Put Agent Skills in the Prompt, or Watch the Model Guess
Learn why prompt-scoped agent skills make AI assistance more reliable, how they compare to unskilled prompting, and how to implement them in your own workflow.
Writing desk
Essays and field notes on frontend architecture, performance, and practical AI coding workflows.
Learn why prompt-scoped agent skills make AI assistance more reliable, how they compare to unskilled prompting, and how to implement them in your own workflow.
A practical breakdown of how developers can use AI to improve coding productivity with concrete workflows, real token savings, and trade-offs that nobody talks about.
Improve Largest Contentful Paint (LCP) in Next.js with image optimisation, resource preloading, font strategies, and server rendering techniques.
Fix Cumulative Layout Shift (CLS) on Next.js landing pages — image dimensions, fonts, dynamic content, and skeleton patterns that eliminate unexpected shifts.
Learn what IndexedDB is, how it differs from localStorage, when to use it, and how to implement it in a framework-agnostic way.
Learn how to lazy-load Lottie, Framer Motion, and GSAP animation components in Next.js using dynamic imports — with real bundle reduction results.
Real Next.js performance case study: 66% bundle size reduction, Core Web Vitals improvement, and the exact code changes that achieved it in a production app.
Practical JavaScript guide for choosing Object, Array, Set, and Map by access pattern, with performance trade-offs and migration examples.
Improve Interaction to Next Paint (INP) in React — identify slow event handlers, defer non-critical work, and use scheduler patterns for responsive UIs.
How we scaled a React codebase from 2 to 20 engineers without rewrites — feature-based modules, shared conventions, and the architectural decisions that held up under pressure.
Most apps do not have a state management problem.
RTK is a Rust shell wrapper for AI coding agents that compresses CLI output before it hits model context windows, reducing token usage and cost.
Learn when to use showSaveFilePicker for large browser downloads, how streaming writes reduce memory pressure, and how to implement a safe fallback.
Compare Server-Sent Events and WebSockets in web development, including communication direction, reconnect behavior, browser APIs, scaling, security, and practical use cases.
Learn a practical framework for reducing unnecessary useEffect usage in React and improving component predictability, readability, and performance.
A complete web performance checklist for product teams covering Core Web Vitals, images, JavaScript, fonts, caching, and monitoring — downloadable and actionable.
How I use AI agents on performance tasks where numbers matter: code-splitting boundaries, lazy-loaded animation, WebP and preloads for LCP/TBT, and bundle diffs humans can trust.
Practical habits for faster frontend work with AI agents: prompt contracts, tight context, written specs, and tasks small enough to verify with one command.
Autonomous agents can implement, review, and test frontend code, but humans should keep explicit approval gates for merges, money UI, security, and irreversible changes. A practical checklist.
How QA agents run Maestro flows on iOS and Android for a fintech H5 app, write test cases, and feed failures back to implementer agents inside a closed PR loop.
How I tuned a three-agent PR review team on a mobile fintech frontend so comments are blocking when they matter and silent when they do not.
A frontend-first multi-agent setup for a fintech H5 app: implementers, PR reviewers, DevOps watchers, Maestro QA, architecture docs, and chat mention bots, with clear handoffs between them.
AI loop engineering means wiring coding agents into closed build, review, test, and deploy cycles instead of one-off prompts. Here is how I structure those loops on a production H5 frontend codebase.