What problem does it solve? Raw benchmark numbers are easy to produce and easy to get wrong: a closed port can report 1626 RPS, thermal drift can fake a 30% regression, and a variance of 27% makes two measurements unrankable. This Skill turns load-test scripts into trustworthy measurements by enforcing protocols, validity gates, and controlled environments. ## Core Features & Use Cases - Versioned Vitest load suites: CI-stable WebSocket connection/message load tests and memory-leak gates, with rupture probes gated behind NF_RUN_WS_RUPTURE=1. - 31 standalone benchmark scripts: WebSocket connection ceilings, throughput, RPS and percentiles, pod capacity, soak tests for memory leaks, boot-time and HMR-delay measurement, and browserless e2e proofs (idempotence, TOTP, webhooks, rate limiting, graceful shutdown, cluster). - A/B performance benchmarking: bench-ab-mono.sh measures single-process production pipeline cost with alternating pairs, dispersion refusal above 3%, and thermal/indexer machine guards; bench-pairs.sh compares two camps (e.g. Nodefony vs Express/Fastify) with a separation verdict. - Use Case: Before committing a change to the HTTP pipeline, run alternating A/B pairs with bench-ab-mono.sh to confirm the gain exceeds machine noise, then generate an HTML report with bench-report.mjs for human review. ## Quick Start Ask the agent to run a load test against the dev server, for example: start the Nodefony dev server and measure how many concurrent WebSocket connections it sustains using the load-test scripts.
