references/components-and-state.md
# Components, state, and interaction
## Component TypeScript
- Extend native elements with `ComponentPropsWithoutRef<'button'>`, add custom props via intersection
- Use `React.ReactNode` for children, `React.ReactElement` for single element, render prop `(data: T) => ReactNode`
- Discriminated unions for variant props -- TypeScript narrows automatically in branches
- Generic components: `<T>` with `keyof T` for column keys, `T extends { id: string }` for constraints
- Event types: `React.MouseEvent<HTMLButtonElement>`, `FormEvent<HTMLFormElement>`, `ChangeEvent<HTMLInputElement>`
- `as const` for custom hook tuple returns
- `useRef<HTMLInputElement>(null)` for DOM (use `?.`), `useRef<number>(0)` for mutable values
- Explicit `useState<User | null>(null)` for unions/null
- useReducer actions as discriminated unions: `{ type: 'set'; payload: number } | { type: 'reset' }`
- useContext null guard: throw in custom `useX()` hook if context is null
- **A parameter type whose properties are all optional is a weak type, and a mismatched argument fails to compile rather than passing `undefined`.** TypeScript requires the argument to share at least one property with a weak type, so `(record: { newField?: boolean })` rejects a generated type that does not yet carry `newField` with `TS2559: Type 'X' has no properties in common with type 'Y'`. Intersect with the base type instead (`T & { newField?: boolean }`) -- the shim then deletes cleanly once the field lands upstream
## Concurrency & Race Classes
Five race classes survive type-checking and unit tests -- hunt each one during review (cleanup/cancellation mechanics: Effect rules above):
| Class | Production signal | Fix |
|-------|-------------------|-----|
| Lifecycle cleanup gap | "state update on unmounted component" warnings, leaks under rapid navigation | Return cleanup from every effect that registers a listener/timer/observer |
| Remount-timing mistake | Async callback mutates state/DOM after route change/unmount (`fetch().then(setData)` resolves post-navigation) | Cancel per the cancellation hierarchy |
| Boolean-as-state for non-binary UI | Contradictory combos (`isLoading: true, error: Error`) | State constant (`'idle' \| 'loading' \| 'success' \| 'error'`) + transition function; invalid states unreachable |
| Stale promise/timer, no cancel path | Promise chain or `setTimeout` holds `setState` after the component moved on | Bind every async op to a cancel mechanism; test the cleanup path |
| Per-element handlers on large lists | N closures/subscriptions per row, stale-closure bugs on rapid re-renders | Delegate: one parent handler + `event.target.closest(...)` when >~50 items or frequent updates |
| Gate keyed on a child's success-only callback | Parent's submit stays locked behind "still loading, try again in a moment" copy that never resolves; only a page reload escapes | Report failures up (`onLoadError`) as well as successes, and split the flag into loading / loaded / failed -- a boolean cannot carry a terminal state. Keep the action blocked in both if proceeding on unknown data is unsafe; the fix is honest copy plus a real in-place retry, not unblocking |
| `inert` toggled from a blur-managed flag | First `Tab` *inside* the subtree sends focus to `<body>`; the next `Tab` restarts at the top of the document and skips the (now inert) subtree entirely | React maps `onBlur` to native `focusout`, which fires on intra-subtree moves, and `focusin`/`focusout` are `DiscreteEventPriority` -- React commits `inert` synchronously *between* the two events, landing it on the already-focused incoming control. Stand down only when focus truly leaves: `if (e.currentTarget.contains(e.relatedTarget)) return;` |
**Focus-ownership rules:**
- Never toggle `inert` on a subtree that currently holds focus. The blunt version has no guard at all -- a scroll-driven `inert={!isVisible}` on a sticky bar, drawer, or collapsing panel strands the user on a control that is invisible, inert, and unactivatable. Hand focus to the equivalent visible control before hiding, and pass `focus({ preventScroll: true })` when the handoff fires from a scroll handler, or `focus()` scrolls its target into view and fights the scroll the user is performing. If the handoff target carries the same focus listeners that feed the guard, `focus()` arms the flag as a side effect of doing its job and the subtree never goes inert again -- make the handoff symmetric (two effects guarding opposite values of one flag) rather than adding an exception to the guard. `aria-hidden` without `inert` fixes double announcement but leaves the duplicate tab stops. Assert on what the user can *do* (does the handler fire, where does `Tab` go), not on `document.activeElement` -- Chromium resolves the unfocusing steps lazily and it reads back inconsistently
