@rangojs/router 0.0.0-experimental.15 → 0.0.0-experimental.151

package/skills/document-cache/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: document-cache
-description: Cache full HTTP responses at the edge with Cache-Control headers
+description: Cache the whole HTTP response at the edge with Cache-Control headers. Use when caching an entire page or response at a CDN edge, setting Cache-Control headers, or cutting origin hits for public pages — not for caching a single segment or function.
 argument-hint: [setup]
 ---
 
@@ -8,75 +8,92 @@ argument-hint: [setup]
 
 Caches complete HTTP responses (HTML/RSC) at the edge based on Cache-Control headers. Routes opt-in by setting `s-maxage`.
 
+## Not this skill if…
+
+- You want loaders to stay live while rendered segments are reused — document
+  caching freezes the WHOLE response, loader output included; segment caching
+  is `cache()`: see `/caching`.
+- You want a cached HTML shell with per-request live holes — see `/ppr`.
+- You are unsure which cache layer you need — start at `/cache-guide`.
+
 ## Setup
 
-Configure document cache in router:
+Document caching is a middleware. Add `createDocumentCacheMiddleware()` to the
+router with `.use()`. The cache store it reads from is the app-level store you
+configure on `createRouter({ cache })` (available on the request context as
+`requestCtx._cacheStore`), not a store passed to the middleware.
 
 ```typescript
 import { createRouter } from "@rangojs/router";
-import { CFCacheStore } from "@rangojs/router/cache/cf";
+import {
+  createDocumentCacheMiddleware,
+  CFCacheStore,
+} from "@rangojs/router/cache";
 import { urlpatterns } from "./urls";
 
-const router = createRouter<AppEnv>({
+const router = createRouter<AppBindings>({
   document: Document,
   urls: urlpatterns,
-  documentCache: (env) => ({
-    store: new CFCacheStore({ ctx: env.ctx }),
+  // App-level cache store. The document cache middleware uses this store's
+  // getResponse/putResponse methods.
+  cache: (_env, ctx) => ({ store: new CFCacheStore({ ctx: ctx! }) }),
+});
+
+router.use(
+  createDocumentCacheMiddleware({
     skipPaths: ["/api", "/admin"],
     debug: process.env.NODE_ENV === "development",
   }),
-});
+);
 
 export default router;
 ```
 
-## Route Opt-In with cache()
+## Route Opt-In with Cache-Control
 
-Routes opt-in to document caching using the `cache()` DSL with `documentCache` option:
+Routes opt-in to document caching by setting a `Cache-Control` response header
+with `s-maxage`. The middleware caches responses whose `Cache-Control` includes
+`s-maxage`; `stale-while-revalidate` enables background revalidation (SWR).
 
 ```typescript
-import { urls } from "@rangojs/router";
-
-export const urlpatterns = urls(({ path, cache }) => [
-  // Cache full page for 5 min, serve stale for 1 hour
-  cache({ documentCache: { sMaxAge: 300, swr: 3600 } }, () => [
-    path("/blog", BlogIndex, { name: "blog" }),
-  ]),
-
-  // Long cache for individual posts
-  cache({ documentCache: { sMaxAge: 3600, swr: 86400 } }, () => [
-    path("/blog/:slug", BlogPost, { name: "blogPost" }),
-  ]),
-
-  // No cache for dashboard (no documentCache option)
-  path("/dashboard", Dashboard, { name: "dashboard" }),
-]);
+// Cache full page for 5 min, serve stale for 1 hour
+function BlogIndexHandler(ctx) {
+  ctx.headers.set("Cache-Control", "s-maxage=300, stale-while-revalidate=3600");
+  return <BlogIndex />;
+}
+
+// Long cache for individual posts
+function BlogPostHandler(ctx) {
+  ctx.headers.set("Cache-Control", "s-maxage=3600, stale-while-revalidate=86400");
+  return <BlogPost />;
+}
+
+// Dashboard sets no Cache-Control header, so it is never document-cached.
 ```
 
