k6-load-testing

Overview

performance-optimization is a post-hoc skill — "the prod p99 is 2× last week, hunt the bottleneck". This skill is the pre-deploy acceptance gate: "we just rewrote the auth middleware; prove it holds 1k req/s at p95 < 200ms with <0.1% errors before merging". The two are paired front-back: load-test sets the bar; optimization tells you how to clear it.

This skill uses k6 (Grafana's load-testing tool) as the default runner because it has a single-binary deploy, JS scenario syntax, and rich threshold/threshold-breach output. The runner is interchangeable — the methodology (smoke / load / stress / spike / soak, plus percentile + error-budget thresholds) is the actual content. Adapt the runner if your stack requires Locust, Gatling, Vegeta, or k8s-based tools; the test categories below apply unchanged.

OMO integration: analyze mode for post-run bottleneck diagnosis (p99 tails, GC pressure, connection-pool exhaustion); oracle agent for "is this regression expected given the change set?" judgment; websearch MCP for k6 syntax + threshold pattern lookup.

When to Use

Use when: