### Architecture
// Initialize development swarm with hierarchy
mcp__claude-flow__swarm_init({
"topology": "hierarchical",
"maxAgents": 8,
"strategy": "balanced"
})
// Spawn development team
const devTeam = [
{ type: "architect", name: "System Architect", role: "coordinator" },
{ type: "coder", name: "Backend Developer", capabilities: ["node", "api", "database"] },
{ type: "coder", name: "Frontend Developer", capabilities: ["react", "ui", "ux"] },
{ type: "coder", name: "Database Engineer", capabilities: ["sql", "nosql", "optimization"] },
{ type: "tester", name: "QA Engineer", capabilities: ["unit", "integration", "e2e"] },
{ type: "reviewer", name: "Code Reviewer", capabilities: ["security", "performance", "best-practices"] },
{ type: "documenter", name: "Technical Writer", capabilities: ["api-docs", "guides", "tutorials"] },
{ type: "monitor", name: "DevOps Engineer", capabilities: ["ci-cd", "deployment", "monitoring"] }
]
// Spawn all team members
devTeam.forEach(member => {
mcp__claude-flow__agent_spawn({
type: member.type,
name: member.name,
capabilities: member.capabilities,
swarmId: "dev-swarm"
})
})
### Development Workflow
#### Phase 1: Architecture and Design
// System architecture design
mcp__claude-flow__task_orchestrate({
"task": "design system architecture for REST API",
"strategy": "sequential",
"priority": "critical",
"assignTo": "System Architect"
})
// Store architecture decisions
mcp__claude-flow__memory_usage({
"action": "store",
"key": "architecture-decisions",
"value": JSON.stringify(architectureDoc),
"namespace": "development$design"
})
#### Phase 2: Parallel Implementation
// Parallel development tasks
mcp__claude-flow__parallel_execute({
"tasks": [
{
"id": "backend-api",
"command": "implement REST API endpoints",
"assignTo": "Backend Developer"
},
{
"id": "frontend-ui",
"command": "build user interface components",
"assignTo": "Frontend Developer"
},
{
"id": "database-schema",
"command": "design and implement database schema",
"assignTo": "Database Engineer"
},
{
"id": "api-documentation",
"command": "create API documentation",
"assignTo": "Technical Writer"
}
]
})
// Monitor development progress
mcp__claude-flow__swarm_monitor({
"swarmId": "dev-swarm",
"interval": 5000
})
#### Phase 3: Testing and Validation
// Comprehensive testing
mcp__claude-flow__batch_process({
"items": [
{ type: "unit", target: "all-modules" },
{ type: "integration", target: "api-endpoints" },
{ type: "e2e", target: "user-flows" },
{ type: "performance", target: "critical-paths" }
],
"operation": "execute-tests"
})
// Quality assessment
mcp__claude-flow__quality_assess({
"target": "codebase",
"criteria": ["coverage", "complexity", "maintainability", "security"]
})
#### Phase 4: Review and Deployment
// Code review workflow
mcp__claude-flow__workflow_execute({
"workflowId": "code-review-process",
"params": {
"reviewers": ["Code Reviewer"],
"criteria": ["security", "performance", "best-practices"]
}
})
// CI/CD pipeline
mcp__claude-flow__pipeline_create({
"config": {
"stages": ["build", "test", "security-scan", "deploy"],
"environment": "production"
}
})
### CLI Fallback
Quick development swarm
npx claude-flow swarm "build REST API with authentication" \
--strategy development \
--mode hierarchical \
--monitor \
--output sqlite
## Pattern 3: Testing Swarm
### Purpose
Comprehensive quality assurance through distributed testing.
