Skip to main content

Reinvent Notes

Learn by reinventing the wheel

Reinventing the wheel is often called a waste, but rebuilding things from scratch lets you understand how they really work. A personal tech notebook recording what I learn by reinventing the frontend, backend, infrastructure, AI, and more.

📰 Latest posts

Recent writing, in chronological order

Japanese Web Fonts and Core Web Vitals — Fixing LCP and CLS by Measurement

June 14, 2026

Technical SEO is a perfect score, yet only Lighthouse Performance won't budge. Tracing the cause with chrome-devtools measurements, the LCP culprit was the render-blocking of a Japanese Web font, and the CLS culprit was a navbar logo with no dimensions. A record of rejecting the font-swap-as-culprit theory by measurement and driving CLS to 0 without adding size-adjust.

performancecore-web-vitalsdocusaurus

I built a running-car animation in the hero

June 2, 2026

A record of building a running-car animation from scratch for the Docusaurus top hero. I componentize the SVG line art with svgr and drive the tire rotation and background parallax with pure CSS. From drawing backgrounds with CSS mask, a seamless infinite loop, theme-color following, all the way to prefers-reduced-motion support, explained for beginners.

docusaurusreactcss

A Claude Code Best-Practice Collection: How to Navigate shanraisshan/claude-code-best-practice

May 25, 2026

A map for reading shanraisshan/claude-code-best-practice (445 files), the GitHub Trending #1 curation collection, along five axes: directory structure, concepts, workflows, tips, and reports. Organized so you can pick the elements that match your level.

ai-agentclaude-codebest-practice

I changed the blog domain to notes.rewheel.dev

May 18, 2026

I changed my personal tech blog's domain from notes.rasshii.dev to notes.rewheel.dev. I share the naming, which matches the concept of "learning by reinventing the wheel," along with my decision process.

metabranding

Introduction to Harness Engineering: An AI Agent's Reliability Is Decided by the Design Around It

May 12, 2026

The reliability of an AI agent (including Claude Code) depends not on the model itself but on the design quality of the harness. This post organizes the core equation Agent = Model + Harness, the dichotomy framework, and the staged adoption of Phases 0-6, with real examples from my ~/.claude/ setup.

ai-agentclaude-codeharness-engineering

👤 About

Yuki Igarashi (yuki / rasshii). Marketing engineer at Codmon Inc. Learning across the frontend, backend, infrastructure, and AI while improving the developer experience with an SRE mindset.