codemaxxing 0.2.0 → 0.2.1

package/src/skills/registry.ts

@@ -477,6 +477,960 @@ export const REGISTRY: RegistrySkill[] = [
 - Don't hardcode config — use environment variables or config maps
 - Don't ignore failed CI — fix or revert immediately
 - Don't alert on everything — alert fatigue leads to ignored alerts
+`,
+  },
+  {
+    name: "nextjs-app",
+    description: "Next.js App Router, Server/Client Components, server actions, caching, streaming",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["nextjs", "react", "app-router", "server-components", "streaming"],
+    prompt: `# Next.js App Router Expert
+
+## Server Components vs Client Components
+- Default to Server Components — they run on the server, ship zero JS to the client
+- Add "use client" only when you need: useState, useEffect, event handlers, browser APIs
+- Keep "use client" boundaries as low in the tree as possible — wrap only the interactive leaf
+- Server Components can import Client Components, but NOT the reverse
+- Pass server data to Client Components as serializable props — no functions, no classes
+- Use composition: Server Component fetches data, passes it to a Client Component for interactivity
+
+## Route Handlers & Server Actions
+- Use server actions ("use server") for mutations — forms, data writes, revalidation
+- Prefer server actions over API route handlers for app-internal mutations
+- Use route.ts (GET/POST/PUT/DELETE) for webhooks, external API consumers, and streaming responses
+- Always validate input in server actions with Zod or similar — they're public endpoints
+- Call revalidatePath() or revalidateTag() after mutations to bust the cache
+- Use useActionState (React 19) for form state + pending UI, not manual useState
+
+## Routing & Layouts
+- Use layout.tsx for shared UI that persists across navigations (navbars, sidebars)
+- Use template.tsx instead of layout.tsx when you need fresh state on every navigation
+- Implement loading.tsx per route segment for instant loading UI via Suspense
+- Implement error.tsx per route segment — wraps in an ErrorBoundary automatically
+- Use not-found.tsx for custom 404 pages, call notFound() to trigger programmatically
+- Route groups (parentheses) for organizing without affecting URL: (marketing)/about/page.tsx
+- Parallel routes (@modal) and intercepting routes ((..)photo) for modals and feeds
+
+## Data Fetching & Caching
+- Fetch in Server Components directly — no useEffect, no client-side fetching for initial data
+- Use the extended fetch options: \`fetch(url, { next: { revalidate: 3600, tags: ["posts"] } })\`
+- unstable_cache or "use cache" directive for caching non-fetch operations (DB queries)
+- Understand the cache layers: Request Memoization → Data Cache → Full Route Cache
+- Use \`export const dynamic = "force-dynamic"\` to opt out of static rendering per route
+- generateStaticParams for static generation of dynamic routes at build time
+
+## Streaming & Suspense
+- Wrap slow data fetches in <Suspense> with a fallback to stream the page progressively
+- Use loading.tsx for route-level Suspense — it wraps page.tsx automatically
+- Multiple <Suspense> boundaries let fast content appear while slow content loads
+- Streaming works out of the box — no special config needed with App Router
+
+## Metadata & SEO
+- Export metadata object or generateMetadata function from page.tsx and layout.tsx
+- Use generateMetadata for dynamic metadata (fetching title from DB)
+- Metadata merges from layout → page, with page taking precedence
+- Add opengraph-image.tsx and twitter-image.tsx for dynamic OG images
+- Use sitemap.ts and robots.ts for programmatic SEO files
+
+## Middleware
+- Use middleware.ts at the project root for auth checks, redirects, A/B testing, geo-routing
+- Middleware runs on the Edge — keep it lightweight, no heavy computation or DB queries
+- Use NextResponse.next() to continue, NextResponse.redirect() to redirect, NextResponse.rewrite() to rewrite
+- Match routes with the config.matcher array — don't run middleware on static assets
+
+## Anti-Patterns
+- Don't "use client" on layout.tsx — it defeats the purpose of Server Component layouts
+- Don't fetch data on the client when you can fetch in a Server Component
+- Don't use useEffect for data fetching in the App Router — use Server Components or server actions
+- Don't mix Pages Router patterns (getServerSideProps) with App Router
+- Don't put all components in a single "use client" boundary — isolate interactivity
+- Don't forget to handle the loading and error states per route segment
+`,
+  },
+  {
+    name: "tailwind-ui",
+    description: "Tailwind CSS, Shadcn UI, Radix primitives, responsive design, dark mode",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["tailwind", "shadcn", "radix", "css", "ui", "design-system"],
+    prompt: `# Tailwind CSS & Shadcn UI Expert
+
+## Utility-First Patterns
+- Build UI with utility classes directly — avoid creating CSS files unless truly necessary
+- Group related utilities logically: layout → spacing → sizing → typography → colors → effects
+- Use arbitrary values sparingly: \`w-[327px]\` is a sign you need a design token or a different approach
+- Prefer Tailwind's spacing scale (p-4, m-6) over arbitrary values for consistency
+- Use @apply only in base layer for truly repeated patterns (btn, input) — not for components
+- Prefer component extraction (React/Svelte/Vue components) over @apply for reuse
+
+## Responsive Design
+- Mobile-first: write base styles for mobile, override with sm:, md:, lg:, xl:, 2xl:
+- Breakpoints: sm(640) md(768) lg(1024) xl(1280) 2xl(1536) — learn them
+- Use container mx-auto for centered max-width layouts
+- Responsive grids: \`grid grid-cols-1 md:grid-cols-2 lg:grid-cols-3 gap-6\`
+- Hide/show elements responsively: \`hidden md:block\` or \`md:hidden\`
+- Use responsive typography: \`text-sm md:text-base lg:text-lg\`
+
+## Dark Mode
+- Use the "class" dark mode strategy for user-controlled toggling
+- Apply dark variants: \`bg-white dark:bg-slate-900 text-slate-900 dark:text-slate-100\`
+- Use CSS custom properties with Tailwind for theme-aware colors
+- Define semantic color names in tailwind.config: primary, secondary, muted, accent
+- Use Shadcn's built-in dark mode — it uses CSS variables under the hood
+- Test both modes — don't forget hover/focus states in dark mode
+
+## Shadcn UI & Radix Primitives
+- Install Shadcn components individually: \`npx shadcn@latest add button dialog\`
+- Components live in your codebase (src/components/ui/) — customize them directly
+- Use the cn() helper (clsx + tailwind-merge) for conditional + overridable classes
+- cn() pattern: \`cn("base classes", conditional && "conditional-class", className)\`
+- Radix provides unstyled accessible primitives — Shadcn adds Tailwind styles on top
+- Always pass className through to the root element of custom components for composability
+
+## Component Composition
+- Build complex components by composing Shadcn primitives: Dialog + Form + Button
+- Use Radix's compound component pattern: \`<Select><SelectTrigger><SelectContent>...</SelectContent></SelectTrigger></Select>\`
+- Use cva (class-variance-authority) for multi-variant components: size, color, state
+- Pattern: \`const buttonVariants = cva("base", { variants: { size: { sm: "h-8", lg: "h-12" } } })\`
+- Extend Shadcn components by wrapping them, not by modifying the generated code (easier upgrades)
+- Use Slot from Radix (asChild prop) to merge props onto child elements
+
+## Accessibility
+- Shadcn/Radix components handle ARIA attributes, keyboard navigation, and focus management
