optimization-mastery_skill

Overview

This skill codifies a 2026-grade cross-domain performance playbook that treats performance as a primary feature. It enforces Interaction to Next Paint (INP) limits, Partial Hydration patterns, UUIDv7 indexing, and efficient AI token usage. Use it to audit, implement, and harden frontends, backends, and AI tooling for real-user speed.

How this skill works

The skill inspects runtime interaction paths to detect main-thread blocking and enforces yielding patterns (scheduler.yield(), requestAnimationFrame, requestIdleCallback) to keep INP < 200ms. It evaluates hydration strategy and flags full-hydration anti-patterns, recommending partial hydration or resumability. On the backend, it scans schema and index design, enforcing UUIDv7 for high-insert tables and promoting covering indexes and edge compute where appropriate. For AI integrations it measures prompt token budgets, applies context folding, and recommends semantic caching and credit-based execution.

When to use it

• During frontend performance audits focused on real-user interaction responsiveness.
• When designing or migrating hydration strategy for SSR frameworks (avoid full hydration).
• Before schema design or bulk-write features in high-throughput services.
• When integrating LLMs to enforce token budgets and reduce cost/latency.
• When preparing deployments to edge platforms to lower TTFB.

Best practices

• Prioritize INP over synthetic frame-based metrics; reject any change that risks >200ms INP.
• Use scheduler.yield(), requestAnimationFrame or requestIdleCallback for heavy event-driven work to avoid layout thrashing.
• Adopt Partial Hydration or resumability for static pages; hydrate interactive islands only as needed.
• Use UUIDv7 for primary keys in high-insert tables to keep indexes time-ordered and reduce fragmentation.
• Create covering indexes for critical read paths and target sub-100ms OLTP queries.
• Fold conversational context, set explicit token budgets, and implement semantic caching for repeated LLM calls.

Example use cases

• Audit a single-page app to eliminate INP regressions caused by synchronous state updates.
• Refactor a server-rendered site from full hydration to partial hydration with interactive islands.
• Design a high-throughput event ingestion table using UUIDv7 primary keys and covering indexes.
• Move latency-sensitive endpoints to edge functions to improve TTFB and perceived performance.
• Optimize an AI assistant pipeline by summarizing history, applying token budgets, and caching responses.