agent-v3-performance-engineer

Benchmarks and validates performance targets for Flash Attention, HNSW search, and agent coordination.

70.1k|8.4k|Updated Jun 2, 2025
One-click install
npx skills add https://github.com/ruvnet/claude-flow --skill agent-v3-performance-engineer
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: agent-v3-performance-engineer
Source: https://github.com/ruvnet/claude-flow/tree/main/.agents/skills/agent-v3-performance-engineer
Command: npx skills add https://github.com/ruvnet/claude-flow --skill agent-v3-performance-engineer

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires agentic-flow.

What problem does it solve?

Validating aggressive performance claims for an agent orchestration platform requires systematic benchmarking across attention mechanisms, vector search, memory usage, and swarm coordination, which is difficult to do consistently without a structured framework.

Core Features & Use Cases

  • Benchmark Suites: Provides TypeScript benchmark classes for startup time, memory operations, swarm coordination, Flash Attention, and SONA learning adaptation.
  • Performance Target Validation: Checks results against defined targets such as 2.49x-7.47x Flash Attention speedup, 150x-12,500x search improvement, and sub-500ms cold starts.
  • Regression Detection: Compares current benchmark runs against baselines and flags regressions exceeding a 5% threshold.
  • Use Case: A performance engineer validating a claude-flow v3 release runs the full benchmark suite to confirm AgentDB HNSW search meets the 150x minimum improvement before shipping.

Quick Start

Invoke the v3-performance-engineer agent to run the benchmark suite and validate all performance targets for the current build.

Frequently Asked Questions about agent-v3-performance-engineer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I benchmark Flash Attention speedup in TypeScript?

Compare standard attention against Flash Attention across test sequences of lengths 512 to 4096, measuring execution time and peak heap memory for each. The speedup ratio and memory reduction percentage are then checked against the 2.49x-7.47x and 50-75% targets.

How to measure HNSW vector search improvement over linear search?

Run the same set of test queries through both the baseline linear search and the AgentDB HNSW index, then divide baseline time by HNSW time. The benchmark validates whether the improvement falls within the 150x to 12,500x target range.

What performance targets does this agent validate?

It validates Flash Attention speedup of 2.49x-7.47x, search improvement of 150x-12,500x, memory reduction of 50-75%, cold start under 500ms, and SONA adaptation under 0.05ms. It also checks 15-agent swarm coordination efficiency.

How does performance regression detection work?

The regression detector runs the full benchmark suite and compares each metric against stored baseline values. Any metric degrading by more than 5% is flagged as a regression, and the report includes recommended fixes.

What are the limitations of benchmark-based performance validation?

Benchmark results depend on hardware, dataset size, and test sequence lengths, so targets validated in one environment may not hold elsewhere. The framework detects relative regressions against baselines but cannot guarantee absolute latency in production conditions.