+- Don't remove or override ARIA roles/attributes added by Radix primitives
+- Use sr-only class for screen-reader-only text
+- Ensure sufficient color contrast — Tailwind's default palette mostly passes WCAG AA
+- Test with keyboard navigation: Tab, Enter, Escape, Arrow keys should all work
+
+## Anti-Patterns
+- Don't fight Tailwind by writing custom CSS for things utilities already handle
+- Don't create deeply nested className strings — extract components instead
+- Don't use inline styles alongside Tailwind — pick one approach
+- Don't ignore the cn() helper — raw string concatenation breaks with conflicting utilities
+- Don't copy-paste Shadcn components without understanding the underlying Radix primitives
+- Don't use arbitrary values for spacing when a Tailwind scale value is close enough
+- Don't forget to import Tailwind's base/components/utilities layers in your CSS entry point
+`,
+  },
+  {
+    name: "svelte-kit",
+    description: "Svelte 5 runes, SvelteKit routing, load functions, form actions, SSR/SSG",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["svelte", "sveltekit", "runes", "ssr", "frontend"],
+    prompt: `# Svelte 5 & SvelteKit Expert
+
+## Svelte 5 Runes
+- Use $state() for reactive state: \`let count = $state(0)\` — replaces the old \`let count = 0\` reactivity
+- Use $derived() for computed values: \`let doubled = $derived(count * 2)\` — replaces $: reactive labels
+- Use $derived.by() for multi-line derivations: \`let total = $derived.by(() => { /* compute */ return val })\`
+- Use $effect() for side effects: replaces onMount + $: reactive statements that cause side effects
+- $effect runs after DOM update — use $effect.pre() if you need to run before DOM update
+- Use $props() to declare component props: \`let { name, age = 25 }: Props = $props()\`
+- Use $bindable() for two-way bindable props: \`let { value = $bindable() }: Props = $props()\`
+- $inspect() for debugging reactive values — like console.log but re-runs when values change
+
+## Component Patterns
+- Svelte components are .svelte files with <script>, markup, and <style> sections
+- Use {#snippet name(params)} for reusable markup blocks within a component (replaces slots)
+- Use {@render snippetName(args)} to render snippets
+- Children content is received as a children snippet: \`let { children }: Props = $props()\` then {@render children?.()}
+- Use \`<svelte:component this={Component}\` for dynamic components
+- Scoped styles by default — styles in <style> only affect the current component
+- Use :global() sparingly for styles that must escape component scope
+
+## SvelteKit Routing
+- File-based routing: src/routes/about/+page.svelte → /about
+- Dynamic params: src/routes/users/[id]/+page.svelte → /users/123
+- Layout nesting: +layout.svelte wraps all child routes automatically
+- Use +page.ts for universal load (runs server + client), +page.server.ts for server-only load
+- Use +error.svelte for error pages, +layout.server.ts for layout-level data loading
+- Group routes without URL impact: (group)/route/+page.svelte
+- Rest params: [...slug]/+page.svelte catches all remaining path segments
+
+## Load Functions
+- Load functions in +page.ts export a load function that returns data as props
+- Access params, url, fetch (SvelteKit's enhanced fetch with credentials), and parent data
+- Use depends() to declare custom invalidation keys: \`depends('app:posts')\`
+- Call invalidate('app:posts') or invalidateAll() to re-run load functions
+- Server load functions (+page.server.ts) can access DB, env vars, and secrets directly
+- Data flows: +layout.server.ts → +layout.ts → +page.server.ts → +page.ts → component
+
+## Form Actions
+- Define actions in +page.server.ts: \`export const actions = { default: async ({ request }) => { ... } }\`
+- Use <form method="POST"> — SvelteKit handles it without JS (progressive enhancement)
+- Named actions: \`<form method="POST" action="?/create">\` maps to \`actions: { create: async () => {} }\`
+- Use enhance action for progressive enhancement: \`<form method="POST" use:enhance>\`
+- Return validation errors with fail(): \`return fail(400, { email, error: "Invalid email" })\`
+- Access returned data in the page via $page.form or the form prop from load
+
+## SSR, SSG & Prerendering
+- SSR is on by default — pages render on the server, then hydrate on the client
+- Prerender static pages: \`export const prerender = true\` in +page.ts or +layout.ts
+- Use adapter-static for full SSG (all pages prerendered)
+- Disable SSR for SPA pages: \`export const ssr = false\` — use only when necessary
+- Use adapter-node for Node server, adapter-vercel/adapter-netlify for serverless
+
+## Anti-Patterns
+- Don't use $: reactive labels in Svelte 5 — migrate to $state, $derived, $effect
+- Don't mutate $state values indirectly without assignment — Svelte 5 uses proxies but keep mutations explicit
+- Don't use $effect for derived values — use $derived instead
+- Don't put secrets or DB calls in +page.ts (universal) — use +page.server.ts for server-only code
+- Don't forget use:enhance on forms — without it, you lose SvelteKit's progressive enhancement
+- Don't create stores for local component state — $state() rune handles it
+`,
+  },
+  {
+    name: "react-native",
+    description: "Expo, React Navigation, platform-specific code, performance, EAS builds",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["react-native", "expo", "mobile", "ios", "android", "navigation"],
+    prompt: `# React Native & Expo Expert
+
+## Expo Workflow
+- Start with Expo for new projects — it handles native config, builds, and OTA updates
+- Use Expo Router for file-based routing (built on React Navigation)
+- Use expo-dev-client for custom native modules during development
+- Use EAS Build for cloud builds: \`eas build --platform ios\` — no local Xcode/Android Studio needed
+- Use EAS Submit to automate App Store and Play Store submissions
+- Use EAS Update for OTA JavaScript updates — skip the app store review cycle
+- Config plugins (app.config.ts) let you modify native projects without ejecting
+- Use expo-constants, expo-device, expo-file-system for cross-platform native APIs
+
+## Navigation (React Navigation / Expo Router)
+- Use Expo Router: app/_layout.tsx defines the navigator, app/index.tsx is the home screen
+- Stack navigator for hierarchical flows (push/pop), Tab navigator for main app sections
+- Use typed routes: \`router.push("/users/[id]", { id: "123" })\`
+- Deep linking works automatically with Expo Router — configure in app.config
+- Use navigation state to persist and restore navigation across app restarts
+- Modals: use presentation: "modal" in stack screen options
+- Avoid deeply nested navigators — flatten where possible for performance
+
+## Layout & Styling
+- Always wrap content in SafeAreaView (expo-safe-area-context) to avoid notches/status bars
+- Use StyleSheet.create() for styles — it validates and optimizes at creation time
+- Flexbox is the default layout system — flexDirection defaults to "column" (not "row" like web)
+- Use Dimensions or useWindowDimensions() for responsive layouts
+- Platform-specific code: Platform.select({ ios: value, android: value }) or .ios.tsx / .android.tsx files
+- Use react-native-reanimated for 60fps animations on the native thread
+- Avoid inline styles in render — they create new objects on every render
+
+## State Management
+- Use Zustand for global state — lightweight, works great with React Native
+- Use MMKV (react-native-mmkv) for persistent storage — 30x faster than AsyncStorage
