Frontend Feature Flags

Feature Flags

Purpose

Deliver features gradually. Run A/B tests. Toggle features without redeploys. Roll back instantly.

Agent Protocol

Trigger

Exact user phrases: "feature flags", "LaunchDarkly", "split testing", "A/B testing", "gradual rollout", "canary release", "feature toggle", "flag provider", "targeting rule".

Input Context

Before activating, verify:

• The flag provider (LaunchDarkly, Split.io, custom, etc.).
• Whether the use case is release toggle, A/B test, or operational flag.

Output Artifact

No file output. Produces flag configuration, provider integration, or rollout strategy code as text.

Response Format

Flag: {name}
Type: {boolean/string/number/json}
Strategy: {targeting rule}
Integration: {code block}

No preamble. No postamble. No explanations. No filler/hedging/transitions. Compress output — why use many token when few do trick.

Completion Criteria

• Flags defined with types — not raw strings throughout codebase.
• Provider client initialized once at app root with user context.
• Gradual rollout with percentage targeting configured.
• A/B test flags include exposure tracking event.
• Fallback/default value provided for every flag evaluation.
• Dead flags cleaned up (no flag code older than 2 releases).

Max Response Length

4096 tokens.

Feature Flag Architecture / Decision Trees

Flag Provider Decision Tree

Need real-time flag updates without page reload?
  |-- YES --> LaunchDarkly (SSE-based, best in class) or Flagsmith
  |-- NO --> Do we have existing LaunchDarkly infra?
  |     |-- YES --> Use LaunchDarkly
  |     |-- NO --> Budget/scale constraints?
  |           |-- SMALL --> Custom flag backend (API + localStorage cache)
  |           |-- MEDIUM --> Flagsmith (open-source option available)
  |           |-- LARGE --> LaunchDarkly or Split.io

Flag Type Decision Tree

What does this flag control?
  |-- Show/hide a UI element --> boolean flag
  |     Example: newCheckoutFlow: boolean
  |
  |-- Numerical configuration --> number flag
  |     Example: maxCartItems: number
  |
  |-- Which layout/copy to show --> string flag
  |     Example: homepageLayout: "legacy" | "redesign-v1" | "redesign-v2"
  |
  |-- Complex configuration object --> json flag
        Example: promoBanner: { enabled: boolean, text: string, ctaUrl: string }

Rollout Strategy Decision Tree

What is the audience for this flag?
  |-- Internal testing only -->
  |     Strategy: Target by email domain (@company.com) or SSO group
  |     Proceed when: Internal team validates feature
  |
  |-- Beta testers -->
  |     Strategy: Target by user IDs in beta list
  |     Proceed when: Beta feedback collected, blockers resolved
  |
  |-- Staged public rollout -->
  |     Strategy: Percentage rollout
  |     Phases: 5% -> 10% -> 25% -> 50% -> 100%
  |     Proceed when: Error rate < 0.1% at each phase
  |
  |-- A/B experiment -->
        Strategy: 50/50 split with randomization consistent per user
        Duration: Minimum 1 week, determined by sample size calculator
        Decision: Analyze after reaching statistical significance (p < 0.05)

Workflow

Step 1: Typed Flag Definitions

// Single source of truth for all flags
export const flags = {
  newCheckoutFlow: { key: 'new-checkout-flow', type: 'boolean', default: false },
  maxItems:        { key: 'max-cart-items', type: 'number', default: 10 },
  homepageLayout:  { key: 'homepage-layout', type: 'string', default: 'legacy' },
  promoBanner:     { key: 'promo-banner', type: 'json', default: { enabled: false, text: '' } },
} as const

type FlagKey = keyof typeof flags

Step 2: Provider-Agnostic Client Abstraction

interface FlagClient {
  getBoolean(key: string, defaultVal: boolean): boolean
  getNumber(key: string, defaultVal: number): number
  getString(key: string, defaultVal: string): string
  getJson(key: string, defaultVal: unknown): unknown
  track(event: string, data?: Record<string, unknown>): void
}

// LaunchDarkly implementation
class LaunchDarklyClient implements FlagClient {
  private client: LDClient

  constructor(env: { clientSideId: string; user: LDUser }) {
    this.client = initialize(env.clientSideId, env.user)
  }

  getBoolean(key: string, defaultVal: boolean): boolean {
    return this.client.variation(key, defaultVal)
  }

  getNumber(key: string, defaultVal: number): number {
    return this.client.variation(key, defaultVal)
  }

  getString(key: string, defaultVal: string): string {
    return this.client.variation(key, defaultVal)
  }

  getJson(key: string, defaultVal: unknown): unknown {
    return this.client.variation(key, defaultVal)
  }

  track(event: string, data?: Record<string, unknown>) {
    this.client.track(event, data)
  }
}