 ## Document Cache Options
 
-```typescript
-createRouter({
-  // ...
-  documentCache: (env) => ({
-    // Cache store (required)
-    store: new CFCacheStore({ ctx: env.ctx }),
+`createDocumentCacheMiddleware(options?)` accepts:
 
-    // Skip specific paths
-    skipPaths: ["/api", "/admin"],
+```typescript
+createDocumentCacheMiddleware({
+  // Skip specific paths (matched by pathname prefix)
+  skipPaths: ["/api", "/admin"],
 
-    // Custom cache key
-    keyGenerator: (url) => url.pathname,
+  // Custom cache key generator
+  keyGenerator: (url) => url.pathname,
 
-    // Conditional caching
-    isEnabled: (ctx) => !ctx.request.headers.has("x-preview"),
+  // Conditional caching, evaluated per request
+  isEnabled: (ctx) => !ctx.request.headers.has("x-preview"),
 
-    // Debug logging
-    debug: true,
-  }),
+  // Debug logging (HIT, MISS, STALE, REVALIDATED)
+  debug: true,
 });
 ```
 
+The cache store is not a middleware option — it comes from the app-level
+`createRouter({ cache })` store.
+
 ## How It Works
 
 ```
@@ -89,7 +106,7 @@ Request → Check Cache
     ↓             ↓
   Fresh?      Run handler
     │             │
-   Yes → Return   Has documentCache?
+   Yes → Return   Has s-maxage?
     │             │
    No (stale)    Yes → Cache + Return
     │             │
@@ -102,6 +119,7 @@ Request → Check Cache
 ## Cache Status Header
 
 Response includes `x-document-cache-status`:
+
 - `HIT` - Fresh cache hit
 - `STALE` - Served stale, revalidating in background
 - `MISS` - Cache miss, response was generated fresh
@@ -109,6 +127,7 @@ Response includes `x-document-cache-status`:
 ## Cache Key Generation
 
 Default keys differentiate:
+
 - HTML requests: `{pathname}:html`
 - RSC partials: `{pathname}:{segmentHash}:rsc`
 
@@ -118,13 +137,14 @@ Segment hash ensures different cached responses for navigations from different s
 
 - Full HTML responses (document requests)
 - RSC payloads (client navigation)
-- Only 200 OK responses with documentCache enabled
+- Only 200 OK responses whose `Cache-Control` includes `s-maxage`
 
 ## What's NOT Cached
 
 - Server actions (`_rsc_action`)
 - Loader requests (`_rsc_loader`)
-- Routes without `documentCache` option
+- Non-GET requests
+- Responses without an `s-maxage` `Cache-Control` directive
 - Non-200 responses
 
 ## Complete Example
@@ -132,49 +152,62 @@ Segment hash ensures different cached responses for navigations from different s
 ```typescript
 // router.tsx
 import { createRouter } from "@rangojs/router";
-import { CFCacheStore } from "@rangojs/router/cache/cf";
+import { createDocumentCacheMiddleware, CFCacheStore } from "@rangojs/router/cache";
 import { urlpatterns } from "./urls";
 
-const router = createRouter<AppEnv>({
+const router = createRouter<AppBindings>({
   document: Document,
   urls: urlpatterns,
-  documentCache: (env) => ({
-    store: new CFCacheStore({ ctx: env.ctx }),
+  cache: (_env, ctx) => ({ store: new CFCacheStore({ ctx: ctx! }) }),
+});
+
+router.use(
+  createDocumentCacheMiddleware({
     skipPaths: ["/api"],
     debug: process.env.NODE_ENV === "development",
   }),
-});
+);
 
 export default router;
 
 // urls.tsx
 import { urls } from "@rangojs/router";
 
-export const urlpatterns = urls(({ path, layout, cache, loader }) => [
-  // Blog with document caching
-  cache({ documentCache: { sMaxAge: 300, swr: 3600 } }, () => [
-    layout(<BlogLayout />, () => [
-      path("/blog", BlogIndex, { name: "blog" }),
-      path("/blog/:slug", BlogPost, { name: "blogPost" }, () => [
-        loader(BlogPostLoader),
-      ]),
+export const urlpatterns = urls(({ path, layout, loader }) => [
+  // Blog pages opt into document caching via Cache-Control headers set in
+  // their handlers (see BlogIndex / BlogPost below).
+  layout(<BlogLayout />, () => [
+    path("/blog", BlogIndex, { name: "blog" }),
+    path("/blog/:slug", BlogPost, { name: "blogPost" }, () => [
+      loader(BlogPostLoader),
     ]),
   ]),
 
-  // Dashboard - no document cache (dynamic content)
+  // Dashboard sets no Cache-Control header, so it is never document-cached.
   layout(<DashboardLayout />, () => [
     path("/dashboard", Dashboard, { name: "dashboard" }),
   ]),
 ]);
+
+// Blog handlers set s-maxage to opt into the document cache.
+function BlogIndex(ctx) {
+  ctx.headers.set("Cache-Control", "s-maxage=300, stale-while-revalidate=3600");
+  return <BlogIndexPage />;
+}
+
+function BlogPost(ctx) {
+  ctx.headers.set("Cache-Control", "s-maxage=300, stale-while-revalidate=3600");
+  return <BlogPostPage />;
+}
 ```
 