+- Use TanStack Query for server state — caching, refetching, offline support
+- For offline-first: combine TanStack Query + MMKV persister + network-aware sync
+- Keep navigation state separate from app state — React Navigation manages its own state
+
+## Performance
+- Use FlatList (not ScrollView) for lists — it virtualizes, rendering only visible items
+- FlatList: set keyExtractor, getItemLayout (fixed height), maxToRenderPerBatch, windowSize
+- Use React.memo for list items to prevent re-renders when data hasn't changed
+- Use useCallback for event handlers passed to list items
+- Avoid passing new objects/arrays as props in render — memoize or define outside
+- Use Hermes engine (default in Expo) — it improves startup time and memory usage
+- Profile with React DevTools and Flipper — never optimize without measuring first
+
+## Native Modules & Platform Code
+- Check expo packages first before reaching for community native modules
+- Use expo-modules-api to write custom native modules in Swift/Kotlin
+- Bridge native UI with requireNativeComponent or Expo's native view pattern
+- Handle platform differences explicitly — don't assume iOS behavior works on Android
+- Test on real devices — simulators miss performance issues, permissions, and hardware quirks
+
+## Anti-Patterns
+- Don't use ScrollView for long lists — use FlatList or FlashList
+- Don't use web-specific APIs (window, document, localStorage) — they don't exist
+- Don't ignore the keyboard — use KeyboardAvoidingView or react-native-keyboard-aware-scroll-view
+- Don't hardcode dimensions — use relative sizing (flex) and useWindowDimensions
+- Don't skip testing on Android — platform differences are real and frequent
+- Don't use Animated API for complex animations — use react-native-reanimated instead
+`,
+  },
+  {
+    name: "swift-ios",
+    description: "SwiftUI, async/await, Combine, SwiftData, MVVM, App Store guidelines",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["swift", "ios", "swiftui", "apple", "mobile", "xcode"],
+    prompt: `# Swift iOS Development Expert
+
+## SwiftUI Views
+- Build UIs declaratively with structs conforming to View — return a body computed property
+- Use VStack, HStack, ZStack for layout composition — avoid GeometryReader unless truly needed
+- Prefer LazyVStack/LazyHStack inside ScrollView for long lists — they load items on demand
+- Use List for built-in swipe actions, selection, and pull-to-refresh
+- Modifiers order matters: \`.padding().background().cornerRadius()\` differs from reordering
+- Extract subviews into separate structs to keep body readable (under ~30 lines)
+- Use @ViewBuilder for functions that return opaque view types conditionally
+
+## Property Wrappers
+- @State for local view state — value types owned by the view
+- @Binding for two-way connection to a parent's @State
+- @StateObject for creating ObservableObject instances (create once, view owns it)
+- @ObservedObject for ObservableObject passed in from parent (view doesn't own it)
+- @EnvironmentObject for dependency injection of ObservableObject through the view hierarchy
+- @Environment(\\.colorScheme) for reading system environment values
+- @AppStorage for UserDefaults-backed persistent state
+- In iOS 17+: use @Observable macro instead of ObservableObject for simpler observation
+
+## MVVM Architecture
+- View: SwiftUI views — declarative UI, no business logic
+- ViewModel: @Observable class that holds state and business logic, exposes computed properties
+- Model: plain structs/classes for data — Codable for serialization
+- ViewModels should not import SwiftUI — keep them testable with plain Swift
+- Use protocols for dependencies (networking, storage) to enable testability
+- One ViewModel per screen/feature — avoid god ViewModels
+
+## Async/Await & Concurrency
+- Use async/await for network calls, file I/O, and any asynchronous work
+- Use Task { } in SwiftUI to launch async work from synchronous contexts (onAppear, buttons)
+- Use TaskGroup for concurrent parallel operations with structured cancellation
+- Actor types protect mutable state from data races — use for shared resources
+- @MainActor ensures code runs on the main thread — use for UI-updating classes
+- Use AsyncSequence and for-await loops for streaming data (WebSockets, file reading)
+- Handle cancellation: check Task.isCancelled or use withTaskCancellationHandler
+
+## Data Persistence
+- SwiftData (iOS 17+): use @Model macro on classes, @Query in views for reactive fetches
+- SwiftData replaces Core Data for most use cases — simpler API, better SwiftUI integration
+- Use ModelContainer for configuration, ModelContext for CRUD operations
+- For key-value storage: UserDefaults (@AppStorage) for small data, Keychain for secrets
+- For complex legacy needs: Core Data with NSPersistentContainer and @FetchRequest
+
+## App Store Guidelines
+- Follow Human Interface Guidelines — use standard navigation patterns and system controls
+- Request permissions just-in-time with clear usage descriptions in Info.plist
+- Implement in-app purchases with StoreKit 2 — server-side validation for subscriptions
+- Support Dynamic Type for text accessibility — use system fonts and relative sizing
+- Add VoiceOver labels to all interactive elements: \`.accessibilityLabel("Close button")\`
+- Test on multiple device sizes and orientations — use preview providers
+
+## Anti-Patterns
+- Don't use force unwraps (!) in production code — use guard let, if let, or ?? defaults
+- Don't put networking or heavy logic in View bodies — use ViewModels
+- Don't use GeometryReader for simple layouts — it causes unnecessary complexity
+- Don't ignore memory management — watch for retain cycles with [weak self] in closures
+- Don't use Timer for background work — use Task with proper lifecycle management
+- Don't hardcode strings — use String Catalogs or NSLocalizedString for localization
+`,
+  },
+  {
+    name: "flutter",
+    description: "Dart patterns, widget tree, state management (Riverpod/Bloc), Material Design 3",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["flutter", "dart", "mobile", "widgets", "riverpod", "material"],
+    prompt: `# Flutter & Dart Expert
+
+## Dart Patterns
+- Use null safety: \`String?\` for nullable, \`!\` only when you're certain (prefer null checks)
+- Prefer final for variables that won't be reassigned: \`final name = "Flutter";\`
+- Use named parameters with required keyword: \`void greet({required String name})\`
+- Use extension methods to add functionality to existing types without subclassing
+- Use sealed classes (Dart 3) for exhaustive pattern matching on state variants
+- Cascade notation (..) for chaining operations on the same object
+- Use records for lightweight data grouping: \`(String, int) getNameAndAge()\`
+- Pattern matching with switch expressions: \`final label = switch(status) { Status.ok => "Good" };\`
+
+## Widget Tree & Composition
+- Everything is a widget — compose small, focused widgets into complex UIs
+- StatelessWidget for UI that depends only on constructor parameters (immutable)
+- StatefulWidget when the widget needs to manage mutable state with setState()
+- Keep build() methods lean — extract sub-widget methods into separate widget classes
+- Use const constructors wherever possible — they enable widget tree optimizations
+- Prefer composition over inheritance — wrap widgets rather than extending them
+- Use the Builder pattern for context-dependent widgets: \`Builder(builder: (context) => ...)\`
+
+## State Management