Step 3: React Integration

const FlagContext = createContext<FlagClient | null>(null)

function FlagProvider({ client, children }: { client: FlagClient; children: React.ReactNode }) {
  return <FlagContext.Provider value={client}>{children}</FlagContext.Provider>
}

function useFlag(key: FlagKey): boolean {
  const client = useContext(FlagContext)!
  const def = flags[key]
  return client.getBoolean(def.key, def.default as boolean)
}

function useFlagValue<T>(key: FlagKey): T {
  const client = useContext(FlagContext)!
  const def = flags[key]
  return client.getJson(def.key, def.default) as T
}

// Usage
function CheckoutPage() {
  const useNewFlow = useFlag('newCheckoutFlow')
  return useNewFlow ? <NewCheckout /> : <LegacyCheckout />
}

Step 4: Gradual Rollout

// LaunchDarkly targeting rule (JSON in dashboard, or via config)
{
  "rules": [
    {
      "clauses": [
        {
          "attribute": "email",
          "op": "endsWith",
          "values": ["@internal.com"]
        }
      ],
      "variation": 1  // target internal users first
    },
    {
      "clauses": [
        {
          "attribute": "country",
          "op": "in",
          "values": ["US", "CA"]
        }
      ],
      "variation": 0  // roll out to US/CA at 10%
    }
  ],
  "variations": [false, true],
  "salt": "new-checkout-rollout-v1"
}

Step 5: A/B Test Tracking

function useABTest(flagKey: FlagKey, exposureName: string): boolean {
  const client = useContext(FlagContext)!
  const def = flags[flagKey]
  const value = client.getBoolean(def.key, def.default as boolean)

  useEffect(() => {
    client.track(exposureName, {
      flagKey: def.key,
      variation: value ? 'treatment' : 'control',
    })
  }, []) // fire once on mount

  return value
}

// Usage with experiment tracking
function PricingPage() {
  const showNewPricing = useABTest('newPricingModel', 'pricing-page-exposure')
  return showNewPricing ? <NewPricing /> : <CurrentPricing />
}

Step 6: Flag Cleanup Checklist

• Remove flag evaluation code after feature is 100% rolled out.
• Delete dead flag keys from provider dashboard.
• Keep flag definition file as source of truth — delete the entry.
• Run codemod to remove conditional branches: git grep 'useFlag\|useABTest'.

Step 7: Custom Flag Backend (No Provider)

// Minimal in-memory + localStorage backend
class LocalFlagClient implements FlagClient {
  private flags: Map<string, unknown> = new Map()
  private listeners: Set<() => void> = new Set()

  constructor(private configUrl: string) {}

  async initialize() {
    const response = await fetch(this.configUrl)
    const data = await response.json()
    Object.entries(data).forEach(([key, value]) => {
      this.flags.set(key, value)
    })
    // Persist to localStorage for offline tolerance
    localStorage.setItem('feature-flags-cache', JSON.stringify(data))
  }

  getBoolean(key: string, defaultVal: boolean): boolean {
    return (this.flags.get(key) ?? defaultVal) as boolean
  }

  getNumber(key: string, defaultVal: number): number {
    return (this.flags.get(key) ?? defaultVal) as number
  }

  getString(key: string, defaultVal: string): string {
    return (this.flags.get(key) ?? defaultVal) as string
  }

  getJson(key: string, defaultVal: unknown): unknown {
    return this.flags.get(key) ?? defaultVal
  }

  track(event: string, data?: Record<string, unknown>) {
    console.log('[FlagTrack]', event, data)
    // Send to analytics
  }
}

Step 8: Testing with Feature Flags

// Mock flag client for tests
class MockFlagClient implements FlagClient {
  private overrides: Map<string, unknown> = new Map()

  setFlag(key: string, value: unknown) {
    this.overrides.set(key, value)
  }

  getBoolean(key: string, defaultVal: boolean): boolean {
    return (this.overrides.get(key) ?? defaultVal) as boolean
  }
  // ... other methods
}

// Test
it('shows new checkout when flag is enabled', () => {
  const mockClient = new MockFlagClient()
  mockClient.setFlag('new-checkout-flow', true)

  render(
    <FlagProvider client={mockClient}>
      <CheckoutPage />
    </FlagProvider>
  )

  expect(screen.getByTestId('new-checkout')).toBeInTheDocument()
})

Step 9: SSR / Next.js Integration

// Server-side flag evaluation (Next.js App Router)
import { cookies } from 'next/headers'

async function getServerFlags() {
  const cookieStore = await cookies()
  const userId = cookieStore.get('user_id')?.value

  // Evaluate flags server-side
  const flags = await evaluateFlags(userId, [
    { key: 'new-checkout-flow', default: false },
  ])

  return flags
}

// Pass flags to client via RSC or inline script
export default async function Page() {
  const flags = await getServerFlags()
  return <ClientShell serverFlags={flags} />
}

Common Pitfalls

1. Raw String Keys Throughout Codebase

// BAD -- scattered string keys
if (ldClient.variation('new-checkout-flow', false)) { }

// GOOD -- typed flag definitions
if (useFlag('newCheckoutFlow')) { }

2. No Fallback Value

// BAD -- no default, crashes if provider unavailable
ldClient.variation('my-flag')

// GOOD -- always provide default
ldClient.variation('my-flag', false)

3. Provider Imported Directly in Components

Direct SDK imports couple components to a specific provider. Always wrap in an abstraction layer so you can swap providers without touching component code.

