| name | performance-analysis |
| version | 1.0.0 |
| description | Comprehensive performance analysis, bottleneck detection, and optimization recommendations for Claude Flow swarms |
| category | monitoring |
| tags | performance, bottleneck, optimization, profiling, metrics, analysis |
| author | Claude Flow Team |
Performance Analysis Skill
Comprehensive performance analysis suite for identifying bottlenecks, profiling swarm operations, generating detailed reports, and providing actionable optimization recommendations.
Overview
This skill consolidates all performance analysis capabilities:
- Bottleneck Detection: Identify performance bottlenecks across communication, processing, memory, and network
- Performance Profiling: Real-time monitoring and historical analysis of swarm operations
- Report Generation: Create comprehensive performance reports in multiple formats
- Optimization Recommendations: AI-powered suggestions for improving performance
Quick Start
Basic Bottleneck Detection
npx claude-flow bottleneck detect
Generate Performance Report
npx claude-flow analysis performance-report --format html --include-metrics
Analyze and Auto-Fix
npx claude-flow bottleneck detect --fix --threshold 15
Core Capabilities
1. Bottleneck Detection
Command Syntax
npx claude-flow bottleneck detect [options]
Options
--swarm-id, -s <id>- Analyze specific swarm (default: current)--time-range, -t <range>- Analysis period: 1h, 24h, 7d, all (default: 1h)--threshold <percent>- Bottleneck threshold percentage (default: 20)--export, -e <file>- Export analysis to file--fix- Apply automatic optimizations
Usage Examples
# Basic detection for current swarm
npx claude-flow bottleneck detect
# Analyze specific swarm over 24 hours
npx claude-flow bottleneck detect --swarm-id swarm-123 -t 24h
# Export detailed analysis
npx claude-flow bottleneck detect -t 24h -e bottlenecks.json
# Auto-fix detected issues
npx claude-flow bottleneck detect --fix --threshold 15
# Low threshold for sensitive detection
npx claude-flow bottleneck detect --threshold 10 --export critical-issues.json
Metrics Analyzed
Communication Bottlenecks:
- Message queue delays
- Agent response times
- Coordination overhead
- Memory access patterns
- Inter-agent communication latency
Processing Bottlenecks:
- Task completion times
- Agent utilization rates
- Parallel execution efficiency
- Resource contention
- CPU/memory usage patterns
Memory Bottlenecks:
- Cache hit rates
- Memory access patterns
- Storage I/O performance
- Neural pattern loading times
- Memory allocation efficiency
Network Bottlenecks:
- API call latency
- MCP communication delays
- External service timeouts
- Concurrent request limits
- Network throughput issues
Output Format
🔍 Bottleneck Analysis Report
━━━━━━━━━━━━━━━━━━━━━━━━━━━
📊 Summary
├── Time Range: Last 1 hour
├── Agents Analyzed: 6
├── Tasks Processed: 42
└── Critical Issues: 2
🚨 Critical Bottlenecks
1. Agent Communication (35% impact)
└── coordinator → coder-1 messages delayed by 2.3s avg
2. Memory Access (28% impact)
└── Neural pattern loading taking 1.8s per access
⚠️ Warning Bottlenecks
1. Task Queue (18% impact)
└── 5 tasks waiting > 10s for assignment
💡 Recommendations
1. Switch to hierarchical topology (est. 40% improvement)
2. Enable memory caching (est. 25% improvement)
3. Increase agent concurrency to 8 (est. 20% improvement)
✅ Quick Fixes Available
Run with --fix to apply:
- Enable smart caching
- Optimize message routing
- Adjust agent priorities
2. Performance Profiling
Real-time Detection
Automatic analysis during task execution:
- Execution time vs. complexity
- Agent utilization rates
- Resource constraints
- Operation patterns
Common Bottleneck Patterns
Time Bottlenecks:
- Tasks taking > 5 minutes
- Sequential operations that could parallelize
- Redundant file operations
- Inefficient algorithm implementations
Coordination Bottlenecks:
- Single agent for complex tasks
- Unbalanced agent workloads
- Poor topology selection
- Excessive synchronization points
Resource Bottlenecks:
- High operation count (> 100)
- Memory constraints
- I/O limitations
- Thread pool saturation
MCP Integration
// Check for bottlenecks in Claude Code
mcp__claude-flow__bottleneck_detect({
timeRange: "1h",
threshold: 20,
autoFix: false
})
// Get detailed task results with bottleneck analysis
