Running Claude Code Via Litellm Copilot

This is Vercel's internal performance playbook turned into actionable code guidance. It runs 69 specific checks across React and Next.js codebases, catching things like barrel import performance hits, async waterfall patterns, and unnecessary re-renders. The rules are prioritized by actual impact, so it tackles bundle size killers and server-side bottlenecks before micro-optimizations. Each rule includes before/after code examples showing exactly what to fix. If you're shipping React apps at scale or just want to stop guessing about performance patterns, this codifies what the Vercel team actually follows internally.

When you need to build web interfaces that don't look like every other AI-generated site, this skill pushes Claude to make bold aesthetic choices and implement them with precision. It guides creation of production-ready HTML/CSS/JS or React components with distinctive typography, creative layouts, and thoughtful animations. Instead of defaulting to Inter fonts and purple gradients, it forces commitment to a specific design direction, whether that's brutalist minimalism or maximalist chaos. The code output includes working implementations with CSS variables, animation delays, and contextual visual effects that actually match the chosen aesthetic rather than falling back on generic patterns.

A solid UI audit tool that pulls the latest Web Interface Guidelines from Vercel's GitHub repo and checks your code against them. Point it at your HTML, CSS, or component files and it'll flag accessibility issues, design inconsistencies, and UX problems in a clean file:line format. The guidelines stay current since it fetches fresh rules on each run rather than using stale built-in checks. Handy for catching common web standards violations before they ship, especially if you're working fast and might miss obvious accessibility or usability issues.

A comprehensive browser automation CLI that handles the full spectrum of web interaction tasks through Chrome DevTools Protocol. What impressed me most is the thoughtful authentication handling with six different approaches, from importing existing browser sessions to encrypted credential vaults. The element reference system (@e1, @e2) makes DOM interaction reliable across page changes, and command chaining keeps workflows efficient. The network inspection and download management features push it beyond basic scraping into serious testing territory. If you're doing any programmatic web work beyond simple HTTP requests, this covers the gap between lightweight scraping libraries and heavyweight testing frameworks like Selenium.

Covers Azure's five core storage services with MCP tools for blob operations and CLI fallbacks for everything else. The breakdown of access tiers (hot, cool, cold, archive) and redundancy options (LRS through GZRS) is thorough, and the service comparison table makes it clear when to use blob storage versus file shares versus queues. Heavy focus on blob storage operations since that's what most people actually use day-to-day. The SDK references are comprehensive but the MCP integration only handles basic blob CRUD, so you'll still need CLI for lifecycle policies or advanced features.

This handles the tricky routing problem when you're working with GitHub Copilot SDK apps and Azure deployment. It automatically detects @github/copilot-sdk in your package.json or CopilotClient in your source files, then routes you to the right deployment path instead of generic Azure tools. The skill scaffolds new SDK projects with the official template, adds Azure infrastructure to existing Copilot apps, and configures BYOM (bring your own model) setups with proper DefaultAzureCredential auth. Honestly, it's mainly useful because the detection logic prevents you from accidentally using azure-prepare on Copilot SDK codebases, which breaks token management and model configuration.