+- setState() for local widget state — fine for simple, isolated state
+- Riverpod (recommended): Provider, StateNotifier, AsyncNotifier for scalable state
+- Riverpod patterns: define providers at top level, use ref.watch() in widgets, ref.read() for actions
+- Use AsyncValue (Riverpod) for loading/error/data states from async operations
+- Bloc pattern: Events → Bloc → States — good for complex business logic with clear event flows
+- Use ChangeNotifier + Provider for simpler apps, but Riverpod scales better
+- Avoid global mutable state — use providers/blocs to scope state to features
+
+## Navigation & Routing
+- Use GoRouter for declarative, URL-based routing with deep link support
+- Define routes in a central configuration: \`GoRouter(routes: [GoRoute(path: "/", builder: ...)])\`
+- Use shell routes for persistent navigation (bottom tabs, drawers)
+- Pass parameters via path or query params — avoid passing complex objects through navigation
+- Use guards (redirect) for authentication-protected routes
+
+## Platform Channels & Integration
+- Use MethodChannel for calling native platform code (Swift/Kotlin) from Dart
+- Use EventChannel for streaming data from native to Dart (sensors, Bluetooth)
+- Use Pigeon for type-safe platform channel communication — generates boilerplate
+- Check platform with \`Platform.isIOS\` / \`Platform.isAndroid\` for platform-specific behavior
+- Use federated plugins to share platform-specific implementations across packages
+
+## Material Design 3 & Theming
+- Use Material 3: \`MaterialApp(theme: ThemeData(useMaterial3: true, colorSchemeSeed: Colors.blue))\`
+- Use ColorScheme for systematic color usage: primary, secondary, surface, error variants
+- Use Theme.of(context) to access theme data — never hardcode colors or text styles
+- Use TextTheme for consistent typography: \`Theme.of(context).textTheme.headlineMedium\`
+- Support dark mode: provide both light and dark ThemeData to MaterialApp
+- Use adaptive widgets (e.g., Switch.adaptive) for platform-appropriate appearance
+
+## Performance
+- Use const widgets to avoid unnecessary rebuilds — the framework skips const subtrees
+- Use ListView.builder for long lists — it creates items lazily as they scroll into view
+- Profile with Flutter DevTools: widget rebuild counts, frame rendering times, memory
+- Use RepaintBoundary to isolate frequently updating parts of the widget tree
+- Avoid building widgets in initState — use didChangeDependencies for context-dependent setup
+- Use Isolates for CPU-heavy work — they run in separate threads without blocking UI
+
+## Anti-Patterns
+- Don't put all logic in widgets — separate business logic into repositories/services
+- Don't use setState for state shared across widgets — use a state management solution
+- Don't create deep widget nesting (10+ levels) — extract into named widget classes
+- Don't ignore keys in lists — use ValueKey or ObjectKey for items that move/reorder
+- Don't use BuildContext across async gaps — capture what you need before the await
+- Don't hardcode sizes — use MediaQuery, LayoutBuilder, or Flex for responsive layouts
+`,
+  },
+  {
+    name: "rust-systems",
+    description: "Ownership, borrowing, error handling, async Tokio, Actix/Axum, unsafe guidelines",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["rust", "systems", "tokio", "axum", "ownership", "concurrency"],
+    prompt: `# Rust Systems Programming Expert
+
+## Ownership & Borrowing
+- Each value has exactly one owner — when the owner goes out of scope, the value is dropped
+- Use references (&T) for read-only borrowing, (&mut T) for mutable borrowing
+- Rule: unlimited &T OR exactly one &mut T — never both simultaneously
+- Clone explicitly when you need a separate owned copy — don't fight the borrow checker blindly
+- Use Cow<str> (Clone on Write) when you sometimes need to own, sometimes just borrow
+- Prefer &str over String in function parameters — accepts both String and &str
+- Use .as_ref(), .as_str(), .into() for ergonomic type conversions
+
+## Error Handling
+- Use Result<T, E> for recoverable errors — never panic in library code
+- Use the ? operator to propagate errors up the call chain concisely
+- Define custom error enums with thiserror: \`#[error("not found: {id}")] NotFound { id: u64 }\`
+- Use anyhow::Result for application code where you don't need typed errors
+- Reserve panic!() and .unwrap() for truly unrecoverable situations or tests
+- Use .expect("reason") over .unwrap() — document WHY you believe it won't fail
+- Map errors at boundaries: \`.map_err(|e| MyError::Database(e))?\`
+
+## Lifetimes
+- Lifetimes prevent dangling references — the compiler ensures references outlive their use
+- Elision rules handle most cases — only annotate when the compiler asks you to
+- Named lifetimes: \`fn first<'a>(items: &'a [T]) -> &'a T\` — output lives as long as input
+- 'static means the reference lives for the entire program — owned data or compile-time constants
+- Lifetime bounds on structs: \`struct Parser<'a> { input: &'a str }\` — struct can't outlive the reference
+- When fighting lifetimes: consider if you should own the data instead of borrowing
+
+## Async with Tokio
+- Use #[tokio::main] for the async runtime entry point
+- Use tokio::spawn for concurrent tasks — returns a JoinHandle for awaiting results
+- Use tokio::select! to race multiple futures — first one to complete wins
+- Use tokio::sync::Mutex for async-safe locking (not std::sync::Mutex in async code)
+- Use channels (mpsc, broadcast, oneshot) for communication between tasks
+- Use tokio::time::timeout to prevent indefinite waits on futures
+- Use Stream (futures/tokio-stream) for async iterators — process items as they arrive
+
+## Web with Axum
+- Define handlers as async functions: \`async fn get_user(Path(id): Path<u64>) -> impl IntoResponse\`
+- Use extractors for parsing: Path, Query, Json, State, Headers — composable and type-safe
+- Shared state with State(Arc<AppState>) — wrap in Arc for thread-safe sharing
+- Use tower middleware for logging, CORS, auth, rate limiting — composable layers
+- Error handling: implement IntoResponse for your error type to control HTTP responses
+- Use Router::new().route("/users", get(list).post(create)) for clean route definitions
+- Use axum::serve with graceful_shutdown for production deployments
+
+## Cargo & Project Structure
+- Use Cargo workspaces for multi-crate projects: shared deps, unified builds
+- Feature flags for conditional compilation: \`#[cfg(feature = "postgres")]\`
+- Use clippy: \`cargo clippy -- -W clippy::all\` — it catches common mistakes and idioms
+- Organize: lib.rs for library logic, main.rs for binary entry, mod.rs for module roots
+- Use integration tests in tests/ directory — they test your public API as an external consumer
+- Profile with cargo bench (criterion) and cargo flamegraph for performance optimization
+
+## Unsafe Guidelines
+- Avoid unsafe unless absolutely necessary — most Rust code should be 100% safe
+- Valid reasons for unsafe: FFI, raw pointer manipulation, implementing unsafe traits
+- Document every unsafe block with a SAFETY comment explaining why it's sound
+- Minimize the scope of unsafe blocks — keep the unsafe surface area as small as possible
+- Use safe abstractions around unsafe internals — expose a safe API to callers
+- Prefer well-audited crates (libc, nix, windows-rs) over raw FFI bindings
+
+## Anti-Patterns
+- Don't clone everything to avoid the borrow checker — understand ownership first
+- Don't use Rc/RefCell as a default — they add runtime overhead, use references instead