 ## Document Cache vs Segment Cache
 
-| Feature | Document Cache | Segment Cache |
-|---------|---------------|---------------|
-| Granularity | Full response | Individual segments |
-| Opt-in | `documentCache` in cache() | `cache({ ttl, swr })` |
-| Use case | Static pages | Dynamic compositions |
-| Key includes | URL + segment hash | Route params |
+| Feature      | Document Cache             | Segment Cache         |
+| ------------ | -------------------------- | --------------------- |
+| Granularity  | Full response              | Individual segments   |
+| Opt-in       | `Cache-Control` `s-maxage` | `cache({ ttl, swr })` |
+| Use case     | Static pages               | Dynamic compositions  |
+| Key includes | URL + segment hash         | Route params          |
 
 Use document cache for mostly-static pages. Use segment cache when different parts of a page have different cache requirements.

package/skills/fonts/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: fonts
-description: Load and configure web fonts with preload hints for optimal performance
+description: Load and configure web fonts with preload hints for optimal performance. Use when adding a custom font to a Rango app, fonts flash or swap on load (FOUT/FOIT), or you want font preload for performance.
 argument-hint: [provider]
 ---
 
@@ -24,7 +24,11 @@ export function Document({ children }: { children: ReactNode }) {
       <head>
         {/* Preconnect to Google Fonts */}
         <link rel="preconnect" href="https://fonts.googleapis.com" />
-        <link rel="preconnect" href="https://fonts.gstatic.com" crossOrigin="anonymous" />
+        <link
+          rel="preconnect"
+          href="https://fonts.gstatic.com"
+          crossOrigin="anonymous"
+        />
 
         {/* Load font stylesheet */}
         <link
@@ -37,9 +41,7 @@ export function Document({ children }: { children: ReactNode }) {
         <link rel="stylesheet" href={styles} />
         <MetaTags />
       </head>
-      <body>
-        {children}
-      </body>
+      <body>{children}</body>
     </html>
   );
 }