mcp__claude-flow__task_results({
taskId: "task-123",
format: "detailed"
})
Result Format:
{
"bottlenecks": [
{
"type": "coordination",
"severity": "high",
"description": "Single agent used for complex task",
"recommendation": "Spawn specialized agents for parallel work",
"impact": "35%",
"affectedComponents": ["coordinator", "coder-1"]
}
],
"improvements": [
{
"area": "execution_time",
"suggestion": "Use parallel task execution",
"expectedImprovement": "30-50% time reduction",
"implementationSteps": [
"Split task into smaller units",
"Spawn 3-4 specialized agents",
"Use mesh topology for coordination"
]
}
],
"metrics": {
"avgExecutionTime": "142s",
"agentUtilization": "67%",
"cacheHitRate": "82%",
"parallelizationFactor": 1.2
}
}
3. Report Generation
Command Syntax
npx claude-flow analysis performance-report [options]
Options
--format <type>- Report format: json, html, markdown (default: markdown)--include-metrics- Include detailed metrics and charts--compare <id>- Compare with previous swarm--time-range <range>- Analysis period: 1h, 24h, 7d, 30d, all--output <file>- Output file path--sections <list>- Comma-separated sections to include
Report Sections
Executive Summary
- Overall performance score
- Key metrics overview
- Critical findings
Swarm Overview
- Topology configuration
- Agent distribution
- Task statistics
Performance Metrics
- Execution times
- Throughput analysis
- Resource utilization
- Latency breakdown
Bottleneck Analysis
- Identified bottlenecks
- Impact assessment
- Optimization priorities
Comparative Analysis (when --compare used)
- Performance trends
- Improvement metrics
- Regression detection
Recommendations
- Prioritized action items
- Expected improvements
- Implementation guidance
Usage Examples
# Generate HTML report with all metrics
npx claude-flow analysis performance-report --format html --include-metrics
# Compare current swarm with previous
npx claude-flow analysis performance-report --compare swarm-123 --format markdown
# Custom output with specific sections
npx claude-flow analysis performance-report \
--sections summary,metrics,recommendations \
--output reports/perf-analysis.html \
--format html
# Weekly performance report
npx claude-flow analysis performance-report \
--time-range 7d \
--include-metrics \
--format markdown \
--output docs/weekly-performance.md
# JSON format for CI/CD integration
npx claude-flow analysis performance-report \
--format json \
--output build/performance.json
Sample Markdown Report
# Performance Analysis Report
## Executive Summary
- **Overall Score**: 87/100
- **Analysis Period**: Last 24 hours
- **Swarms Analyzed**: 3
- **Critical Issues**: 1
## Key Metrics
| Metric | Value | Trend | Target |
|--------|-------|-------|--------|
| Avg Task Time | 42s | ↓ 12% | 35s |
| Agent Utilization | 78% | ↑ 5% | 85% |
| Cache Hit Rate | 91% | → | 90% |
| Parallel Efficiency | 2.3x | ↑ 0.4x | 2.5x |
## Bottleneck Analysis
### Critical
1. **Agent Communication Delay** (Impact: 35%)
- Coordinator → Coder messages delayed by 2.3s avg
- **Fix**: Switch to hierarchical topology
### Warnings
1. **Memory Access Pattern** (Impact: 18%)
- Neural pattern loading: 1.8s per access
- **Fix**: Enable memory caching
## Recommendations
1. **High Priority**: Switch to hierarchical topology (40% improvement)
2. **Medium Priority**: Enable memory caching (25% improvement)
3. **Low Priority**: Increase agent concurrency to 8 (20% improvement)
4. Optimization Recommendations
Automatic Fixes
When using --fix, the following optimizations may be applied:
1. Topology Optimization
- Switch to more efficient topology (mesh → hierarchical)
- Adjust communication patterns
- Reduce coordination overhead
- Optimize message routing
2. Caching Enhancement
- Enable memory caching
- Optimize cache strategies
- Preload common patterns
- Implement cache warming
3. Concurrency Tuning
- Adjust agent counts
- Optimize parallel execution
- Balance workload distribution
- Implement load balancing
4. Priority Adjustment
- Reorder task queues
- Prioritize critical paths
- Reduce wait times
- Implement fair scheduling
5. Resource Optimization
- Optimize memory usage
- Reduce I/O operations
- Batch API calls
- Implement connection pooling
Performance Impact
Typical improvements after bottleneck resolution:
- Communication: 30-50% faster message delivery