+- Don't ignore compiler warnings — Rust warnings almost always indicate real issues
+- Don't use String when &str suffices — unnecessary allocation
+- Don't block the async runtime with synchronous I/O — use tokio::task::spawn_blocking
+- Don't write unsafe for convenience — only for correctness when safe Rust can't express it
+`,
+  },
+  {
+    name: "go-backend",
+    description: "Go standard library, error handling, goroutines, interfaces, table-driven tests",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["go", "golang", "backend", "concurrency", "testing"],
+    prompt: `# Go Backend Expert
+
+## Standard Library First
+- Use net/http for HTTP servers — it's production-ready and widely understood
+- Use http.ServeMux (Go 1.22+) with method-aware routing: \`mux.HandleFunc("GET /users/{id}", handler)\`
+- Use encoding/json for JSON: \`json.NewDecoder(r.Body).Decode(&v)\` for requests, json.NewEncoder for responses
+- Use database/sql with a driver (pgx, go-sql-driver) — it handles connection pooling
+- Use log/slog (Go 1.21+) for structured logging: \`slog.Info("user created", "id", user.ID)\`
+- Use html/template for server-rendered HTML — auto-escapes to prevent XSS
+- Use embed for bundling static assets and templates into the binary
+- Reach for third-party libraries only when the stdlib genuinely falls short
+
+## Error Handling
+- Errors are values — return them, don't panic: \`func Open(name string) (*File, error)\`
+- Always check errors: \`if err != nil { return fmt.Errorf("opening config: %w", err) }\`
+- Wrap errors with %w for context: \`fmt.Errorf("creating user %s: %w", name, err)\`
+- Use errors.Is() to check for specific errors: \`if errors.Is(err, sql.ErrNoRows)\`
+- Use errors.As() to extract typed errors: \`var pathErr *os.PathError; errors.As(err, &pathErr)\`
+- Define sentinel errors: \`var ErrNotFound = errors.New("not found")\`
+- Don't use panic for expected failures — reserve panic for truly unrecoverable programmer errors
+
+## Goroutines & Channels
+- Use goroutines for concurrent work — they're lightweight (2KB stack) and managed by the runtime
+- Always ensure goroutines can terminate — use context.Context for cancellation
+- Use sync.WaitGroup to wait for a group of goroutines to finish
+- Use channels for communication between goroutines: \`ch := make(chan Result, bufSize)\`
+- Buffered channels for producer/consumer, unbuffered for synchronization
+- Use select for multiplexing: waiting on multiple channels or timeouts
+- Use errgroup (golang.org/x/sync) for groups of goroutines that return errors
+- Avoid goroutine leaks: always close channels, cancel contexts, and handle done signals
+
+## Interfaces
+- Interfaces are small: \`type Reader interface { Read(p []byte) (n int, err error) }\`
+- Accept interfaces, return structs — keep interfaces at the consumer, not the producer
+- Don't create interfaces until you need them — premature abstraction is costly in Go
+- Implicit implementation: types satisfy interfaces without explicit declaration
+- Use io.Reader, io.Writer, fmt.Stringer — standard interfaces enable composition
+- Empty interface (any) should be rare — use generics (Go 1.18+) for type-safe polymorphism
+
+## Project Structure
+- Keep it simple: cmd/ for entry points, internal/ for private packages, pkg/ only if truly reusable
+- Use internal/ to prevent external packages from importing your implementation details
+- One package per directory — package name matches the directory name
+- Avoid circular dependencies — they're compile errors in Go
+- Use Go modules: go.mod at the root, run go mod tidy to clean up dependencies
+- Follow standard naming: lowercase packages, MixedCaps for exported, camelCase for unexported
+
+## Table-Driven Tests
+- Use table-driven tests for exhaustive input/output coverage:
+- Pattern: define []struct{ name, input, want }, loop with t.Run(tc.name, func(t *testing.T) { ... })
+- Use t.Parallel() for independent tests — speeds up test suites
+- Use testify/assert for readable assertions, or stick with stdlib for zero dependencies
+- Use t.Helper() in test helper functions to fix file/line reporting
+- Use t.Cleanup() for teardown — it runs after the test and all its subtests
+- Use testcontainers-go for integration tests with real databases and services
+
+## Anti-Patterns
+- Don't ignore errors with \`_ = someFunc()\` — handle or explicitly document why it's safe
+- Don't use init() functions — they make testing hard and hide initialization order
+- Don't use global mutable state — pass dependencies explicitly via struct fields or function params
+- Don't overuse channels — a mutex is simpler when you just need to protect shared data
+- Don't create interfaces for single implementations — that's Java, not Go
+- Don't return interfaces — return concrete types and let consumers define their own interfaces
+`,
+  },
+  {
+    name: "node-backend",
+    description: "Express/Fastify, middleware, input validation, JWT auth, error handling, graceful shutdown",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["node", "express", "fastify", "backend", "api", "middleware"],
+    prompt: `# Node.js Backend Expert
+
+## Framework Choice
+- Express for simplicity and ecosystem maturity — huge middleware library, well-documented
+- Fastify for performance — 2-3x faster than Express, built-in schema validation, TypeScript-first
+- Use Fastify's plugin system for encapsulated, reusable modules
+- Both support async handlers — always use async/await, never callbacks
+- For new projects: Fastify is recommended for its speed, schema validation, and TypeScript support
+
+## Middleware Patterns
+- Middleware executes in order — put auth before route handlers, error handlers last
+- Express: \`app.use(middleware)\` for global, \`router.use(middleware)\` for scoped
+- Fastify: use hooks (onRequest, preHandler, onSend) or plugins for middleware-like behavior
+- Common middleware stack: cors → helmet → rate-limiter → body-parser → auth → routes → error-handler
+- Keep middleware focused: one concern per middleware (logging, auth, validation)
+- Use express-async-errors or wrap handlers to catch async errors automatically
+
+## Input Validation
+- Validate ALL incoming data at the boundary — never trust req.body, req.params, req.query
+- Use Zod for runtime validation + TypeScript type inference: \`const UserSchema = z.object({...})\`
+- Validate in middleware: parse input before it reaches the handler, reject invalid requests early
+- Pattern: \`const data = UserSchema.parse(req.body)\` — throws ZodError on invalid input
+- Fastify: use JSON Schema in route definitions for automatic validation + serialization
+- Validate path params and query strings too — not just the body
+- Return structured validation errors: \`{ errors: [{ field: "email", message: "Invalid email" }] }\`
+
+## Authentication
+- Use bcrypt (cost factor 12+) for password hashing — never store plaintext passwords
+- JWTs for stateless auth: short-lived access tokens (15min), longer refresh tokens (7 days)
+- Store refresh tokens in httpOnly cookies — never in localStorage (XSS vulnerability)
+- Verify JWTs with jose or jsonwebtoken: check signature, expiry, issuer, audience
+- Middleware pattern: decode JWT → find user → attach to req.user → call next()
+- Use passport.js only if you need multiple OAuth providers — otherwise it's overhead
+- Implement token rotation: issue new refresh token on each use, revoke old one
+
+## Error Handling
+- Use a centralized error handler middleware — the last app.use() with (err, req, res, next)