package/skills/handler-use/SKILL.md

@@ -0,0 +1,364 @@
+---
+name: handler-use
+description: Attach default loaders, middleware, parallels, and other use items directly to handlers via handler.use, and compose them with explicit use() at mount sites. Use when a handler needs its own default middleware/loader without repeating use() at every mount site, or deciding between handler-level defaults and per-route use().
+argument-hint: "[handler]"
+---
+
+# Handler-Attached `.use`
+
+A handler function (or branded `Static`/`Prerender`/`Passthrough` definition) can carry its own defaults via a `.use` callback that returns an array of `use` items (loader, middleware, parallel, intercept, layout, loading, etc.). The mount-site DSL (`path()`, `layout()`, `parallel()`, `intercept()`) merges those defaults with any explicit `use()` callback supplied at the registration site.
+
+This lets handlers be **self-contained, reusable units** — a page brings its own loader, a layout brings its own middleware, a parallel slot brings its own data + skeleton — without forcing every caller to wire the same items at every mount site.
+
+Canonical implementation reference:
+[src/route-definition/resolve-handler-use.ts](../../src/route-definition/resolve-handler-use.ts)
+
+## Defining a handler with `.use`
+
+Attach `.use` to the function (or to the branded definition for `Static()`/`Prerender()`/`Passthrough()`):
+
+```typescript
+import {
+  loader,
+  middleware,
+  loading,
+  createLoader,
+  type Handler,
+} from "@rangojs/router";
+
+export const ProductLoader = createLoader(async (ctx) =>
+  fetchProduct(ctx.params.slug),
+);
+
+const ProductPage: Handler<"/product/:slug"> = async (ctx) => {
+  const product = await ctx.use(ProductLoader);
+  return <ProductView product={product} />;
+};
+
+ProductPage.use = () => [
+  loader(ProductLoader),
+  loading(<ProductSkeleton />),
+  middleware(async (ctx, next) => {
+    await next();
+    ctx.headers.set("Cache-Control", "private, max-age=60");
+  }),
+];
+```
+
+Now `ProductPage` carries its loader, loading state, and response-header middleware regardless of where it is mounted.
+
+## Allowed items per mount site
+
+`handler.use()` is the same callback shape regardless of where the handler runs, but the runtime validates that the items it returns are valid for the mount site. Driven by `MOUNT_SITE_ALLOWED_TYPES` in [resolve-handler-use.ts](../../src/route-definition/resolve-handler-use.ts):
+
+| Mount site                                        | Allowed item types                                                                                                                                                                        |
+| ------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `path()` / `route()`                              | `layout`, `parallel`, `intercept`, `middleware`, `revalidate`, `loader`, `loading`, `errorBoundary`, `notFoundBoundary`, `cache`, `transition`                                            |
+| `layout()`                                        | All of the above, plus `route`, `include`                                                                                                                                                 |
+| `parallel()` (per slot)                           | `revalidate`, `loader`, `loading`, `errorBoundary`, `notFoundBoundary`, `transition`                                                                                                      |
+| `intercept()`                                     | `middleware`, `revalidate`, `loader`, `loading`, `errorBoundary`, `notFoundBoundary`, `layout`, `route`, `transition` (conditional activation is config-only: `intercept(..., { when })`) |
+| Response routes (`path.json()`, `path.text()`, …) | `middleware`, `cache`                                                                                                                                                                     |
+
+For per-item semantics see the dedicated skills: [middleware](../middleware/SKILL.md), [loader](../loader/SKILL.md), [parallel](../parallel/SKILL.md), [intercept](../intercept/SKILL.md), [layout](../layout/SKILL.md), [view-transitions](../view-transitions/SKILL.md).
+
