Web Meta Framework Remix

Remix / React Router v7 Framework Patterns

Quick Guide: A route exports a loader for reads and an action for writes, both server-only, so a component never fetches its own data. Forms submit without JavaScript and nested routes load in parallel. The version fact that changes every example below: React Router v7 deprecates json() and defer() — return raw objects and raw Promises, and use data() only when you need a custom status or header.

Detailed Resources:

• examples/core.md — the routes directory, file naming, a route end to end
• examples/loaders.md — auth, pagination, search params, caching headers
• examples/actions.md — validation, accessible error display, delete with confirmation
• examples/forms.md — several forms in one route through the intent field
• examples/nested-routes.md — layouts, index routes, pathless layouts
• examples/error-handling.md — error boundaries that branch on status
• examples/optimistic.md — optimistic UI and debounced search with useFetcher
• examples/deferred.md — streaming with Suspense and Await
• examples/resource-routes.md — JSON APIs, webhooks, file downloads
• examples/meta.md — titles, Open Graph, Twitter cards, canonical URLs
• examples/react-router-v7.md — routes.ts, generated types, clientAction, Single Fetch
• reference.md — decision trees, route module exports, hooks, response utilities

Which path applies

• On Remix v2 — the examples in this skill are written for it: @remix-run/* imports, json(), defer() and useLoaderData<typeof loader>().
• On React Router v7 framework mode — the concepts carry over unchanged and the API does not. Read examples/react-router-v7.md first and translate as you go; the mapping table is below.
• Building a page — a loader, a default export, and an ErrorBoundary; follow examples/core.md.
• Building an endpoint — omit the default export and the route becomes a resource route; follow examples/resource-routes.md.

Remix v2 to React Router v7

Remix has merged into React Router v7. What was planned as Remix v3 is React Router v7's "framework mode".

Migration guide: Upgrading from Remix

Before writing Remix code

Export loader and action from route modules only. The build wires them up by route, so the same export in a helper file is dead code that silently never runs.

Throw a Response for an expected failure — 404, 403 — and let the ErrorBoundary render it. Returning null instead pushes a null check onto every consumer and loses the status code.

Await the data the page cannot render without, and return the rest as Promises. Anything awaited delays the first byte; anything returned as a Promise streams in behind it.

Name HTTP status codes as constants. HTTP_NOT_FOUND says what the branch is for, where 404 has to be recognised.

Auto-detection: loader, action, clientLoader, clientAction, useLoaderData, useActionData, useFetcher, useNavigation, useRouteError, isRouteErrorResponse, ErrorBoundary, HydrateFallback, shouldRevalidate, defer, Await, Outlet, NavLink, meta function, links function, @remix-run/node, @remix-run/react, react-router, ServerRouter, HydratedRouter, Route.LoaderArgs, Route.ComponentProps, routes.ts, Single Fetch, app/routes/

Applies to:

• File-based routing, including nested layouts, index routes and pathless layouts
• Server-side data loading in loaders, and mutations in actions
• Forms that work before hydration and better after it
• Streaming non-critical data behind Suspense and Await
• Non-navigating mutations, optimistic UI and debounced search with useFetcher
• Route-scoped error boundaries that branch on HTTP status
• SEO through meta, and stylesheets and preloads through links
• Resource routes: JSON APIs, webhooks, file downloads

Handled elsewhere:

• Persistence — a loader queries and an action writes; neither the client nor the query shape is settled here
• Session and password handling — this skill's examples call an auth layer and read what it returns
• Schema validation — an action parses FormData; which library defines the schema is a separate choice
• Styling — links returns stylesheet descriptors, and what is in them is someone else's concern
• React component authoring itself, outside the route module's own exports

Philosophy

Remix collapses full-stack development to one mental model: a route exports a loader for reads and an action for writes, and both run only on the server. That is what lets a route query a database directly without a secret reaching the browser, and it is why there is no client-side fetching library in the picture.

Four things follow:

1. No fetch waterfalls — loaders run before the component renders, and nested loaders run in parallel with each other rather than in sequence down the tree
2. Progressive enhancement is the default — a <Form> is a real form; JavaScript makes the submission smoother rather than making it possible
3. HTTP semantics rather than framework ones — caching is Cache-Control, errors are status codes, and requests and responses are the platform's own objects
4. The URL is the state — a nested URL maps to a nested component tree, and search params are where filter and pagination state lives

URL change -> loaders run in parallel -> component renders -> user submits
                                                                  |
                                              action runs -> loaders revalidate

Core patterns

Pattern 1: File-Based Routing

Files in app/routes/ become URLs; the naming characters control nesting and dynamic segments.

