performance-analysis

Run npx claude-flow bottleneck detect to analyze the current swarm over the last hour. Use --swarm-id to target a specific swarm, --time-range for longer periods like 24h or 7d, and --threshold to control detection sensitivity.

Use npx claude-flow analysis performance-report with a --format option of json, html, or markdown. Add --include-metrics for detailed charts, --compare to benchmark against a previous swarm, and --output to save the report to a file.

Yes, adding the --fix flag applies automatic optimizations such as topology changes, cache warming, concurrency tuning, and priority adjustments. Review suggested fixes first and test them in a development environment before applying to production swarms.

It measures communication latency, agent utilization rates, task completion times, cache hit rates, memory access patterns, and parallelization efficiency. Results include impact percentages per bottleneck and expected improvement estimates for each recommendation.

Common causes include single agents handling complex tasks, unbalanced workloads, poor topology selection, and memory cache misses. Run bottleneck detection with a low threshold like 10-15% to surface these issues and get targeted recommendations.

Yes, export reports in JSON format and run bottleneck detection with --export in GitHub Actions or similar pipelines. You can fail builds when critical bottlenecks exceed defined thresholds and upload reports as build artifacts.