+- Define custom error classes: AppError extends Error with statusCode and isOperational
+- Distinguish operational errors (bad input, not found) from programmer errors (null reference)
+- Never send stack traces to clients in production — log them, return a generic message
+- Use process.on('unhandledRejection') and process.on('uncaughtException') as safety nets
+- Return consistent error format: \`{ error: { code: "NOT_FOUND", message: "User not found" } }\`
+
+## Rate Limiting & Security
+- Use express-rate-limit or @fastify/rate-limit — essential for public APIs
+- Apply stricter limits to auth endpoints (login, register, password reset)
+- Use helmet for security headers: CSP, HSTS, X-Frame-Options, X-Content-Type-Options
+- Use cors with explicit origin allowlist — never use \`origin: "*"\` in production
+- Implement request ID (uuid) for tracing requests across services and logs
+
+## Structured Logging & Graceful Shutdown
+- Use pino (Fastify default) or winston — never console.log in production
+- Log as JSON with consistent fields: timestamp, level, requestId, message, metadata
+- Log at boundaries: incoming request, outgoing response, external API calls, errors
+- Graceful shutdown: handle SIGTERM/SIGINT → stop accepting requests → finish in-flight → close DB/Redis → exit
+- Pattern: \`process.on("SIGTERM", async () => { await server.close(); await db.end(); process.exit(0); })\`
+- Set a shutdown timeout (10-30s) — force exit if graceful shutdown hangs
+
+## Anti-Patterns
+- Don't use callback-based APIs — use promisify() or native promise alternatives
+- Don't block the event loop with synchronous operations (fs.readFileSync, crypto.pbkdf2Sync)
+- Don't catch errors silently: \`catch (e) {}\` — always log or handle meaningfully
+- Don't use \`res.send()\` after \`next()\` — it causes "headers already sent" errors
+- Don't store sessions in memory — use Redis or a database for production session storage
+- Don't skip input validation because "the frontend validates" — the frontend is untrusted
+`,
+  },
+  {
+    name: "sql-master",
+    description: "Query optimization, indexing, schema design, migrations, CTEs, window functions",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["sql", "postgres", "mysql", "database", "indexing", "performance"],
+    prompt: `# SQL Mastery Expert
+
+## Schema Design
+- Use appropriate data types: INTEGER for IDs, TEXT/VARCHAR for strings, TIMESTAMPTZ for dates
+- Always use TIMESTAMPTZ (not TIMESTAMP) for dates — store in UTC, convert in the app
+- Add NOT NULL constraints by default — make columns nullable only when NULL has business meaning
+- Use UUID (gen_random_uuid) for public-facing IDs, BIGSERIAL for internal primary keys
+- Define foreign keys with appropriate ON DELETE behavior: CASCADE, SET NULL, or RESTRICT
+- Add created_at and updated_at columns to every table — use triggers for updated_at
+- Use CHECK constraints for data validation: \`CHECK (price >= 0)\`, \`CHECK (status IN ('active','inactive'))\`
+- Normalize to 3NF by default — denormalize intentionally for read-heavy queries with measured need
+
+## Indexing Strategies
+- Index columns used in WHERE, JOIN, ORDER BY, and GROUP BY clauses
+- B-tree indexes (default) work for equality and range queries: =, <, >, BETWEEN, LIKE 'prefix%'
+- Composite indexes: put equality columns first, range columns last: \`(status, created_at)\`
+- Covering indexes include all columns a query needs — avoids heap lookups (INDEX ... INCLUDE)
+- Partial indexes for filtered queries: \`CREATE INDEX ON orders (user_id) WHERE status = 'pending'\`
+- GIN indexes for JSONB, full-text search, and array columns
+- Don't over-index — each index slows writes and consumes storage. Index what you query.
+- Use CONCURRENTLY for production index creation: \`CREATE INDEX CONCURRENTLY ...\`
+
+## Query Optimization
+- Use EXPLAIN ANALYZE to see actual execution plans — not just EXPLAIN (which estimates)
+- Look for: Seq Scan (missing index?), Nested Loop (N+1?), Sort (missing index?), high row estimates vs actuals
+- Avoid SELECT * — select only the columns you need
+- Use JOINs instead of subqueries where possible — the optimizer handles JOINs better
+- Avoid functions on indexed columns in WHERE: \`WHERE created_at > '2024-01-01'\` not \`WHERE YEAR(created_at) = 2024\`
+- Use EXISTS instead of IN for correlated subqueries — it short-circuits on first match
+- LIMIT early in subqueries to reduce the working set for outer queries
+- Use connection pooling (PgBouncer, built-in pool) — don't open a connection per request
+
+## Common Table Expressions (CTEs)
+- Use CTEs for readability: \`WITH active_users AS (SELECT ...) SELECT ... FROM active_users\`
+- CTEs are optimization fences in some databases — the optimizer may not push predicates into them
+- Recursive CTEs for hierarchical data: org charts, category trees, threaded comments
+- Pattern: \`WITH RECURSIVE tree AS (base UNION ALL SELECT ... FROM tree JOIN ... ) SELECT * FROM tree\`
+- Use CTEs to break complex queries into named, readable steps — each CTE is a logical unit
+
+## Window Functions
+- ROW_NUMBER() for pagination, deduplication, and picking the latest record per group
+- RANK() and DENSE_RANK() for ranking with ties (RANK skips, DENSE_RANK doesn't)
+- LAG/LEAD for comparing rows to previous/next rows: \`LAG(amount) OVER (ORDER BY date)\`
+- SUM/AVG/COUNT as window functions for running totals: \`SUM(amount) OVER (ORDER BY date)\`
+- Use PARTITION BY to compute within groups: \`ROW_NUMBER() OVER (PARTITION BY user_id ORDER BY created_at DESC)\`
+- FILTER clause for conditional aggregation: \`COUNT(*) FILTER (WHERE status = 'active')\`
+
+## Migrations
+- Use a migration tool: Flyway, Liquibase, golang-migrate, Prisma Migrate, or Knex
+- Every schema change is a numbered, version-controlled migration — never modify production DDL manually
+- Make migrations backward-compatible: add columns with defaults, avoid renaming in one step
+- Large table changes: add column → backfill → add constraint, not ALTER + NOT NULL in one migration
+- Always test migrations on a copy of production data before applying to production
+- Include both up and down migrations — even if you rarely use rollbacks
+
+## Avoiding N+1 Queries
+- N+1: fetching a list then querying related data per item — O(N) queries instead of O(1)
+- Fix with JOINs: \`SELECT users.*, orders.* FROM users JOIN orders ON ...\`
+- Fix with IN: \`SELECT * FROM orders WHERE user_id IN (1,2,3,...)\` — batch the lookups
+- ORMs: use eager loading (include, joinedload, preload) to fetch relations in one query
+- Monitor query counts in development — log the number of queries per request
+
+## Anti-Patterns
+- Don't use SELECT * in production queries — specify columns explicitly
+- Don't store CSV or JSON when you need relational data — normalize it
+- Don't skip foreign keys for "flexibility" — they enforce data integrity
+- Don't use OFFSET for deep pagination — use cursor-based (WHERE id > last_id) instead
+- Don't run schema changes without a migration tool — you'll lose track of database state
+- Don't optimize queries without EXPLAIN ANALYZE — measure before changing
+`,
+  },
+  {
+    name: "supabase",
+    description: "Auth, Row Level Security, edge functions, real-time, Postgres functions, storage",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["supabase", "postgres", "auth", "realtime", "edge-functions", "baas"],
+    prompt: `# Supabase Expert
+
+## Auth Setup
+- Use Supabase Auth for email/password, magic link, OAuth (Google, GitHub, etc.)
