Бенчмарк производительности воркеров в agentic-flow

testing

worker-benchmarks is a Claude Code skill that runs comprehensive performance benchmarks for the agentic-flow worker system and produces detailed analysis reports. It covers six benchmark types: trigger detection across 12 worker triggers (1,000 iterations, p95 < 5ms target), worker registry CRUD operations (p95 < 10ms), performance-based agent selection (p95 < 1ms), model cache hit rate and eviction stats (p95 < 0.5ms), concurrent creation of 10 workers (< 1,000ms total), and memory key generation (5,000 iterations, p95 < 0.1ms). Tests run via `npx agentic-flow workers benchmark` with an optional `--type` flag; thresholds are configured in `.claude/settings.json` and the TypeScript API allows programmatic use. Ideal for developers tuning latency and throughput in agentic pipelines or catching performance regressions in the worker layer.