### Architecture
// Initialize testing swarm with star topology
mcp__claude-flow__swarm_init({
"topology": "star",
"maxAgents": 7,
"strategy": "parallel"
})
// Spawn testing team
const testingTeam = [
{
type: "tester",
name: "Unit Test Coordinator",
capabilities: ["unit-testing", "mocking", "coverage", "tdd"]
},
{
type: "tester",
name: "Integration Tester",
capabilities: ["integration", "api-testing", "contract-testing"]
},
{
type: "tester",
name: "E2E Tester",
capabilities: ["e2e", "ui-testing", "user-flows", "selenium"]
},
{
type: "tester",
name: "Performance Tester",
capabilities: ["load-testing", "stress-testing", "benchmarking"]
},
{
type: "monitor",
name: "Security Tester",
capabilities: ["security-testing", "penetration-testing", "vulnerability-scanning"]
},
{
type: "analyst",
name: "Test Analyst",
capabilities: ["coverage-analysis", "test-optimization", "reporting"]
},
{
type: "documenter",
name: "Test Documenter",
capabilities: ["test-documentation", "test-plans", "reports"]
}
]
// Spawn all testers
testingTeam.forEach(tester => {
mcp__claude-flow__agent_spawn({
type: tester.type,
name: tester.name,
capabilities: tester.capabilities,
swarmId: "testing-swarm"
})
})
### Testing Workflow
#### Phase 1: Test Planning
// Analyze test coverage requirements
mcp__claude-flow__quality_assess({
"target": "test-coverage",
"criteria": [
"line-coverage",
"branch-coverage",
"function-coverage",
"edge-cases"
]
})
// Identify test scenarios
mcp__claude-flow__pattern_recognize({
"data": testScenarios,
"patterns": [
"edge-case",
"boundary-condition",
"error-path",
"happy-path"
]
})
// Store test plan
mcp__claude-flow__memory_usage({
"action": "store",
"key": "test-plan-" + Date.now(),
"value": JSON.stringify(testPlan),
"namespace": "testing$plans"
})
#### Phase 2: Parallel Test Execution
// Execute all test suites in parallel
mcp__claude-flow__parallel_execute({
"tasks": [
{
"id": "unit-tests",
"command": "npm run test:unit",
"assignTo": "Unit Test Coordinator"
},
{
"id": "integration-tests",
"command": "npm run test:integration",
"assignTo": "Integration Tester"
},
{
"id": "e2e-tests",
"command": "npm run test:e2e",
"assignTo": "E2E Tester"
},
{
"id": "performance-tests",
"command": "npm run test:performance",
"assignTo": "Performance Tester"
},
{
"id": "security-tests",
"command": "npm run test:security",
"assignTo": "Security Tester"
}
]
})
// Batch process test suites
mcp__claude-flow__batch_process({
"items": testSuites,
"operation": "execute-test-suite"
})
#### Phase 3: Performance and Security
// Run performance benchmarks
mcp__claude-flow__benchmark_run({
"suite": "comprehensive-performance"
})
// Bottleneck analysis
mcp__claude-flow__bottleneck_analyze({
"component": "application",
"metrics": ["response-time", "throughput", "memory", "cpu"]
})
// Security scanning
mcp__claude-flow__security_scan({
"target": "application",
"depth": "comprehensive"
})
// Vulnerability analysis
mcp__claude-flow__error_analysis({
"logs": securityScanLogs
})
#### Phase 4: Monitoring and Reporting
// Real-time test monitoring
mcp__claude-flow__swarm_monitor({
"swarmId": "testing-swarm",
"interval": 2000
})
// Generate comprehensive test report
mcp__claude-flow__performance_report({
"format": "detailed",
"timeframe": "current-run"
})
// Get test results
mcp__claude-flow__task_results({
"taskId": "test-execution-001"
})
// Trend analysis
mcp__claude-flow__trend_analysis({
"metric": "test-coverage",
"period": "30d"
})
### CLI Fallback
Quick testing swarm
npx claude-flow swarm "test application comprehensively" \
--strategy testing \
--mode star \
--parallel \
--timeout 600
## Pattern 4: Analysis Swarm
### Purpose
Deep code and system analysis through specialized analyzers.