- A blur-managed "focus is inside" flag cannot be cleared by a blur that never fires. Headless popover primitives restore focus to the element that *opened* the content on close; when the popover is anchored to an input with no trigger element, that restore no-ops, the library preventDefaults the focus-scope restore, and focus lands on `document.body` -- the input never receives another `blur`, so the flag sticks `true` for the component's lifetime and anything gated on it (a "re-seed local text from `props.value`" effect, for instance) is silently dead. Clear it explicitly in the select handler and on close when `document.activeElement` is not the input. The same design usually adds `onOpenAutoFocus={e => e.preventDefault()}` to avoid stealing typing focus, which leaves the content pointer-only: no trigger to Tab to and no auto-focus in. Add an explicit affordance plus `aria-haspopup`/`aria-expanded`. Browser-dependent -- Chromium and Firefox blur the input on `mousedown`, Safari does not, so "works on my machine" from Safari proves nothing
## State Management
```
Local UI state → useState, useReducer
Shared client state → Zustand (simple) | Redux Toolkit (complex)
Atomic/granular → Jotai
Server/remote data → React Query (TanStack Query)
URL state → nuqs, router search params
Form state → React Hook Form
```
**Key patterns:**
- Zustand: `create<State>()(devtools(persist((set) => ({...}))))` -- use slices for scale, selective subscriptions to prevent re-renders
- React Query: query keys factory (`['users', 'detail', id] as const`), `staleTime`/`gcTime`, optimistic updates with `onMutate`/`onError` rollback
- React Query `isError` means *a fetch failed*, not *there is no data* -- a failed refetch sets `status: 'error'` while retaining the last successful payload, so the usual `isLoading ? spinner : isError ? errorPanel : content` ladder routes a working, fully cached list into the error panel. The defaults compose into it: `refetchOnMount: true` + `staleTime: 0` refetch on every mount, `retry: false` makes one blip terminal, and `gcTime: 5min` keeps the cache alive across a modal's unmount/remount. Gate the branch on data-absence -- `isLoadingError` (`isError && !hasData`), or `isError && derived.length === 0` when the suite mocks the hook and leaves `isLoadingError` undefined. Audit the side effects with it: `useEffect(() => { if (isError) toast(...) }, [isError])` fires over the live list too
- The mirror case has the same tell: under `retry: false` a first load that fails leaves `data` `undefined` forever, so a branch gated on `data !== undefined` folds *failed* into *pending* and renders "Loading…" permanently. Whether it is recoverable is decided by mount topology, not by open/closed state -- a component rendered unconditionally inside a ref-driven popup is mounted for the life of the page, so `refetchOnMount` never fires again. Any remedy that adds a branch on `isError` must itself be gated on data-absence, or it re-introduces the previous bullet. Enumerate all four states before writing either gate -- `isLoading` (first load, no data yet), `isFetching` (any fetch in flight, including a background refetch over a warm cache), `isError` (last fetch failed, cache may still be present), and loaded-and-genuinely-empty -- and state the action in each: `!isFetching` misses the error state, `!isFetching && !isError` mishandles a background refetch of a legitimately empty result. The write path has the same shape, inverted: a fail-closed save gate keyed on `isError` blocks a valid submit whenever cached data is present
- A mutate-scoped `onSuccess` survives an unmount that the mutation itself caused, so "the success toast is dropped on promote/archive/delete" is usually a false finding. The mutation dispatches success on the microtask after the hook-level `onSuccess` await resolves, while query-observer notifications -- the ones that re-render the list and unmount the row -- flush on a zero-delay timer; that is an ordering guarantee, not a race. The documented "callbacks do not fire on unmount" caveat describes a different unmount (navigation, closing a drawer). Read the scheduler and run a known-bad control (remove the observer before dispatch) before accepting the finding
- Never duplicate server data (React Query) in a client store (Zustand)
- Colocate state close to where it's used
references/e2e-testing.md
# E2E Testing with Playwright
> When to read: when authoring Playwright end-to-end tests — directory layout, fixtures, page objects, network mocking, CI integration.
## Directory Structure
```
e2e/
├── playwright.config.ts
├── fixtures/
│ ├── auth.fixture.ts
│ └── test-data.fixture.ts
├── pages/
│ ├── base.page.ts
│ └── <page-name>.page.ts
├── tests/
│ ├── auth/
│ │ └── login.spec.ts
│ └── smoke/
│ └── critical-paths.spec.ts
└── utils/
└── api-helpers.ts
```
Naming: tests `<feature>.spec.ts`, page objects `<page>.page.ts`, fixtures `<concern>.fixture.ts`.
