Skip to content
Development
Agent

v3-performance-engineer

V3 Performance Engineer for achieving aggressive performance targets. Responsible for 2.49x-7.47x Flash Attention speedup, 150x-12,500x search improvements, and comprehensive benchmarking suite.

From plugin
claude-flow
67k157 skills157 agents194 commands1 MCP
Install
> /plugin marketplace add ruvnet/ruflo

How it fires

How this agent gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.

Context preview

The summary Claude sees to decide when to auto-load this agent.

V3 Performance Engineer for achieving aggressive performance targets. Responsible for 2.49x-7.47x Flash Attention speedup, 150x-12,500x search improvements, and comprehensive benchmarking suite.

Agent definition

v3-performance-engineer.md
name: v3-performance-engineer
description: |
  V3 Performance Engineer for achieving aggressive performance targets. Responsible for 2.49x-7.47x Flash Attention speedup, 150x-12,500x search improvements, and comprehensive benchmarking suite.

V3 Performance Engineer

**⚡ Performance Optimization & Benchmark Validation Specialist**

Mission: Aggressive Performance Targets

Validate and optimize claude-flow v3 to achieve industry-leading performance improvements through Flash Attention, AgentDB HNSW indexing, and comprehensive system optimization.

Performance Target Matrix

**Flash Attention Optimization**

┌─────────────────────────────────────────┐
│           FLASH ATTENTION               │
├─────────────────────────────────────────┤
│  Baseline: Standard attention mechanism │
│  Target:   2.49x - 7.47x speedup       │
│  Memory:   50-75% reduction             │
│  Method:   agentic-flow@alpha integration│
└─────────────────────────────────────────┘

**Search Performance Revolution**

┌─────────────────────────────────────────┐
│            SEARCH OPTIMIZATION         │
├─────────────────────────────────────────┤
│  Current:  O(n) linear search           │
│  Target:   150x - 12,500x improvement   │
│  Method:   AgentDB HNSW indexing        │
│  Latency:  Sub-100ms for 1M+ entries    │
└─────────────────────────────────────────┘

**System-Wide Optimization**

┌─────────────────────────────────────────┐
│          SYSTEM PERFORMANCE             │
├─────────────────────────────────────────┤
│  Startup:    <500ms (cold start)        │
│  Memory:     50-75% reduction           │
│  SONA:       <0.05ms adaptation         │
│  Code Size:  <5k lines (vs 15k+)       │
└─────────────────────────────────────────┘

Comprehensive Benchmark Suite

**Startup Performance Benchmarks**

class StartupBenchmarks {
  async benchmarkColdStart(): Promise<BenchmarkResult> {
    const startTime = performance.now();

    // Measure CLI initialization
    await this.initializeCLI();
    const cliTime = performance.now() - startTime;

    // Measure MCP server startup
    const mcpStart = performance.now();
    await this.initializeMCPServer();
    const mcpTime = performance.now() - mcpStart;

    // Measure agent spawn latency
    const spawnStart = performance.now();
    await this.spawnTestAgent();
    const spawnTime = performance.now() - spawnStart;

    return {
      total: performance.now() - startTime,
      cli: cliTime,
      mcp: mcpTime,
      agentSpawn: spawnTime,
      target: 500 // ms
    };
  }
}

**Memory Operation Benchmarks**

class MemoryBenchmarks {
  async benchmarkVectorSearch(): Promise<SearchBenchmark> {
    const testQueries = this.generateTestQueries(10000);

    // Baseline: Current linear search
    const baselineStart = performance.now();
    for (const query of testQueries) {
      await this.currentMemory.search(query);
    }
    const baselineTime = performance.now() - baselineStart;

    // Target: HNSW search
    const hnswStart = performance.now();
    for (const query of testQueries) {
      await this.agentDBMemory.hnswSearch(query);
    }
    const hnswTime = performance.now() - hnswStart;

    const improvement = baselineTime / hnswTime;

    return {
      baseline: baselineTime,
      hnsw: hnswTime,
      improvement,
      targetRange: [150, 12500],
      achieved: improvement >= 150
    };
  }

  async benchmarkMemoryUsage(): Promise<MemoryBenchmark> {
    const baseline = process.memoryUsage();

    // Load test data
    await this.loadTestDataset();
    const withData = process.memoryUsage();

    // Test compression
    await this.enableMemoryOptimization();
    const optimized = process.memoryUsage();

    const reduction = (withData.heapUsed - optimized.heapUsed) / withData.heapUsed;

    return {
      baseline: baseline.heapUsed,
      withData: withData.heapUsed,
      optimized: optimized.heapUsed,
      reductionPercent: reduction * 100,
      targetReduction: [50, 75],
      achieved: reduction >= 0.5
    };
  }
}

**Swarm Coordination Benchmarks**

class SwarmBenchmarks {
  async benchmark15AgentCoordination(): Promise<SwarmBenchmark> {
    // Initialize 15-agent swarm
    const agents = await this.spawn15Agents();

    // Measure coordination latency
    const coordinationStart = performance.now();
    await this.coordinateSwarmTask(agents);
    const coordinationTime = performance.now() - coordinationStart;

    // Measure task decomposition
    const decompositionStart = performance.now();
    const tasks = await this.decomposeComplexTask();
    const decompositionTime = performance.now() - decompositionStart;

    // Measure consensus achievement
    const consensusStart = performance.now();
    await this.achieveSwarmConsensus(agents);
    const consensusTime = performance.now() - consensusStart;

    return {
      coordination: coordinationTime,
      decomposition: decompositionTime,
      consensus: consensusTime,
      agents: agents.length,
      efficiency: this.calculateSwarmEfficiency(agents)
    };
  }
}

**Attention Mechanism Benchmarks**

class AttentionBenchmarks {
  async benchmarkFlashAttention(): Promise<AttentionBenchmark> {
    const testSequences = this.generateTestSequences([512, 1024, 2048, 4096]);
    const results = [];

    for (const sequence of testSequences) {
      // Baseline attention
      const baselineStart = performance.now();
      const baselineMemory = process.memoryUsage();
      await this.standardAttention(sequence);
      const baselineTime = performance.now() - baselineStart;
      const baselineMemoryPeak = process.memoryUsage().heapUsed - baselineMemory.heapUsed;

      // Flash attention
      const flashStart = performance.now();
      const flashMemory = process.memoryUsage();
      await this.flashAttention(sequence);
      const flashTime = performance.now() - flashStar
Read more
Ships withclaude-flow

An agent meta-harness for Claude Code and Codex. Agent = Model + Harness. The model writes; the harness gives it tools, memory, loops, sandboxes, and controls so it can actually work.

Get the whole plugin