+- Client: \`const { data, error } = await supabase.auth.signUp({ email, password })\`
+- Always check for errors after auth operations — don't assume success
+- Use onAuthStateChange listener to react to sign-in/sign-out events globally
+- Store user metadata in a separate profiles table linked by auth.uid() — don't overload auth.users
+- Use auth.getUser() (server-side verified) over auth.getSession() (client JWT only) for security
+- Configure email templates and redirect URLs in the Supabase dashboard
+- Set up proper redirect handling for OAuth: signInWithOAuth({ options: { redirectTo } })
+
+## Row Level Security (RLS)
+- ALWAYS enable RLS on every table: \`ALTER TABLE posts ENABLE ROW LEVEL SECURITY\`
+- Without RLS policies, NO rows are accessible — it's deny-by-default
+- Use auth.uid() in policies to scope access to the authenticated user
+- SELECT policy: \`CREATE POLICY "users read own" ON posts FOR SELECT USING (user_id = auth.uid())\`
+- INSERT policy: \`CREATE POLICY "users insert own" ON posts FOR INSERT WITH CHECK (user_id = auth.uid())\`
+- Use auth.jwt() -> 'user_role' for role-based access in policies
+- Test RLS policies thoroughly — use Supabase SQL editor with \`SET request.jwt.claims = ...\`
+- For public data, create a policy with \`USING (true)\` — but be intentional about it
+
+## Database & Postgres Functions
+- Use the Supabase client for CRUD: \`supabase.from("posts").select("*, author:profiles(name)")\`
+- Foreign key relations are automatic in .select() — use the relation name for joins
+- Use Postgres functions for complex operations: \`supabase.rpc("function_name", { param: value })\`
+- Write functions in SQL or PL/pgSQL: \`CREATE FUNCTION get_stats(...) RETURNS TABLE (...) AS $$ ... $$\`
+- Use SECURITY DEFINER functions to bypass RLS when needed (admin operations) — with caution
+- Set search_path in SECURITY DEFINER functions to prevent search path injection
+- Use database triggers for automated side effects: updated_at timestamps, audit logs, notifications
+
+## Edge Functions
+- Write edge functions in Deno/TypeScript — they run on Supabase's edge infrastructure
+- Use for: webhooks, third-party API calls, custom auth flows, scheduled jobs
+- Create a Supabase client inside edge functions with the service_role key for admin access
+- Validate incoming requests: check auth headers, parse and validate request bodies
+- Use Deno's built-in fetch for external API calls — edge functions have internet access
+- Deploy with \`supabase functions deploy function-name\`
+- Set secrets with \`supabase secrets set API_KEY=value\` — access via Deno.env.get()
+
+## Real-Time Subscriptions
+- Subscribe to database changes: \`supabase.channel("posts").on("postgres_changes", { event: "INSERT", schema: "public", table: "posts" }, callback).subscribe()\`
+- Filter subscriptions: \`filter: "user_id=eq.abc123"\` — don't subscribe to all rows
+- Use Broadcast for ephemeral messages (typing indicators, cursor positions)
+- Use Presence for tracking online users and their state
+- Unsubscribe when components unmount: \`supabase.removeChannel(channel)\`
+- RLS policies apply to real-time — users only receive changes they're authorized to see
+
+## Storage
+- Use Supabase Storage for file uploads: images, documents, videos
+- Create buckets with appropriate policies: public (avatars) vs private (documents)
+- Upload: \`supabase.storage.from("avatars").upload(path, file, { contentType })\`
+- Use signed URLs for time-limited access to private files: \`createSignedUrl(path, expiresIn)\`
+- Use image transformations for thumbnails: \`getPublicUrl(path, { transform: { width: 200 } })\`
+- Set file size limits and allowed MIME types per bucket in storage policies
+
+## Typed Client Generation
+- Generate TypeScript types from your database: \`supabase gen types typescript --project-id xxx > types/supabase.ts\`
+- Use generated types with the client: \`createClient<Database>(url, key)\`
+- Types provide autocomplete for table names, column names, and filter operations
+- Regenerate types after every migration — keep types in sync with the schema
+- Use Database['public']['Tables']['posts']['Row'] for row types in your app code
+
+## Anti-Patterns
+- Don't skip RLS — an unprotected table is a data breach waiting to happen
+- Don't use the service_role key on the client side — it bypasses all RLS
+- Don't query without filters on large tables — always scope your queries
+- Don't use real-time for everything — polling is simpler for low-frequency updates
+- Don't store business logic only in the client — use Postgres functions or edge functions
+- Don't hardcode Supabase URLs and keys — use environment variables
+`,
+  },
+  {
+    name: "unity-csharp",
+    description: "MonoBehaviour lifecycle, ScriptableObjects, ECS, object pooling, performance profiling",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["unity", "csharp", "gamedev", "ecs", "performance", "game-engine"],
+    prompt: `# Unity & C# Game Development Expert
+
+## MonoBehaviour Lifecycle
+- Awake() → OnEnable() → Start() → FixedUpdate() → Update() → LateUpdate() → OnDisable() → OnDestroy()
+- Use Awake() for self-initialization (cache component references with GetComponent<T>())
+- Use Start() for initialization that depends on other objects being ready
+- Use FixedUpdate() for physics — it runs at a fixed timestep (default 50Hz), independent of framerate
+- Use Update() for input handling and non-physics game logic — runs once per frame
+- Use LateUpdate() for camera follow and anything that must happen after all Update() calls
+- Cache component references in Awake: \`private Rigidbody _rb; void Awake() => _rb = GetComponent<Rigidbody>();\`
+- Never use GetComponent in Update — it's an expensive lookup every frame
+
+## ScriptableObjects
+- Use ScriptableObjects for shared, data-driven assets: enemy stats, item definitions, game config
+- Create with: \`[CreateAssetMenu(fileName = "NewWeapon", menuName = "Game/Weapon")] public class WeaponData : ScriptableObject\`
+- ScriptableObjects persist in the project (not the scene) — changes in editor persist, changes at runtime don't (in builds)
+- Use them to decouple data from behavior — MonoBehaviours reference ScriptableObjects, not each other
+- Event channels: use ScriptableObject-based events to decouple systems without direct references
+- Pattern: ScriptableObject holds a list of listeners, Raise() notifies all — no singleton needed
+
+## Object Pooling
+- Instantiate/Destroy is expensive — pool frequently created/destroyed objects (bullets, particles, enemies)
+- Unity's built-in ObjectPool<T> (UnityEngine.Pool): \`new ObjectPool<GameObject>(createFunc, onGet, onRelease, onDestroy)\`
+- On Get: activate the object, reset its state (position, velocity, health)
+- On Release: deactivate the object, return it to the pool — don't Destroy it
+- Pre-warm pools at scene load to avoid frame hitches during gameplay
+- Use pool for particle systems, audio sources, and UI elements — not just game objects
+
+## Entity Component System (ECS)
+- Unity DOTS: Entities for data, Systems for logic, Components for pure data (no methods)
+- IComponentData structs are blittable data — no references, no managed types
+- SystemBase or ISystem process entities with matching component queries
+- Use Burst compiler for performance-critical systems — compiles to optimized native code
+- Jobs system for multi-threaded work: IJobEntity, IJobChunk for parallel processing
+- Use ECS for massive entity counts (10K+ units) — traditional MonoBehaviours for simpler needs
+
+## Physics Best Practices
+- Move Rigidbodies with forces (AddForce) or velocity — never transform.position for physics objects