## Configuration
```typescript
import { defineConfig, devices } from '@playwright/test';
export default defineConfig({
testDir: './e2e/tests',
fullyParallel: true,
forbidOnly: !!process.env.CI,
retries: process.env.CI ? 2 : 0,
workers: process.env.CI ? 1 : undefined,
reporter: 'html',
use: {
baseURL: 'http://localhost:5173',
trace: 'on-first-retry',
screenshot: 'only-on-failure',
},
projects: [
{ name: 'setup', testDir: './e2e/fixtures', testMatch: 'auth.fixture.ts' },
{
name: 'chromium',
use: { ...devices['Desktop Chrome'], storageState: 'e2e/.auth/user.json' },
dependencies: ['setup'],
},
],
webServer: {
command: 'npm run dev',
url: 'http://localhost:5173',
reuseExistingServer: !process.env.CI,
},
});
```
## Page Object Model
Tests never use selectors directly -- page objects encapsulate all locators and actions.
```typescript
// e2e/pages/base.page.ts
import { type Page, type Locator } from '@playwright/test';
export abstract class BasePage {
constructor(protected readonly page: Page) {}
abstract goto(): Promise<void>;
async waitForLoad() { await this.page.waitForLoadState('networkidle'); }
get toast(): Locator { return this.page.getByRole('alert'); }
}
// e2e/pages/users.page.ts
export class UsersPage extends BasePage {
readonly createButton: Locator;
readonly searchInput: Locator;
constructor(page: Page) {
super(page);
this.createButton = page.getByRole('button', { name: /create/i });
this.searchInput = page.getByRole('searchbox', { name: /search/i });
}
async goto() {
await this.page.goto('/users');
await this.waitForLoad();
}
async searchFor(query: string) {
await this.searchInput.fill(query);
await this.page.waitForResponse('**/api/users?*');
}
}
```
Rules: locators as public readonly properties, actions as async methods with internal waits, no assertions in page objects, one PO per page.
## Selector Priority
| Priority | Method | Use when |
|----------|--------|----------|
| 1 | `getByRole` | Buttons, links, headings, inputs |
| 2 | `getByLabel` | Form inputs with labels |
| 3 | `getByPlaceholder` | Search inputs |
| 4 | `getByText` | Static text content |
| 5 | `getByTestId` | No accessible selector available |
Never use CSS selectors, XPath, or DOM structure selectors. When adding `data-testid`, use `<action>-<entity>-<type>` pattern: `create-user-btn`.
## Filling Inputs
Prefer `locator.fill(value)` to `page.keyboard.type()`. Synthesised keystrokes drop characters intermittently under a browser-attached session (CDP against an already-running browser), and the driver reports the full string as typed while the DOM holds a short value -- so the assertion that would catch it is the one nobody writes. Rich-text editors whose state lives outside the element's `value` (ProseMirror, Slate, TipTap) ignore programmatic writes and still need `type()`; there, assert `input_value()` (or the editor's own serialized content) after typing and retry on a short read.
## Wait Strategies
Never use `waitForTimeout` or `setTimeout`. Use explicit conditions:
```typescript
await page.getByRole('heading', { name: 'Dashboard' }).waitFor();
await page.waitForURL('/dashboard');
await page.waitForResponse(
(r) => r.url().includes('/api/users') && r.status() === 200,
);
await page.getByTestId('spinner').waitFor({ state: 'hidden' });
```
## Auth State Reuse
Save auth state once, reuse across all tests:
```typescript
// e2e/fixtures/auth.fixture.ts
import { test as setup } from '@playwright/test';
setup('authenticate', async ({ page }) => {
await page.goto('/login');
await page.getByLabel('Email').fill('testuser@example.com');
await page.getByLabel('Password').fill('TestPassword123!');
await page.getByRole('button', { name: /sign in/i }).click();
await page.waitForURL('/dashboard');
await page.context().storageState({ path: 'e2e/.auth/user.json' });
});
```
Tests receive auth state via `storageState` in config projects.