+If `handler.use()` returns a disallowed item for a mount site, registration throws:
+
+```
+handler.use() returned middleware() which is not valid inside parallel().
+Allowed types: revalidate, loader, loading, errorBoundary, notFoundBoundary, transition.
+```
+
+The narrowest contract is `parallel()` — slots cannot bring their own middleware or layout; only data, loading, error/notFound boundaries, revalidation, and transitions.
+
+## Composition with explicit `use()`
+
+Every mount site that takes a `use` callback merges in this order:
+
+1. **`handler.use()` items first** — the handler's defaults.
+2. **Explicit `use()` items second** — overrides specified at the mount site.
+
+Items of the same kind from the explicit `use()` follow the existing override rules of that item type. The most important ones for composition:
+
+- **`loading()`** — last definition wins, so explicit `loading()` replaces the handler's default.
+- **`parallel({ "@slot": … })`** — the last `parallel()` call wins per slot name. Other slots from earlier calls are preserved (see `skills/parallel`).
+- **`loader()`, `middleware()`, etc.** — accumulate; both the handler's and the explicit ones run.
+
+Skip the boilerplate: if neither `handler.use` nor explicit `use()` is provided, no merge happens.
+
+```typescript
+// Handler brings a loader + a (placeholder) loading; explicit use replaces loading.
+const SidebarSlot: Handler = async (ctx) => {
+  const data = await ctx.use(SidebarLoader);
+  return <Sidebar data={data} />;
+};
+SidebarSlot.use = () => [
+  loader(SidebarLoader),
+  loading(<DefaultSidebarSkeleton />),
+];
+
+parallel({ "@sidebar": SidebarSlot }, () => [
+  // Replaces the default skeleton; SidebarLoader from handler.use still runs.
+  loading(<SiteSpecificSidebarSkeleton />),
+]);
+```
+
+## Composable parallel slots (the main pay-off)
+
+The parallel slot site is where `handler.use` shines. A slot handler that owns its data/loading lets a layout declare **just** the slot names — every loader, skeleton, and revalidation contract travels with the slot itself.
+
+### Without `handler.use` (every caller wires it up)
+
+```typescript
+layout(<DashboardLayout />, () => [
+  parallel({ "@cart": CartSummary }, () => [
+    loader(CartLoader),
+    loading(<CartSkeleton />),
+    revalidate(revalidateCartData),
+  ]),
+  parallel({ "@notifs": NotificationPanel }, () => [
+    loader(NotificationsLoader),
+    loading(<NotifsSkeleton />),
+    revalidate(revalidateNotifs),
+  ]),
+  path("/dashboard", DashboardIndex, { name: "dashboard.index" }),
+]);
+```
+
+Every layout that wants `@cart` must repeat the same loader/loading/revalidate triplet.
+
+### With `handler.use` (slot owns its dependencies)
+
+```typescript
+const CartSummary: Handler = async (ctx) => {
+  const cart = await ctx.use(CartLoader);
+  return <CartSummaryView cart={cart} />;
+};
+CartSummary.use = () => [
+  loader(CartLoader),
+  loading(<CartSkeleton />),
+  revalidate(revalidateCartData),
+];
+
+const NotificationPanel: Handler = async (ctx) => {
+  const items = await ctx.use(NotificationsLoader);
+  return <NotificationsView items={items} />;
+};
+NotificationPanel.use = () => [
+  loader(NotificationsLoader),
+  loading(<NotifsSkeleton />),
+  revalidate(revalidateNotifs),
+];
+
+// Mount sites become declarative — no per-call data wiring.
+layout(<DashboardLayout />, () => [
+  parallel({ "@cart": CartSummary, "@notifs": NotificationPanel }),
+  path("/dashboard", DashboardIndex, { name: "dashboard.index" }),
+]);
+
+layout(<AccountLayout />, () => [
+  // Same slot, same defaults, zero re-wiring.
+  parallel({ "@cart": CartSummary }),
+  path("/account", AccountIndex, { name: "account.index" }),
+]);
+```
+
+Each slot handler is now a portable, self-contained unit. Different layouts can use the same slot without copying data plumbing.
+
+### Streaming behavior is per-slot
+
+A slot's `loading()` (whether from `handler.use` or explicit) makes that slot an independent streaming unit — its loader does not block the parent layout. Two slot handlers with their own loading skeletons stream independently.
+
+```typescript
+parallel({
+  "@cart": CartSummary, // handler.use loading() → streams independently