### Architecture
// Initialize analysis swarm
mcp__claude-flow__swarm_init({
"topology": "mesh",
"maxAgents": 5,
"strategy": "adaptive"
})
// Spawn analysis specialists
const analysisTeam = [
{
type: "analyst",
name: "Code Analyzer",
capabilities: ["static-analysis", "complexity-analysis", "dead-code-detection"]
},
{
type: "analyst",
name: "Security Analyzer",
capabilities: ["security-scan", "vulnerability-detection", "dependency-audit"]
},
{
type: "analyst",
name: "Performance Analyzer",
capabilities: ["profiling", "bottleneck-detection", "optimization"]
},
{
type: "analyst",
name: "Architecture Analyzer",
capabilities: ["dependency-analysis", "coupling-detection", "modularity-assessment"]
},
{
type: "documenter",
name: "Analysis Reporter",
capabilities: ["reporting", "visualization", "recommendations"]
}
]
// Spawn all analysts
analysisTeam.forEach(analyst => {
mcp__claude-flow__agent_spawn({
type: analyst.type,
name: analyst.name,
capabilities: analyst.capabilities
})
})
### Analysis Workflow
// Parallel analysis execution
mcp__claude-flow__parallel_execute({
"tasks": [
{ "id": "analyze-code", "command": "analyze codebase structure and quality" },
{ "id": "analyze-security", "command": "scan for security vulnerabilities" },
{ "id": "analyze-performance", "command": "identify performance bottlenecks" },
{ "id": "analyze-architecture", "command": "assess architectural patterns" }
]
})
// Generate comprehensive analysis report
mcp__claude-flow__performance_report({
"format": "detailed",
"timeframe": "current"
})
// Cost analysis
mcp__claude-flow__cost_analysis({
"timeframe": "30d"
})
## Advanced Techniques
### Error Handling and Fault Tolerance
// Setup fault tolerance for all agents
mcp__claude-flow__daa_fault_tolerance({
"agentId": "all",
"strategy": "auto-recovery"
})
// Error handling pattern
try {
await mcp__claude-flow__task_orchestrate({
"task": "complex operation",
"strategy": "parallel",
"priority": "high"
})
} catch (error) {
// Check swarm health
const status = await mcp__claude-flow__swarm_status({})
// Analyze error patterns
await mcp__claude-flow__error_analysis({
"logs": [error.message]
})
// Auto-recovery attempt
if (status.healthy) {
await mcp__claude-flow__task_orchestrate({
"task": "retry failed operation",
"strategy": "sequential"
})
}
}
### Memory and State Management
// Cross-session persistence
mcp__claude-flow__memory_persist({
"sessionId": "swarm-session-001"
})
// Namespace management for different swarms
mcp__claude-flow__memory_namespace({
"namespace": "research-swarm",
"action": "create"
})
// Create state snapshot
mcp__claude-flow__state_snapshot({
"name": "development-checkpoint-1"
})
// Restore from snapshot if needed
mcp__claude-flow__context_restore({
"snapshotId": "development-checkpoint-1"
})
// Backup memory stores
mcp__claude-flow__memory_backup({
"path": "$workspaces$claude-code-flow$backups$swarm-memory.json"
})
### Neural Pattern Learning
// Train neural patterns from successful workflows
mcp__claude-flow__neural_train({
"pattern_type": "coordination",
"training_data": JSON.stringify(successfulWorkflows),
"epochs": 50
})
// Adaptive learning from experience
mcp__claude-flow__learning_adapt({
"experience": {
"workflow": "research-to-report",
"success": true,
"duration": 3600,
"quality": 0.95
}
})
// Pattern recognition for optimization
mcp__claude-flow__pattern_recognize({
"data": workflowMetrics,
"patterns": ["bottleneck", "optimization-opportunity", "efficiency-gain"]
})
### Workflow Automation
// Create reusable workflow
mcp__claude-flow__workflow_create({
"name": "full-stack-development",
"steps": [
{ "phase": "design", "agents": ["architect"] },
{ "phase": "implement", "agents": ["backend-dev", "frontend-dev"], "parallel": true },
{ "phase": "test", "agents": ["tester", "security-tester"], "parallel": true },