## Test Data & Network
- Tests create own data via API helpers (faster than UI), clean up in `finally` blocks
- Mock responses with `page.route('**/api/path', route => route.fulfill({ ... }))`
- Simulate errors with `route.abort('failed')`
- Wait for responses: `const resp = page.waitForResponse('**/api/users'); await click; await resp;`
## Flaky Test Fixes
| Cause | Fix |
|-------|-----|
| Hardcoded waits | Explicit wait conditions |
| Shared test data | Each test creates its own |
| Animations | `animations: 'disabled'` in config |
| Race conditions | Wait for API responses before assertions |
**Quarantine workflow** -- confirm flakiness before quarantining:
```bash
npx playwright test --repeat-each=10 path/to/test.spec.ts # Confirm flakiness
npx playwright test --retries=3 path/to/test.spec.ts # Check if retries help
```
Mark confirmed flaky tests with an issue reference:
```typescript
test.fixme(true, 'Flaky - Issue #123'); // Always skip
test.skip(!!process.env.CI, 'Flaky in CI only - Issue #123'); // Skip in CI only
```
```bash
npx playwright test --headed --debug # Debug mode
npx playwright show-trace trace.zip # Trace viewer
npx playwright test --ui # Interactive UI
```
references/rendering-and-frameworks.md
# Rendering and framework patterns
## Performance
**Critical -- eliminate waterfalls:**
- `Promise.all()` for independent async operations
- Move `await` into branches where actually needed
- Suspense boundaries to stream slow content
**Critical -- bundle size:**
- Import directly from modules, avoid barrel files (`index.ts` re-exports)
- `next/dynamic` or `React.lazy()` for heavy components
- Defer third-party scripts (analytics, logging) until after hydration
- Preload on hover/focus for perceived speed
- `content-visibility: auto` + `contain-intrinsic-size` on long lists -- skips off-screen layout/paint
**Re-render optimization:**
- Never define a component inside another component. Each parent render creates a new function identity, and React compares element *types* to decide whether to update or replace -- a new type means the whole subtree unmounts and remounts, so local state is lost, effects re-run, and DOM nodes are recreated. Symptoms are behavioral, not slow: an input loses focus on every keystroke, animations restart, scroll position resets. Hoist the component to module scope and pass what it needed via props. The React Compiler does not save this one -- the type identity changes before memoization applies
- Derive state during render, not in effects
- Subscribe to derived booleans, not raw objects (`state.items.length > 0` not `state.items`)
- Functional setState for stable callbacks: `setCount(c => c + 1)`
- Lazy state init: `useState(() => expensiveComputation())`
- `useTransition` for non-urgent updates (search filtering)
- `useDeferredValue` for expensive derived UI
- Don't subscribe to searchParams/state read only in callbacks -- read on demand
- Use ternary (`condition ? <A /> : <B />`), not `&&` for conditionals
- `React.memo` only for expensive subtrees with stable props
- Hoist static JSX outside components
**React Compiler** (React 19): auto-memoizes -- write idiomatic React, remove manual `useMemo`/`useCallback`/`memo`. Enable via `reactCompiler: true` in next.config (non-framework: `babel-plugin-react-compiler`). Keep components pure.
## React 19
- **ref as prop** -- `forwardRef` deprecated. Accept `ref?: React.Ref<HTMLElement>` as regular prop
- **useActionState** -- replaces `useFormState`: `const [state, formAction, isPending] = useActionState(action, initialState)`
- **use()** -- unwrap Promise or Context during render (not in callbacks/effects). Enables conditional context reads
- **useOptimistic** -- `const [optimistic, addOptimistic] = useOptimistic(state, mergeFn)` for instant UI feedback
- **useFormStatus** -- `const { pending } = useFormStatus()` in child of `<form action={...}>`
- **Server Components** -- default in App Router. Async, access DB/secrets directly. No hooks, no event handlers
- **Server Actions** -- `'use server'` directive. Validate inputs (Zod), `revalidateTag`/`revalidatePath` after mutations. **Server Actions are public endpoints** -- always verify auth/authz inside each action, not just in middleware or layout guards
- **`<Activity mode='visible'|'hidden'>`** -- preserves state/DOM for toggled components (experimental)
## Next.js App Router
**File conventions:** `page.tsx` (route UI), `layout.tsx` (shared wrapper), `template.tsx` (re-mounted on navigation, unlike layout), `loading.tsx` (Suspense), `error.tsx` (error boundary), `not-found.tsx` (404), `default.tsx` (parallel route fallback), `route.ts` (API endpoint)
**Rendering modes:** Server Components (default) | Client (`'use client'`) | Static (build) | Dynamic (request) | Streaming (progressive)
**Decision:** Server Component unless it needs hooks, event handlers, or browser APIs. Split: server parent + client child. Isolate interactive components as `'use client'` leaf components -- keep server components static with no global state or event handlers.
