React State Machines with XState v5

Overview

State machines make impossible states unrepresentable by modeling UI behavior as explicit states, transitions, and events. XState v5 (2.5M+ weekly npm downloads) unifies state machines with the actor model—every machine is an independent entity with its own lifecycle, enabling sophisticated composition patterns.

When to Use This Skill

Trigger patterns:

• Boolean flag explosion: multiple isLoading, isError, isSuccess flags
• Implicit states: writing if (isLoading && !isError && data) to derive mode
• Defensive coding: guards before state updates to prevent invalid transitions
• Timing coordination: timeouts, delays, debouncing across states
• State dependencies: one state depends on another to update correctly

Do not use for:

• Simple boolean toggles with no async (useState is simpler)
• Single form fields with basic validation (useReducer suffices)
• Server state caching (React Query/TanStack Query handles this)
• Static data transformations (useMemo is better)
• Simple counters or toggles (useState is clearer)

See decision-trees.md for comprehensive decision guidance

Core Mental Model

Finite states represent modes of behavior: idle, loading, success, error. A component can only be in ONE state at a time.

Context (extended state) stores quantitative data that doesn't define distinct states. The finite state says "playing"; context says what at what volume.

Events trigger transitions between states. Events are objects: { type: 'SUBMIT', data: formData }.

Guards conditionally allow/block transitions: { guard: 'hasValidInput' }.

Actions are fire-and-forget side effects during transitions or state entry/exit.

Invoked actors are long-running processes (API calls, subscriptions) with lifecycle management and cleanup.

Quick Start: XState v5 setup() Pattern

import { setup, assign, fromPromise } from 'xstate';

const fetchMachine = setup({
  types: {
    context: {} as { data: User | null; error: string | null },
    events: {} as 
      | { type: 'FETCH'; userId: string }
      | { type: 'RETRY' }
  },
  actors: {
    fetchUser: fromPromise(async ({ input, signal }) => {
      const res = await fetch(`/api/users/${input.userId}`, { signal });
      if (!res.ok) throw new Error(res.statusText);
      return res.json();
    })
  },
  actions: {
    setData: assign({ data: ({ event }) => event.output }),
    setError: assign({ error: ({ event }) => event.error.message })
  }
}).createMachine({
  id: 'fetch',
  initial: 'idle',
  context: { data: null, error: null },
  states: {
    idle: { on: { FETCH: 'loading' } },
    loading: {
      invoke: {
        src: 'fetchUser',
        input: ({ event }) => ({ userId: event.userId }),
        onDone: { target: 'success', actions: 'setData' },
        onError: { target: 'failure', actions: 'setError' }
      }
    },
    success: { on: { FETCH: 'loading' } },
    failure: { on: { RETRY: 'loading' } }
  }
});

React Integration Decision Tree

Basic pattern:

import { useMachine } from '@xstate/react';

function Toggle() {
  const [snapshot, send] = useMachine(toggleMachine);
  return (
    <button onClick={() => send({ type: 'TOGGLE' })}>
      {snapshot.matches('inactive') ? 'Off' : 'On'}
    </button>
  );
}

Performance pattern:

import { useActorRef, useSelector } from '@xstate/react';

const selectCount = (s) => s.context.count;
const selectLoading = (s) => s.matches('loading');

function Counter() {
  const actorRef = useActorRef(counterMachine);
  const count = useSelector(actorRef, selectCount);
  const loading = useSelector(actorRef, selectLoading);
  // Only re-renders when count or loading changes
}

Anti-Patterns to Avoid

❌ State explosion: Flat states for orthogonal concerns. Use parallel states instead.

❌ Sending events from actions: Never send() inside assign. Use raise for internal events.

❌ Impure guards: Guards must be pure—no side effects, no external mutations.

❌ Subscribing to entire state: Use focused selectors with useSelector.

❌ Not memoizing model:

// WRONG
const model = Model.fromJson(layout);  // New model every render

// CORRECT
const modelRef = useRef(Model.fromJson(layout));

Navigation to References

Core Patterns

• xstate-v5-patterns.md: Complete v5 API, statecharts (hierarchy/parallel/history), promise actors
• react-integration.md: useMachine vs useActorRef, Context patterns, side effect handling
• testing-patterns.md: Unit testing, mocking actors, visualization debugging

Decision Making & Best Practices

• decision-trees.md: When to use state machines vs useState/useReducer/React Query, machine splitting strategies
• real-world-patterns.md: Complete examples - auth flows, file uploads, wizards, undo/redo, shopping carts
• error-handling.md: Error boundaries, retry strategies, circuit breakers, graceful degradation
• performance.md: Selector memoization, React.memo integration, machine splitting for performance

Advanced Topics

• persistence-hydration.md: localStorage persistence, SSR/Next.js hydration, snapshot serialization
• migration-guide.md: Step-by-step migration from useState/useReducer with before/after examples
• composition-patterns.md: Actor communication, machine composition, higher-order machines, systemId
• skills-architecture.md: Input/output parameterization, library structure

Key Reminders

1. setup() is the v5 way: Strong TypeScript inference, actor registration, action definitions
2. Invoke for async, actions for sync: Actions are fire-and-forget; invoked actors have lifecycle
3. Finite states for modes, context for data: Don't create states for every data variation
4. Visualize first: Stately Studio (stately.ai/editor) makes machines living documentation

Red Flags

• More than 3-4 boolean flags → Need state machine
• Writing if (a && !b && c) to determine mode → States should be explicit
• Bugs from invalid state combinations → Machine prevents impossible states
• Can't explain state transitions to stakeholders → Visualization solves this

Related Skills

• react: Parent skill for React patterns
• nextjs: Server/client state coordination
• test-driven-development: Test machines with createActor before UI integration