HyperFlow, Linting and Examples
HyperFlow, Linting and Examples
See your sockets, lint 16x faster, start from 88 apps.
client.addPlugin(
DevtoolsPlugin({
appName: "My App",
sockets: [socket],
}),
);
A library is only as good as the tools around it. In the v10 cycle HyperFlow learned to inspect sockets and Node.js servers, the lint rules got an Oxlint plugin and a fix for typescript-eslint 8, and the docs gained 88 runnable examples and agent skills for every package.
This post covers HyperFetch 10. Read the release overview for the rest of the series.
- Socket Inspector: HyperFlow shows every socket's connection, messages and errors.
- Server devtools: the devtools plugin reports from Node.js, Deno and Bun.
- Live observers: HyperFlow shows how many subscriptions each cache key really has.
- Oxlint plugin: the HyperFetch lint rules for Oxlint, with the same rule ids.
- ESLint fixes: the rules report on typescript-eslint 8 and ship a flat recommended config.
- 88 examples: runnable apps with StackBlitz and CodeSandbox previews.
- Agent skills: one skill per package for coding agents.
Socket Inspector
Requests were always visible in HyperFlow, but the realtime side of an app was not. When a chat stops updating, the
first question is whether the socket is still connected, and the answer used to live in console.log.
Pass your sockets to the devtools plugin with the new sockets option:
import { createClient } from "@hyper-fetch/core";
import { Socket } from "@hyper-fetch/sockets";
import { DevtoolsPlugin } from "@hyper-fetch/plugin-devtools";
const socket = new Socket({ url: "ws://localhost:3000" });
const client = createClient({ url: "http://localhost:3000" }).addPlugin(
DevtoolsPlugin({
appName: "My App",
sockets: [socket],
}),
);
The new Sockets page lists each socket with its URL, adapter and status (connecting, connected, reconnecting, disconnected or failed), and counts the messages, bytes, errors, failed sends and reconnection attempts. Below it, a log shows received and sent messages, connection events and errors, with search, a direction filter and per-socket selection. It works with the WebSocket and Server-Sent Events adapters.
When HyperFlow restarts, the plugin reports the sockets and their state again, so you do not lose them. Events waiting for HyperFlow are capped at the newest 1000, so an app running without HyperFlow open no longer keeps every event in memory.
Devtools on the server
Backends send requests too, and debugging them meant adding logs. The devtools plugin now uses the runtime's global
WebSocket when there is no window, so it reports from Node.js 22+, Deno and Bun with no extra setup. Use
environment to tell the backend apart from the frontend in HyperFlow:
import { createClient } from "@hyper-fetch/core";
import { DevtoolsPlugin } from "@hyper-fetch/plugin-devtools";
const client = createClient({ url: "http://localhost:3000" }).addPlugin(
DevtoolsPlugin({
appName: "My API",
environment: "backend",
}),
);
On Node.js 18 and 20, install ws and pass its class in the WebSocket option.
Invalidating a cache entry from HyperFlow now marks the data stale in your app, the same as cache.invalidate(). Before,
it only notified listeners and the data stayed fresh.
Real observer counts
HyperFlow mirrors your app's cache with its own client, and that client has no subscriptions, so the observers column always showed 0. The plugin now reports how many subscriptions your app has per cache key and query key, on every change and as a snapshot each time HyperFlow connects. You can see which data is still on screen and which is safe to drop.
A tested compatibility table
HyperFlow inspects requests from any adapter, but "any" is a claim. Every HyperFlow flow now runs end to end against each adapter: fetch, axios, GraphQL, tRPC, oRPC, Firebase, Firebase Admin and a custom one. A real client reports through the devtools plugin to a real HyperFlow, and the compatibility page shows the results. Every mark there is a test that runs on each change.
Writing those tests found real bugs, and they are fixed in the current build:
- The server survives crashes, invalid ports, restarts and reopened windows.
- Canceled requests are counted, and paused requests get their own icon.
- Plain string, number and boolean values can be edited in the cache view.
- Dialogs, crash badges and queue actions reset to the real state of your app.
- Charts label their legends and draw every particle, even when one leaves the canvas.
The download page now finds the newest release that carries HyperFlow installers and offers the one for your system in one click.
Oxlint plugin
The lint rules catch what TypeScript lets through: every generic key on createClient and createRequest is optional,
so a typo like rsponse silently types your data as unknown. Teams moving to Oxlint lost that
check. oxlint-plugin-hyper-fetch brings both rules over with the same rule ids:
npm install -D oxlint oxlint-plugin-hyper-fetch
{
"jsPlugins": ["oxlint-plugin-hyper-fetch"],
"rules": {
"hyper-fetch/client-generic-types": "error",
"hyper-fetch/request-generic-types": "error"
}
}
The rules skip files that cannot contain a HyperFetch client or request before Oxlint builds their AST, which is nearly all of the cost. On the 4960 files of the HyperFetch repository it lints in 1.09s, against 17.8s for ESLint, with the same findings. Moving from ESLint only changes how the plugin is registered.
ESLint plugin fixes
eslint-plugin-hyper-fetch read generics from typeParameters, which typescript-eslint 8 removed in favor of
typeArguments. On current parsers the rules found no generics and reported nothing. They now read typeArguments and
fall back for older parsers.
configs.recommended is now a flat config object, the format of ESLint 9 and the only one ESLint 10 reads:
import hyperFetch from "eslint-plugin-hyper-fetch";
import tseslint from "typescript-eslint";
export default [
{
files: ["**/*.{ts,tsx,mts,cts}"],
languageOptions: { parser: tseslint.parser },
},
hyperFetch.configs.recommended,
];
If you still use .eslintrc on ESLint 8, switch to the preset that moved:
module.exports = {
extends: ["plugin:hyper-fetch/recommended"],
extends: ["plugin:hyper-fetch/recommended-legacy"],
};
88 runnable examples
Reading an API reference tells you what exists. An app that runs tells you how the pieces fit. The docs now have an examples catalog with 88 standalone apps in ten groups: getting started, data fetching, caching, mutations, request control, offline, realtime, integrations, frameworks and complete apps.
Each example has a page with its real source files and a click-to-run preview, and opens in StackBlitz or CodeSandbox straight from GitHub. CI type-checks and builds every one of them against the current source, so the code you read works with the latest release. To copy one locally:
npx tiged BetterTyped/hyper-fetch/examples/offline/offline-first my-app
The offline-first app is the flagship: it persists both the cache and the request queue, so a whole offline session survives a reload.
Agent skills
Coding agents write a lot of data-fetching code, and they write it from what they remember about a library. Every HyperFetch package now has a public agent skill with setup, golden rules, patterns, pitfalls and type safety, each checked against the package source:
npx skills add BetterTyped/hyper-fetch/skills
Or install only the packages you use:
npx skills add BetterTyped/hyper-fetch/skills --skill hyper-fetch-core --skill hyper-fetch-react
There are fifteen skills, one each for core, React, sockets, axios, GraphQL, tRPC, oRPC, Firebase, Firebase Admin, Hono, Elysia, the CLI, devtools, ESLint and Oxlint.
CLI fixes
- Generated SDKs call the real paths of the schema. The generator turns path segments into camelCase keys, so
sdk.pet.findByStatus.$getused to call/pet/find-by-status. The generated file now passespathSegmentsfor every key the conversion would get wrong. initwrites aliases that point to the directories it creates, andgeneratecreates the output directory when it is missing. Before, a project without tsconfigpathsfailed its firstgeneratewithENOENT.