{ "phase": "review", "agents": ["reviewer"] },
{ "phase": "deploy", "agents": ["devops"] }
],
"triggers": ["on-commit", "scheduled-daily"]
})
// Setup automation rules
mcp__claude-flow__automation_setup({
"rules": [
{
"trigger": "file-changed",
"pattern": "*.js",
"action": "run-tests"
},
{
"trigger": "PR-created",
"action": "code-review-swarm"
}
]
})
// Event-driven triggers
mcp__claude-flow__trigger_setup({
"events": ["code-commit", "PR-merge", "deployment"],
"actions": ["test", "analyze", "document"]
})
### Performance Optimization
// Topology optimization
mcp__claude-flow__topology_optimize({
"swarmId": "current-swarm"
})
// Load balancing
mcp__claude-flow__load_balance({
"swarmId": "development-swarm",
"tasks": taskQueue
})
// Agent coordination sync
mcp__claude-flow__coordination_sync({
"swarmId": "development-swarm"
})
// Auto-scaling
mcp__claude-flow__swarm_scale({
"swarmId": "development-swarm",
"targetSize": 12
})
### Monitoring and Metrics
// Real-time swarm monitoring
mcp__claude-flow__swarm_monitor({
"swarmId": "active-swarm",
"interval": 3000
})
// Collect comprehensive metrics
mcp__claude-flow__metrics_collect({
"components": ["agents", "tasks", "memory", "performance"]
})
// Health monitoring
mcp__claude-flow__health_check({
"components": ["swarm", "agents", "neural", "memory"]
})
// Usage statistics
mcp__claude-flow__usage_stats({
"component": "swarm-orchestration"
})
// Trend analysis
mcp__claude-flow__trend_analysis({
"metric": "agent-performance",
"period": "7d"
})
## Best Practices
### 1. Choosing the Right Topology
- **Mesh**: Research, brainstorming, collaborative analysis
- **Hierarchical**: Structured development, sequential workflows
- **Star**: Testing, validation, centralized coordination
- **Ring**: Pipeline processing, staged workflows
### 2. Agent Specialization
- Assign specific capabilities to each agent
- Avoid overlapping responsibilities
- Use coordination agents for complex workflows
- Leverage memory for agent communication
### 3. Parallel Execution
- Identify independent tasks for parallelization
- Use sequential execution for dependent tasks
- Monitor resource usage during parallel execution
- Implement proper error handling
### 4. Memory Management
- Use namespaces to organize memory
- Set appropriate TTL values
- Create regular backups
- Implement state snapshots for checkpoints
### 5. Monitoring and Optimization
- Monitor swarm health regularly
- Collect and analyze metrics
- Optimize topology based on performance
- Use neural patterns to learn from success
### 6. Error Recovery
- Implement fault tolerance strategies
- Use auto-recovery mechanisms
- Analyze error patterns
- Create fallback workflows
## Real-World Examples
### Example 1: AI Research Project
// Research AI trends, analyze findings, generate report
mcp__claude-flow__swarm_init({ topology: "mesh", maxAgents: 6 })
// Spawn: 2 researchers, 2 analysts, 1 synthesizer, 1 documenter
// Parallel gather → Analyze patterns → Synthesize → Report
### Example 2: Full-Stack Application
// Build complete web application with testing
mcp__claude-flow__swarm_init({ topology: "hierarchical", maxAgents: 8 })
// Spawn: 1 architect, 2 devs, 1 db engineer, 2 testers, 1 reviewer, 1 devops
// Design → Parallel implement → Test → Review → Deploy
### Example 3: Security Audit
// Comprehensive security analysis
mcp__claude-flow__swarm_init({ topology: "star", maxAgents: 5 })
// Spawn: 1 coordinator, 1 code analyzer, 1 security scanner, 1 penetration tester, 1 reporter
// Parallel scan → Vulnerability analysis → Penetration test → Report
### Example 4: Performance Optimization
// Identify and fix performance bottlenecks
mcp__claude-flow__swarm_init({ topology: "mesh", maxAgents: 4 })
// Spawn: 1 profiler, 1 bottleneck analyzer, 1 optimizer, 1 tester
// Profile → Identify bottlenecks → Optimize → Validate