+- Use layers and the collision matrix to control which objects interact
+- Use Physics.OverlapSphereNonAlloc() to avoid allocations in physics queries
+- Prefer discrete collision detection; use continuous only for fast-moving small objects
+- Use FixedUpdate for all physics code — Update runs at variable rate and causes jittery physics
+- Avoid moving static colliders at runtime — it rebuilds the physics tree (expensive)
+
+## Coroutines vs Async/Await
+- Coroutines (yield return): tied to MonoBehaviour, stop when GameObject is disabled/destroyed
+- Use yield return new WaitForSeconds(t) for delays, WaitUntil(() => condition) for polling
+- Async/await (UniTask recommended): proper cancellation, no MonoBehaviour dependency, awaitable
+- Use CancellationToken with async/await and link to destroyCancellationToken for auto-cleanup
+- Prefer async/await for complex async flows; coroutines for simple sequences and visual scripting
+
+## Asset Management
+- Use Addressables for loading assets on demand — reduces build size and startup time
+- Never use Resources.Load in production — it forces all Resources/ assets into the build
+- Unload unused assets: Addressables.Release() for counted references, Resources.UnloadUnusedAssets() as fallback
+- Use asset bundles / Addressable groups to organize content for DLC or streaming
+- Sprite atlases for UI and 2D — reduce draw calls by batching sprites into a single texture
+
+## Performance Profiling
+- Use Unity Profiler (Window > Analysis > Profiler) — identify CPU, GPU, memory, and rendering bottlenecks
+- Profile on target hardware — editor performance is not representative of builds
+- Watch for GC allocations in hot paths: use struct over class, avoid LINQ in Update, pre-allocate collections
+- Reduce draw calls: batching, atlasing, LODs, occlusion culling
+- Use Frame Debugger to inspect individual draw calls and shader passes
+- Profile with Deep Profile only when needed — it adds significant overhead
+
+## Anti-Patterns
+- Don't use Find("name") or FindObjectOfType at runtime — cache references or use events
+- Don't allocate in Update: no new List<T>, no LINQ, no string concatenation in hot loops
+- Don't use SendMessage — it's slow and stringly typed, use direct references or events
+- Don't put everything on one GameObject — split responsibilities across child objects
+- Don't ignore the profiler — gut feelings about performance are usually wrong
+- Don't use singletons for everything — use dependency injection or ScriptableObject events
+`,
+  },
+  {
+    name: "git-workflow",
+    description: "Conventional commits, branching strategies, PR best practices, rebase, hooks",
+    version: "1.0.0",
+    author: "codemaxxing",
+    tags: ["git", "workflow", "commits", "branching", "code-review", "hooks"],
+    prompt: `# Git Workflow Expert
+
+## Conventional Commits
+- Format: \`type(scope): description\` — e.g., \`feat(auth): add OAuth2 login flow\`
+- Types: feat (new feature), fix (bug fix), docs, style, refactor, perf, test, build, ci, chore
+- Breaking changes: add ! after type or BREAKING CHANGE in footer: \`feat(api)!: rename /users to /accounts\`
+- Scope is optional but helpful: module, component, or feature area
+- Description: imperative mood, lowercase, no period — "add login page" not "Added login page."
+- Body: wrap at 72 chars, explain WHAT and WHY (not HOW — the diff shows how)
+- Use commitlint + husky to enforce conventional commits in CI and locally
+
+## Branching Strategies
+- Trunk-based: short-lived feature branches (1-2 days), merge to main frequently, feature flags for WIP
+- GitFlow: main + develop + feature/ + release/ + hotfix/ — suited for versioned releases, more overhead
+- GitHub Flow: branch from main → PR → review → merge to main — simple, good for continuous deployment
+- For most teams: trunk-based or GitHub Flow — GitFlow adds complexity that slows teams down
+- Branch naming: type/description — \`feat/user-auth\`, \`fix/login-redirect\`, \`chore/update-deps\`
+- Delete branches after merge — don't accumulate stale branches
+
+## Pull Request Best Practices
+- Keep PRs small (under 400 lines changed) — large PRs get rubber-stamped, not reviewed
+- Write descriptive PR titles and descriptions: what changed, why, how to test
+- One concern per PR — don't mix a feature with a refactor with a bug fix
+- Add screenshots/recordings for UI changes
+- Self-review your PR before requesting review — catch obvious issues yourself
+- Request reviewers who own the affected code area
+- Respond to review comments promptly — don't let PRs go stale
+
+## Interactive Rebase
+- Use \`git rebase -i HEAD~N\` to clean up commits before opening a PR
+- Squash WIP commits into logical units: one commit per meaningful change
+- Reword commit messages to follow conventions: pick → reword
+- Fixup: absorb small fixes into the commit they belong to: pick → fixup
+- Never rebase commits that are already on shared branches (main, develop)
+- Prefer rebase over merge for feature branches — cleaner linear history
+- After rebase, force-push to your branch only: \`git push --force-with-lease\` (safer than --force)
+
+## Cherry-Pick & Bisect
+- Cherry-pick to apply specific commits to another branch: \`git cherry-pick <sha>\`
+- Use cherry-pick for hotfixes: fix on main, cherry-pick to release branch
+- Use -x flag to record the source commit: \`git cherry-pick -x <sha>\`
+- Bisect to find the commit that introduced a bug: \`git bisect start\`, \`git bisect bad\`, \`git bisect good <sha>\`
+- Automate bisect with a test script: \`git bisect run npm test\`
+- Bisect is O(log n) — it finds the bad commit in ~10 steps for 1000 commits
+
+## Git Hooks
+- Use husky (npm) or pre-commit (Python) to manage hooks across the team
+- pre-commit: lint staged files (lint-staged), format code, check for secrets
+- commit-msg: validate commit message format (commitlint)
+- pre-push: run tests, type-check — catch issues before they hit CI
+- Don't put slow operations in pre-commit — keep it under 5 seconds
+- Use lint-staged to only process staged files: \`"*.ts": ["eslint --fix", "prettier --write"]\`
+
+## .gitattributes & Configuration
+- Use .gitattributes for consistent line endings: \`* text=auto\` normalizes to LF
+- Mark binary files: \`*.png binary\`, \`*.zip binary\` — prevents diff/merge issues
+- Use .gitattributes for diff drivers: \`*.lockfile binary\` to skip noisy lockfile diffs
+- Use .gitignore for build artifacts, dependencies, env files, editor config
+- Global gitignore (~/.gitignore_global) for personal editor files (.vscode, .idea, .DS_Store)
+- Configure merge strategies per file: \`package-lock.json merge=ours\` to auto-resolve lockfile conflicts
+
+## Advanced Techniques
+- Use git stash for quick context switches: \`git stash push -m "WIP: feature X"\`
+- Use git worktree for working on multiple branches simultaneously without stashing
+- Use git reflog to recover lost commits — it tracks every HEAD movement for 90 days
+- Use git blame -w to ignore whitespace changes when tracking down who wrote a line
+- Use git log --all --graph --oneline for a visual branch/merge history
+- Use git diff --stat for a quick summary of changes before committing
+
+## Anti-Patterns
+- Don't commit directly to main — use branches and PRs for all changes
+- Don't use \`git add .\` blindly — review staged files to avoid committing secrets or artifacts
+- Don't rewrite shared history — rebase and amend only your own unpushed commits
+- Don't write commit messages like "fix" or "WIP" or "asdf" — future you will be confused
+- Don't keep long-lived branches — merge or rebase frequently to avoid painful conflicts
+- Don't skip code review — even small changes benefit from a second pair of eyes
 `,
   },
 ];