+  "@cartBadge": CartBadge, // no loading() anywhere → awaited before paint
+});
+```
+
+### Two scopes for explicit `use` at the mount site: shared (broadcast) and slot-local
+
+`parallel()` accepts an explicit `use()` callback that **broadcasts** to every slot in the call ([dsl-helpers.ts](../../src/route-definition/dsl-helpers.ts)). That's the right behavior for the items the parallel allow-list permits and that accumulate (`loader`, `revalidate`, `errorBoundary`, `notFoundBoundary`, `transition`) — every slot gets them. (Note: `middleware` is not allowed inside `parallel()`; see the allowed-types table above.)
+
+For single-assignment items like `loading()`, broadcasting overwrites every slot's `handler.use` default. Pass a **slot descriptor** `{ handler, use }` instead: items in the descriptor's `use` apply only to that slot.
+
+```typescript
+parallel({
+  "@meta": MetaSlot,
+  "@sidebar": {
+    handler: SidebarSlot,
+    use: () => [loading(<SidebarSkeleton />)], // ← only @sidebar
+  },
+});
+```
+
+Per-slot merge order is **handler.use → shared use → slot-local use** (narrowest scope wins for last-write-wins items like `loading()`):
+
+```typescript
+parallel(
+  {
+    "@cart": {
+      handler: Cart,
+      use: () => [loading(<CartSkeleton />)], // wins for @cart
+    },
+    "@notifs": Notifs, // gets <BroadcastSkeleton />
+  },
+  () => [
+    loader(SharedAnalyticsLoader), // accumulates on every slot
+    loading(<BroadcastSkeleton />), // applies to slots without slot-local
+  ],
+);
+```
+
+Use the descriptor's `use` for `loading(false)` too — opting one slot out of streaming without affecting siblings:
+
+```typescript
+parallel(
+  {
+    "@cart": { handler: Cart, use: () => [loading(false)] }, // @cart awaits
+    "@notifs": Notifs, // @notifs still streams with broadcast skeleton
+  },
+  () => [loading(<BroadcastSkeleton />)],
+);
+```
+
+Rule of thumb: shared `use` is for items that legitimately apply to every slot. Slot-local `use` is for per-slot precision — especially `loading()` and `loading(false)`.
+
+### Replacing a whole slot from a parent's `handler.use`
+
+A handler can publish a default `parallel({...})` set via its `.use`, and the mount site can replace any individual slot by re-declaring it. Last `parallel()` per slot name wins (see `skills/parallel` § Slot Override Semantics).
+
+```typescript
+const ProductPage: Handler<"/product/:slug"> = (ctx) => (
+  <article>
+    <ProductHero slug={ctx.params.slug} />
+    <ParallelOutlet name="@related" />
+    <ParallelOutlet name="@reviews" />
+  </article>
+);
+ProductPage.use = () => [
+  parallel({
+    "@related": DefaultRelatedProducts,
+    "@reviews": DefaultReviews,
+  }),
+];
+
+path("/product/:slug", ProductPage, { name: "product" }, () => [
+  // Override @related only; @reviews keeps the default from handler.use.
+  parallel({ "@related": SiteSpecificRelated }),
+]);
+```
+
+## Other mount sites
+
+### Pages (`path()`)
+
+Page handlers can carry middleware, loaders, error boundaries, parallel slots, etc. — anything from the `path` row of the table above.
+
+```typescript
+const CheckoutPage: Handler<"/checkout"> = async (ctx) => { /* … */ };
+CheckoutPage.use = () => [
+  middleware(requireAuth),
+  loader(CartLoader),
+  errorBoundary(<CheckoutError />),
+  notFoundBoundary(<CheckoutNotFound />),
+];
+```
+
+### Layouts (`layout()`)
+
+Layout handlers can carry middleware that runs for every child route, plus default parallels, includes, etc.
+
+```typescript
+const AdminLayout: Handler = (ctx) => {
+  const user = ctx.get(CurrentUser);
+  return <Admin user={user} />;
+};
+AdminLayout.use = () => [
+  middleware(requireAdmin),
+  parallel({ "@adminNotifs": AdminNotifsSlot }),
+];
+```
+
+### Intercepts (`intercept()`)
+
+Intercept handlers can carry their own middleware chain, loaders, and even nested layouts/routes for the modal shell.
+
+```typescript
+const QuickViewModal: Handler = async (ctx) => {
+  const product = await ctx.use(ProductLoader);
+  return <QuickView product={product} />;
+};
+QuickViewModal.use = () => [
+  loader(ProductLoader),
+  loading(<QuickViewSkeleton />),
+  layout(<ModalChrome />),
+];
+```
+
+## `loading()` is a single-assignment item — scope it correctly
+
