Performance Analysis

Performance Analysis Skill

Identify bottlenecks, profile swarm operations, generate performance reports, and apply optimisation recommendations for Claude Flow swarms.

When To Use

• Profiling swarm performance or diagnosing slow agents
• Detecting communication / processing / memory / network bottlenecks
• Generating performance reports (json/html/markdown) for review or CI/CD
• Tuning topology, concurrency, caching, and resource allocation

When Not To Use

• CUDA GPU kernel profiling → use the cuda skill
• Full dev pipelines with quality gates and testing → use build-with-quality
• Truth scoring and automatic rollback of bad changes → use verification-quality
• General swarm orchestration without a performance focus → use swarm-advanced
• AgentDB vector-search performance tuning → use agentdb-vector-search

Claude Code only: the mcp__claude-flow__* tool call below needs the claude-flow MCP server registered (already wired in this container). On Codex / GPT-6 Astra: use the CLI form instead.

Quick Start

Verified against the deployed CLI (claude-flow performance --help and claude-flow performance bottleneck --help, ruflo v3.38.21) — there is no top-level bottleneck or analysis command; both live under performance:

# Detect bottlenecks in the current swarm (quick pass; -d full for deep analysis)
claude-flow performance bottleneck -d full

# View/export metrics (there is no `analysis performance-report`; use metrics)
claude-flow performance metrics -t 24h -f prometheus

# Run optimisation recommendations and apply them
claude-flow performance optimize --apply

From Claude Code (MCP):

mcp__claude-flow__bottleneck_analyze({ timeRange: "1h", threshold: 20, autoFix: false })

Core Capabilities

1. Bottleneck detection — communication, processing, memory, and network, with real-time profiling and severity-ranked output.
2. Report generation — executive summary, metrics, bottleneck analysis, and prioritized recommendations in json/html/markdown.
3. Optimisation — --fix applies topology, caching, concurrency, priority, and resource optimisations; review before applying.

References

Load these on demand for the full detail:

• references/bottleneck-detection.md — performance bottleneck options, the metric taxonomy, output format, common patterns, and MCP integration (JSON result shapes).
• references/reporting.md — performance metrics formats, sections, examples, and a sample report.
• references/optimisation-and-operations.md — performance optimize catalog, expected performance impact, continuous monitoring, CI/CD integration, custom scripts, best practices, and troubleshooting recipes.

Related

Verified against claude-flow --help (ruflo v3.38.21) — swarm monitor, token, and cache are not real top-level or subcommands in this build:

• claude-flow performance metrics — real-time and historical metrics (no swarm monitor exists)
• claude-flow agent metrics — agent performance metrics
• claude-flow performance benchmark — benchmark suites (wasm/neural/memory/search)
• swarm-advanced · agentdb-memory-patterns

Version: 1.1.0 · Last Updated: 2026-07-28 · Maintainer: Claude Flow Team