**Server → client boundary:** pass only the fields a client component actually uses, not whole ORM rows or fetch objects. Every prop crossing the `'use client'` boundary is serialized into the payload, so a 50-field `user` object read for one field still ships all 50.
**Client-only state that drives first paint** (theme, locale, feature flag, auth hint): reading `localStorage` during render breaks SSR, and reading it in `useEffect` paints the default first, so the correct value arrives one frame later as a visible flash. Set the value on the document with a small synchronous inline script that runs before hydration -- typically writing a `class` or `data-` attribute on `<html>` that CSS already keys on. The script is developer-authored and must never interpolate user, request, or database data; it is the one place `dangerouslySetInnerHTML` is warranted, and only for a literal string.
**Routing patterns:**
- Route groups `(name)` -- organize without affecting URL
- Parallel routes `@slot` -- independent loading states in same layout
- Intercepting routes `(.)` -- modal overlays with full-page fallback
**Caching:**
- `fetch(url, { cache: 'force-cache' })` -- static
- `fetch(url, { next: { revalidate: 60 } })` -- ISR
- `fetch(url, { cache: 'no-store' })` -- dynamic
- Tag-based: `fetch(url, { next: { tags: ['products'] } })` then `revalidateTag('products')`
**Data fetching:**
- Fetch in Server Components where data is used
- Use Suspense boundaries for slow queries
- `React.cache()` for per-request dedup
- `generateStaticParams` for static generation
- `generateMetadata` for dynamic SEO
- Static metadata with `title: { default: 'App', template: '%s | App' }` for cascading page titles
- `after()` for non-blocking side effects (logging, analytics) -- runs after response is sent
- Hoist static I/O (fonts, config) to module level -- runs once, not per request
- Never hold request-scoped or user data in module-level mutable state -- server renders run concurrently in one process, so shared module state leaks across requests (one user's data surfacing in another's response). Hoist only immutable static I/O; keep request data local to the render tree (pass as props)
## Tailwind Integration
For Tailwind v4 configuration, utility patterns, dark mode, and component variants, see the `ia-tailwind-css` skill.
**Class sorting in JSX**: keep Tailwind classes in canonical order (enforce via `eslint-plugin-better-tailwindcss`).
references/test-selection.md
# Test selection
## Testing (Vitest + React Testing Library)
- **Component tests**: Vitest + RTL, co-located `*.test.tsx`. Default for React components.
- **Hook tests**: `renderHook` + `act`, co-located `*.test.ts`
- **Unit tests**: Vitest for pure functions, utilities, services
- **E2E**: Playwright for user flows and critical paths
- **Query priority**: `getByRole` > `getByLabelText` > `getByPlaceholderText` > `getByText` > `getByTestId`
- Mock API services and external providers; render child components real for integration confidence
- One behavior per test with AAA structure. Name: `should <behavior> when <condition>`
- Use `userEvent` over `fireEvent` for realistic interactions
- `findBy*` for async elements, `waitFor` after state-triggering actions
- `vi.clearAllMocks()` in `beforeEach`. Recreate state per test.
- Timing (`useLayoutEffect` vs `useEffect` report races), engine fidelity (jsdom/happy-dom vs a real browser engine for parser/layout-dependent behavior), interaction-mode pitfalls (`userEvent` delay, fake-timer incompatibility, `fireEvent` vs `userEvent` tradeoffs), and runner/environment failures (`vmThreads` OOM, happy-dom swallowing `console.*`): see [testing.md](./testing.md)
General testing discipline (anti-patterns, rationalization resistance): see the `ia-writing-tests` skill.
See [testing patterns and examples](./testing.md) for component, hook, and mocking examples.
See [e2e testing](./e2e-testing.md) for Playwright patterns.
references/testing.md
# Testing React (Vitest + RTL)
> When to read: when adding or fixing component tests with Vitest + React Testing Library — setup, queries, user-event, async assertions, mocking.
## Setup
Vitest config: `environment: 'jsdom'`, `globals: true`, `setupFiles` pointing to a file that imports `@testing-library/jest-dom/vitest`. Use `@vitejs/plugin-react` and mirror path aliases from `tsconfig.json`.