- Processing: 20-40% reduced task completion time
- Memory: 40-60% fewer cache misses
- Network: 25-45% reduced API latency
- Overall: 25-45% total performance improvement
Advanced Usage
Continuous Monitoring
# Monitor performance in real-time
npx claude-flow swarm monitor --interval 5
# Generate hourly reports
while true; do
npx claude-flow analysis performance-report \
--format json \
--output logs/perf-$(date +%Y%m%d-%H%M).json
sleep 3600
done
CI/CD Integration
# .github/workflows/performance.yml
name: Performance Analysis
on: [push, pull_request]
jobs:
analyze:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v2
- name: Run Performance Analysis
run: |
npx claude-flow analysis performance-report \
--format json \
--output performance.json
- name: Check Performance Thresholds
run: |
npx claude-flow bottleneck detect \
--threshold 15 \
--export bottlenecks.json
- name: Upload Reports
uses: actions/upload-artifact@v2
with:
name: performance-reports
path: |
performance.json
bottlenecks.json
Custom Analysis Scripts
// scripts/analyze-performance.js
const { exec } = require('child_process');
const fs = require('fs');
async function analyzePerformance() {
// Run bottleneck detection
const bottlenecks = await runCommand(
'npx claude-flow bottleneck detect --format json'
);
// Generate performance report
const report = await runCommand(
'npx claude-flow analysis performance-report --format json'
);
// Analyze results
const analysis = {
bottlenecks: JSON.parse(bottlenecks),
performance: JSON.parse(report),
timestamp: new Date().toISOString()
};
// Save combined analysis
fs.writeFileSync(
'analysis/combined-report.json',
JSON.stringify(analysis, null, 2)
);
// Generate alerts if needed
if (analysis.bottlenecks.critical.length > 0) {
console.error('CRITICAL: Performance bottlenecks detected!');
process.exit(1);
}
}
function runCommand(cmd) {
return new Promise((resolve, reject) => {
exec(cmd, (error, stdout, stderr) => {
if (error) reject(error);
else resolve(stdout);
});
});
}
analyzePerformance().catch(console.error);
Best Practices
1. Regular Analysis
- Run bottleneck detection after major changes
- Generate weekly performance reports
- Monitor trends over time
- Set up automated alerts
2. Threshold Tuning
- Start with default threshold (20%)
- Lower for production systems (10-15%)
- Higher for development (25-30%)
- Adjust based on requirements
3. Fix Strategy
- Always review before applying --fix
- Test fixes in development first
- Apply fixes incrementally
- Monitor impact after changes
4. Report Integration
- Include in documentation
- Share with team regularly
- Track improvements over time
- Use for capacity planning
5. Continuous Optimization
- Learn from each analysis
- Build performance budgets
- Establish baselines
- Set improvement goals
Troubleshooting
Common Issues
High Memory Usage
# Analyze memory bottlenecks
npx claude-flow bottleneck detect --threshold 10
# Check cache performance
npx claude-flow cache manage --action stats
# Review memory metrics
npx claude-flow memory usage
Slow Task Execution
# Identify slow tasks
npx claude-flow task status --detailed
# Analyze coordination overhead
npx claude-flow bottleneck detect --time-range 1h
# Check agent utilization
npx claude-flow agent metrics
Poor Cache Performance
# Analyze cache hit rates
npx claude-flow analysis performance-report --sections metrics
# Review cache strategy
npx claude-flow cache manage --action analyze
# Enable cache warming
npx claude-flow bottleneck detect --fix
Integration with Other Skills
- swarm-orchestration: Use performance data to optimize topology
- memory-management: Improve cache strategies based on analysis
- task-coordination: Adjust scheduling based on bottlenecks
- neural-training: Train patterns from performance data
Related Commands
npx claude-flow swarm monitor- Real-time monitoringnpx claude-flow token usage- Token optimization analysisnpx claude-flow cache manage- Cache optimizationnpx claude-flow agent metrics- Agent performance metricsnpx claude-flow task status- Task execution analysis
See Also
- Bottleneck Detection Guide
- Performance Report Guide
- Performance Bottlenecks Overview
- Swarm Monitoring Documentation
- Memory Management Documentation
Version: 1.0.0 Last Updated: 2025-10-19 Maintainer: Claude Flow Team