// app/routes/blog.$slug.tsx
const HTTP_NOT_FOUND = 404;

export async function loader({ params }: LoaderFunctionArgs) {
  const post = await getPostBySlug(params.slug);
  if (!post) throw new Response("Not Found", { status: HTTP_NOT_FOUND });
  return { post };
}

Full code: examples/core.md and examples/nested-routes.md

Pattern 2: Loaders

A loader runs on the server for the initial render and over fetch on every client navigation afterwards.

export async function loader({ params, request }: LoaderFunctionArgs) {
  const user = await getUser(params.userId);
  if (!user) {
    throw json({ message: "User not found" }, { status: HTTP_NOT_FOUND });
  }
  return json({ user });
}

Parent loaders re-run when a child route changes — shouldRevalidate is the opt-out.

Full code: examples/loaders.md

Pattern 3: Actions

An action handles every non-GET method, runs before the loaders, and the loaders revalidate after it.

export async function action({ request }: ActionFunctionArgs) {
  const formData = await request.formData();

  switch (formData.get("intent")) {
    case "update":
      return json({ success: true });
    case "delete":
      return redirect("/items");
    default:
      throw new Error(`Unknown intent`);
  }
}

A hidden intent field is how one route serves several forms. Redirect after a successful mutation so a refresh does not resubmit.

Full code: examples/actions.md and examples/forms.md

Pattern 4: Streaming

Await what the page needs; hand back the rest as Promises.

// Remix v2
return defer({ user, analytics: getAnalytics() });

// React Router v7 — Single Fetch streams a raw Promise
return { user, analytics: getAnalytics() };

<Suspense fallback={<Skeleton />}>
  <Await resolve={analytics} errorElement={<p>Failed to load</p>}>
    {(data) => <Chart data={data} />}
  </Await>
</Suspense>

Stream analytics, comments and recommendations. Await auth state, the page title and anything a crawler reads.

Full code: examples/deferred.md

Pattern 5: useFetcher

A fetcher submits or loads without navigating, which is what inline interactions need.

const fetcher = useFetcher();

// Optimistic UI: the in-flight submission is already in fetcher.formData
const optimisticIsLiked = fetcher.formData
  ? fetcher.formData.get("liked") === "true"
  : isLiked;

<Form> for anything that should change the URL; useFetcher for likes, toggles, inline edits and autocomplete.

Full code: examples/optimistic.md

Pattern 6: Error Boundaries

An exported ErrorBoundary catches everything below it in the route tree, and the first branch tells a thrown Response from an unexpected exception.

export function ErrorBoundary() {
  const error = useRouteError();

  if (isRouteErrorResponse(error)) {
    return (
      <div role="alert">
        <h1>{error.status}</h1>
      </div>
    );
  }

  return (
    <div role="alert">
      <h1>Unexpected Error</h1>
    </div>
  );
}

The boundary is route-scoped, so the rest of the page keeps working.

Full code: examples/error-handling.md

Pattern 7: Meta and Links

meta builds the head tags from the loader's data; links declares stylesheets and preloads.

export const meta: MetaFunction<typeof loader> = ({ data }) => {
  if (!data) return [{ title: "Not Found" }];
  return [
    { title: `${data.post.title} | ${SITE_NAME}` },
    { property: "og:title", content: data.post.title },
    { tagName: "link", rel: "canonical", href: url },
  ];
};

meta receives undefined data when the loader threw, and links cannot see loader data at all — a dynamic <link> goes through meta with tagName: "link".

Full code: examples/meta.md

Pattern 8: Resource Routes

A route module with no default export renders nothing and returns whatever its loader or action does.

// app/routes/api.health.ts
export async function loader() {
  return json({ status: "healthy", timestamp: new Date().toISOString() });
}

Full code: examples/resource-routes.md

Pattern 9: Nested Routes

Nested routes share the parent's layout and load alongside it rather than after it, so an auth check in a parent layout protects every child.

Full code: examples/nested-routes.md

Red flags

Breaks at runtime:

• A loader or action exported from a file that is not a route module — nothing calls it
• <Form> with no method="post" — it submits as a GET, so the action never runs
• defer() without a <Suspense> and <Await> around the consumer
• A form action targeting an index route without ?index — the parent's action receives it
• meta reading data without a null branch — data is undefined when the loader threw
• A secret in a module a route component imports — route modules are bundled for the browser, unlike loaders and actions

Surprising behaviour:

• Every parent loader re-runs on a child navigation, so an expensive parent query runs far more often than expected — shouldRevalidate is the control
• All loaders revalidate after an action, whether or not they relate to what it changed
• links cannot access loader data, so a dynamic stylesheet URL has to go through meta
• On React Router v7, clientAction takes priority when both exist, and the server action is skipped unless the clientAction calls it
• Returning null from a loader compiles and type-checks, and pushes the failure to whatever renders it
• A useFetcher with no optimistic read of fetcher.formData shows nothing until the round trip finishes