## Component Test
```tsx
import { render, screen, waitFor } from '@testing-library/react';
import userEvent from '@testing-library/user-event';
vi.mock('@/api/client');
describe('UserForm', () => {
beforeEach(() => { vi.clearAllMocks(); });
it('should submit valid form data', async () => {
const onSubmit = vi.fn();
render(<UserForm onSubmit={onSubmit} />);
await userEvent.type(screen.getByLabelText(/email/i), 'test@example.com');
await userEvent.click(screen.getByRole('button', { name: /submit/i }));
await waitFor(() => {
expect(onSubmit).toHaveBeenCalledWith(
expect.objectContaining({ email: 'test@example.com' }),
);
});
});
});
```
## Hook Test
```typescript
import { renderHook, act } from '@testing-library/react';
it('should debounce value updates', () => {
vi.useFakeTimers();
const { result, rerender } = renderHook(
({ value }) => useDebounce(value, 300),
{ initialProps: { value: 'initial' } },
);
rerender({ value: 'updated' });
expect(result.current).toBe('initial');
act(() => { vi.advanceTimersByTime(300); });
expect(result.current).toBe('updated');
vi.useRealTimers();
});
```
## Mocking Patterns
```typescript
// Service mock -- mock the module, not the transport layer
vi.mock('@/server-api/me/me.service', () => ({
MeService: { retrieveMe: vi.fn() },
}));
// QueryClient wrapper for components using TanStack Query
const createWrapper = () => {
const queryClient = new QueryClient({
defaultOptions: { queries: { retry: false } },
});
return ({ children }: { children: React.ReactNode }) => (
<QueryClientProvider client={queryClient}>{children}</QueryClientProvider>
);
};
render(<Component />, { wrapper: createWrapper() });
```
## Test Classification
| Type | Tool | Target | File pattern |
|------|------|--------|-------------|
| Unit | Vitest | Pure functions, utilities, services | Co-located `*.test.ts` |
| Component | Vitest + RTL | React components | Co-located `*.test.tsx` |
| Hook | Vitest + RTL | Custom hooks | Co-located `*.test.ts` |
| E2E | Playwright | User flows, critical paths | Separate `e2e/` directory |
## Running Tests
```bash
npx vitest # Watch mode
npx vitest run # Single run (CI)
npx vitest run src/features/ # Test specific directory
npx vitest --coverage # Coverage report
```
## Timing, Fidelity, and Interaction-Mode Pitfalls
Report from `useLayoutEffect`, not `useEffect`, in a child stub whose report feeds a gate under test. A `useEffect` report lands a tick late: `render()` returns, the test queries and clicks before the report's `setState` has committed, and the click reads the pre-report state. RTL flushes layout effects *and their state updates* synchronously inside `render()`'s `act`. Awaiting a settled-state text papers over it only where there is a positive DOM signal -- a gate-*unsatisfiable* case has none, so it flakes or silently asserts the wrong branch.
jsdom and happy-dom are not evidence about a browser-rendered artifact. Where the behavior is decided by a parser, layout, or print engine -- a serialized DOM re-parsed by Chromium, `inert`, `postMessage` across an opaque origin -- execute it in the *sink's* engine (Playwright/Chromium against the repo's own dependency) and carry a known-bad control in the same batch. jsdom does not enforce the radio-group invariant on parse and does not enforce the opaque-origin `targetOrigin` check, so the natural regression test passes while the bug ships. Engine fidelity and path fidelity are separate requirements: load the committed artifact verbatim and drive it through its real entry point, because a hand-rebuilt reconstruction can invert the result while looking like executed evidence.
Use `userEvent.setup({ delay: null })` in component tests. A bare `setup()` inserts a `setTimeout(0)` macrotask between every simulated keystroke and pointer event -- pure idle time in a test DOM, and the single largest source of a slow suite's tail.
`vi.useFakeTimers()` freezes `findBy`/`waitFor` polling **and** `userEvent` (v14 awaits real timers internally), so the two do not compose. Under fake timers drive with `fireEvent` plus an explicit settle -- `await act(async () => { await vi.runOnlyPendingTimersAsync(); })`, or `advanceTimersByTimeAsync(N)` for a debounce window -- and assert both sides of the boundary: not-called at N-1ms, called once at Nms. A one-sided assertion passes against an eager implementation.
Fill forms with one `fireEvent.change` per field when typing is not the behavior under test. `user.type` into a `mode: 'onChange'` react-hook-form field pays a full-schema re-validation per keystroke, and it focuses the input, which mounts whatever popover is attached (a date field re-renders a 42-day calendar grid on every keystroke). Keep one keystroke-level test per form for validation coverage and convert the rest. `fireEvent` cannot drive components that need a real pointer sequence (a combobox opening on input click, a menu opening on `pointerdown`) -- those keep `userEvent`.
