What problem does it solve? Getting real Google PageSpeed Insights scores for both mobile and desktop, then figuring out exactly which code causes each failing audit, is slow and error-prone — the PSI REST API hits anonymous quota limits, the rendered report truncates element tables, and a stubborn 99-vs-100 score is hard to diagnose without understanding Lighthouse's rounding and variance behavior. ## Core Features & Use Cases - UI-driven PSI runs: Drives pagespeed.web.dev with the Playwright MCP (no API key or quota needed) and extracts the complete Lighthouse JSON for both form factors from the page's window globals. - Source-mapped fix lists: Turns every failing audit into a fix list tied to the metric it moves, using node selectors and labels to grep back to the responsible source code. - Road-to-100 playbook: Includes dead-weight audits (unused fonts, wasted preloads), pre-FCP CPU profiling via CDP, throttled-pipe arbitration, interleaved A/B prototype discipline, variance math, and an automated sampling loop to capture a verified 100. - Use Case: A developer asks "why are we not 100 on pagespeed" — the skill runs both form factors, detects a deterministic LCP ceiling caused by a dead webfont preload, removes it, and verifies quad-100 scores with screenshots. ## Quick Start Ask the assistant to run a PageSpeed audit on your deployed production URL and list the failing audits with their source files.