+Most `use` items accumulate when merged: `handler.use` `middleware()` runs _and_ explicit `middleware()` runs; both `loader()` registrations apply. `loading()` is different — it mutates `entry.loading` directly, last call wins ([dsl-helpers.ts `loading`](../../src/route-definition/dsl-helpers.ts)).
+
+For pages, layouts, and intercepts that's straightforward: explicit `loading()` at the mount site replaces any `loading()` from `handler.use`. The merge order is `handler.use → explicit`, so the explicit one is the last writer and wins.
+
+For parallel slots, the shared `parallel(..., () => [...])` callback is **broadcast** to every slot in the call. A single `loading()` placed there lands on every slot, overwriting each slot's `handler.use` default. To scope `loading()` to one slot, use the **slot descriptor** form:
+
+```typescript
+const Cart: Handler = async (ctx) => { /* … */ };
+Cart.use = () => [loader(CartLoader), loading(<CartSkeleton />)];
+
+const Notifs: Handler = async (ctx) => { /* … */ };
+Notifs.use = () => [loader(NotifsLoader), loading(<NotifsSkeleton />)];
+
+// ✅ @cart gets a custom skeleton; @notifs keeps its handler.use default.
+parallel({
+  "@cart": {
+    handler: Cart,
+    use: () => [loading(<CustomCartSkeleton />)],
+  },
+  "@notifs": Notifs,
+});
+
+// ✅ Opt one slot out of streaming while siblings still stream the broadcast.
+parallel(
+  {
+    "@cart": { handler: Cart, use: () => [loading(false)] },
+    "@notifs": Notifs,
+  },
+  () => [loading(<BroadcastSkeleton />)],
+);
+```
+
+Per-slot merge order is **handler.use → shared use → slot-local use**. Slot-local is the narrowest scope, so it wins for last-write-wins items like `loading()`. Items that accumulate within the parallel allow-list (`loader`, `revalidate`, `errorBoundary`, `notFoundBoundary`, `transition`) compose across all three layers regardless.
+
+Other things to keep in mind about `loading()`:
+
+- Any `loading()` (regardless of source) makes the segment a streaming unit. A handler that includes `loading()` in its `.use` opts every mount site into streaming by default. To opt back out, pass `loading(false)` at the mount site (`loading: false` handling in [match-middleware/segment-resolution.ts](../../src/router/match-middleware/segment-resolution.ts)) — use the slot descriptor form for parallel slots so the opt-out doesn't broadcast.
+
+Rule of thumb: only put `loading()` in `handler.use` if you genuinely want every mount site to stream by default. Use the slot descriptor's `use` for any per-slot intent at a `parallel()` call.
+
+## Edge cases & gotchas
+
+- **ReactNode handlers cannot have `.use`.** A bare JSX element passed as a handler (e.g., `path("/about", <About />)`) has no function to attach properties to. Pass a function or branded definition instead.
+- **Branded handlers** — `Static()`, `Prerender()`, and `Passthrough()` are positional constructors (not object-arg). Construct first, then attach `.use` to the returned definition:
+
+  ```typescript
+  const ProductPage = Prerender(async (ctx) => {
+    const product = await fetchProduct(ctx.params.slug);
+    return <ProductView product={product} />;
+  });
+  ProductPage.use = () => [loader(ProductLoader)];
+  ```
+
+- **Items can be flat or nested arrays.** `handler.use()` results are flattened with `.flat(3)` before validation, so factory helpers that return arrays inline work the same as in regular `use()` callbacks.
+- **Validation runs at registration / first match**, not at handler definition. A handler doesn't know its mount site at definition time — the same handler used in a `path()` and an `intercept()` is validated against each mount's allowed-types set when registered.
+- **No silent shadowing.** If a disallowed item slips through (e.g., a layout factory returning `cache()` from a slot's `handler.use`), the runtime throws with the offending type and mount site named.
+
+## Cross-references
+
+- `skills/route` — `path()` mount site basics
+- `skills/layout` — `layout()` mount site basics
+- `skills/parallel` — parallel slot semantics, slot override rules, streaming behavior
+- `skills/intercept` — intercept mount site basics
+- `skills/loader` — defining `createLoader` and reading via `ctx.use()`
+- `skills/middleware` — middleware semantics and ordering