## Runner and Environment Failures
Under `pool: 'vmThreads'` a worker's RSS grows monotonically across files, and the recycler's default threshold is a share of `os.totalmem()` -- the *host's* RAM, not the container's cgroup limit -- so in CI the kernel OOM-kills the worker before Vitest decides to recycle it. Fingerprint: `Worker exited unexpectedly` late in the run, every test that ran reporting green, and **no `Test Files ... | Tests ...` summary line at all**. Set `test.vmMemoryLimit` in the Vitest config (512MB is a sane starting point) on every app using a VM pool; it is a recycle threshold rather than a hard cap, and it is inert on non-VM pools, so it is safe to set repo-wide.
happy-dom installs a virtual console that swallows `console.*`, so debug output written from a test never reaches the runner and a diagnostic that prints nothing looks like a code path that never ran. Write diagnostics through `process.stdout.write`. Any console-reporting safety net (a setup file that fails the suite on an unexpected `console.error`, for instance) must write to stdout the same way, and must throw when the environment hook it depends on is missing rather than silently reporting nothing.
SKILL.md
---
name: react-frontend
class: language
description: >-
React architecture patterns, TypeScript, Next.js, hooks, and testing. Use when
working with React component structure, state management, Next.js routing,
Vitest, React Testing Library, or reviewing React code. For visual design and
aesthetic direction, use frontend-design instead.
paths: "**/*.tsx,**/*.jsx,**/*.ts"
---
# React Frontend
**Verify before implementing**: For App Router patterns, React 19 APIs, or version-specific behavior, look up current docs (Context7 `query-docs` if available, else the framework's official docs via web search) before writing code. Training data may lag current releases.
## Working rules
- Keep derived state in render and user actions in event handlers; use effects for external synchronization.
- Give async work a lifecycle and cancellation policy; represent failure separately from pending and empty data.
- Preserve focus when hiding interactive regions and exercise keyboard navigation in a real browser.
- Validate and authorize every public server action; send only needed fields across server/client boundaries.
- Measure performance changes and test user-visible behavior, not type-checking alone.
## Effects Decision Tree
Effects are escape hatches -- most logic should NOT use effects.
| Need | Solution |
|------|----------|
| Derived value from props/state | Calculate during render (useMemo if expensive) |
| Reset state on prop change | `key` prop on component |
| Respond to user event | Event handler |
| Notify parent of state change | Call onChange in event handler, or fully controlled component |
| Chain of state updates | Calculate all next state in one event handler |
| Sync with external system | Effect with cleanup |
**Effect rules:**
- Never suppress the linter -- fix the code instead
- Use updater functions (`setItems(prev => [...prev, item])`) to remove state dependencies
- Move objects/functions inside effects to stabilize dependencies
- `useEffectEvent` for non-reactive values (e.g., theme in a connection effect)
- Always return cleanup for subscriptions, connections, listeners
- Data fetching cancellation (pick by situation): `AbortController` for fetch; `ignore` flag for non-cancellable promises; React Query handles both automatically
## Discipline
- Simplicity first -- every change as simple as possible, impact minimal code
- Only touch what's necessary -- avoid introducing unrelated changes
- No hacky workarounds -- if a fix feels wrong, step back and implement the clean solution
- Before adding a new abstraction, verify it appears in 3+ places
## References
- [testing.md](./references/testing.md) -- Component, hook, and mocking test examples
- [e2e-testing.md](./references/e2e-testing.md) -- Playwright E2E patterns
## Verify
- TypeScript compiles with zero errors
- No suppressed lint rules (`eslint-disable`, `@ts-ignore`) in new code
- `useEffect` dependency arrays not manually overridden
- No `forwardRef` usage in React 19+ projects (use `ref` prop directly)
## Task-specific references
Read the relevant reference before implementing or reviewing the matching behavior:
- For component types, state ownership, async races, focus, or cached query behavior: [components-and-state.md](./references/components-and-state.md).
- For performance, React APIs, Next.js boundaries, caching, or Tailwind integration: [rendering-and-frameworks.md](./references/rendering-and-frameworks.md).
- For component, hook, browser, or integration test changes: [test-selection.md](./references/test-selection.md).
Existing specialized references, when the corresponding topic applies:
SPEC.md
# ia-react-frontend Specification
## Intent
`ia-react-frontend` is a `language`-class skill (stack-specific patterns and idioms). React architecture patterns, TypeScript, Next.js, hooks, and testing. Use when working with React component structure, state management, Next.js routing, Vitest, React Testing Library, or reviewing React code. For visual design and aesthetic direction, use frontend-design instead.
## Scope
In scope:
- Behaviors described in `SKILL.md` and routed via the should_trigger phrasings in `distillery/tests/fixtures/triggers/ia-react-frontend.jsonl`.
- Updates to runtime behavior, structure, trigger precision, references, and validation.
Out of scope:
- Acting as the runtime instructions themselves (those live in `SKILL.md`).
- Trigger phrasings already covered by adjacent `ia-*` skills (`validate-plugin` flags >70% description overlap as DUPLICATE_TRIGGER).
- <!-- to fill in: domain-specific exclusions when the skill drifts -->
## Trigger Context
- Class: `language`
- Hook regex: `plugins/whetstone/hooks/skill-patterns.sh` -> `SKILL_PATTERNS[ia-react-frontend]`
- Common requests (from fixture should_trigger):
- "create a React component with state management"
- "create a React hook for form validation"
- "fix the Next.js routing issue"
- Should not trigger for (from fixture should_not_trigger):
- "write a Laravel migration for the orders table"
- "optimize the database indexes"
- "write a bash script for deployment"
## Source And Evidence Model
Authoritative sources:
- `SKILL.md` -- runtime instructions and reference routing.
- `references/*.md` -- bundled supplementary content (2 file(s)).
- `distillery/tests/fixtures/triggers/ia-react-frontend.jsonl` -- positive and negative trigger phrasings under regression test.
- `plugins/whetstone/hooks/skill-patterns.sh` -- regex pattern that fires this skill.
- `distillery/.eval-data/ia-react-frontend/` -- harvested session examples (when present).
Data that must not be stored in this skill or its references:
- Secrets, credentials, tokens.
- Machine-specific filesystem paths (`/home/...`, `/Users/...`, `~/ai/...`). The validator (`MACHINE_PATH_LEAK`) flags these as HIGH.
- Private URLs, customer data, or unredacted personal information.
### Coverage matrix
| Dimension | Status | Evidence |
|---|---|---|
| Trigger fixtures | complete | distillery/tests/fixtures/triggers/ia-react-frontend.jsonl (>=5 should_trigger, >=5 should_not_trigger) |
| Hook regex pattern | complete | plugins/whetstone/hooks/skill-patterns.sh (`SKILL_PATTERNS[ia-react-frontend]`) |
| Reference architecture | complete | 2 file(s) under references/ |
| Real-usage signal | <!-- populated by harvest-sessions when sessions exist --> | distillery/.eval-data/ia-react-frontend/ (created by harvest-sessions) |
## Evaluation
Lightweight (run on every change):
```bash
python3 distillery/scripts/distiller.py validate-plugin --component ia-react-frontend
python3 distillery/scripts/distiller.py test-triggers --skill ia-react-frontend
```
Deeper (when behavior risk warrants):
```bash
python3 distillery/scripts/distiller.py dspy-eval ia-react-frontend
python3 distillery/scripts/distiller.py diagnose-negatives ia-react-frontend
```
Acceptance gates:
- `validate-plugin --component ia-react-frontend` returns 0 HIGH findings.
- `test-triggers --skill ia-react-frontend` returns F1 = 1.0 with floors of 5 should_trigger and 5 should_not_trigger.
- For dspy-eval, the composite score does not regress against the most recent saved baseline (see `distillery/.eval-data/ia-react-frontend/history.json`).
## Known Limitations
<!-- to fill in over time as drift surfaces. Default rule: any time diagnose-negatives
surfaces a recurring failure pattern, document it here so future maintainers
understand the trade-off the current implementation accepts. -->
## Maintenance Notes
- Update `SKILL.md` when the runtime workflow, branch conditions, or output contract changes.
- Update this `SPEC.md` when intent, scope, evidence model, evaluation gates, or maintenance expectations change.
- Update the trigger fixture when adding new positive phrasings, removing stale ones, or expanding scope (the 5/5 floor is a hard validator gate).
- Update the hook regex in `skill-patterns.sh` whenever fixture positives expose a missed phrasing; verify F1 = 1.0 with `eval-triggers` before committing.
- Run the full release pipeline via `/release` -- never bump versions or update CHANGELOG.md from a per-skill edit.