4. Forgetting to Clean Up Dead Flags

Old flag conditions accumulate as technical debt. Dead flag code hides bugs and confuses developers. Set a 2-release maximum lifespan.

5. No Exposure Tracking for A/B Tests

Without exposure tracking, you cannot analyze experiment results. Every A/B test flag must fire an exposure event when the user is assigned.

Compared With

Performance Considerations

SDK Bundle Size

LaunchDarkly client-side SDK is 35KB gzipped. Split.io is ~25KB gzipped. If bundle size is a concern, consider a custom backend with a tiny client (2KB).

Flag Evaluation Latency

LaunchDarkly evaluates flags client-side in ~1-5ms once initialized. Initialization requires a network call (100-500ms). Always set timeout to avoid blocking rendering.

Polling vs SSE

• SSE (LaunchDarkly): Instant updates, persistent connection. Good for real-time flags.
• Polling (custom/Flagsmith): Configurable interval (30s-300s). Lower server cost but staleness.
• Hybrid: Evaluate once on page load, then subscribe to SSE for updates.

Accessibility Considerations

• Flag-controlled UI changes should be announced to screen readers (use aria-live regions)
• Don't hide critical functionality behind a flag without fallback for users who miss the feature
• A/B test variants should maintain the same accessibility level as control

Security Considerations

• Never expose flag evaluation context (user attributes, targeting rules) to unauthorized users
• Flag admin dashboards require authentication and audit logging
• Flag values received from the provider must be validated before use (especially JSON flags)
• Don't use feature flags to hide security vulnerabilities

Rules

• All flags defined in one typed file — no raw string keys in components.
• Every flag evaluation must have a fallback default value.
• Wrap provider client in an abstraction — never import SDK directly in components.
• A/B tests must fire an exposure event when the user is assigned.
• Gradual rollout: internal users → staged rollout (10%, 25%, 50%, 100%) → full release.
• Clean up flags within 2 releases — dead flag code is technical debt.

References

• references/feature-flag-systems.md — Feature Flag Systems
• references/feature-flag-testing.md — Feature Flag Testing
• references/flag-lifecycle-management.md — Flag Lifecycle Management
• references/flag-providers.md — Feature Flag Providers
• references/flag-sdk-customization.md — Flag SDK Customization
• references/flag-strategies.md — Feature Flag Strategies

Handoff

No artifact produced. Next skill: testing — mock flag client for component tests. Carry forward: flag definition file pattern, provider abstraction, targeting strategy.

Implementation Patterns

Observer Pattern for Event Handling

` interface EventObserver { onEvent(event: T): Promise; }

class EventBus { private observers: Set<EventObserver> = new Set(); subscribe(observer: EventObserver): void { this.observers.add(observer); } unsubscribe(observer: EventObserver): void { this.observers.delete(observer); } async emit(event: T): Promise { const results = Array.from(this.observers).map(o => o.onEvent(event)); await Promise.allSettled(results); } } `

Configuration-Driven Approach

config: defaults: timeout: 30s retryCount: 3 overrides: production: timeout: 60s retryCount: 5 development: timeout: 300s retryCount: 1

Production Considerations

Deployment Checklist

• Configuration validated against schema before startup
• Health check endpoints registered and monitored
• Graceful shutdown with draining period (30s timeout)
• Resource limits configured (CPU, memory, file descriptors)
• Log level set appropriate for environment
• Metrics endpoint secured and exposed
• Rate limiting configured per-tier
• TLS certificates valid and auto-renewing
• Database migrations run as separate deployment step
• Feature flags ready for gradual rollout

Monitoring and Alerting

Anti-Patterns

Performance Optimization

Caching Strategy

Cache hierarchy: L1 (in-memory local) → L2 (distributed Redis/Memcached) → L3 (CDN/Edge). Cache invalidation: TTL-based (simple, stale), event-based (complex, fresh), write-through (consistent, higher write latency), write-behind (fast writes, eventual consistency).

Resource Pooling

• Database connections: Pool of reusable connections (HikariCP, pgBouncer)
• HTTP connections: Keep-alive + connection pooling for external calls
• Thread pool: Bounded thread pools for async task execution

Profiling Methodology

1. Establish baseline with production traffic profile
2. Profile CPU with sampling profiler (pprof, perf, async-profiler)
3. Profile memory with heap dumps and allocation tracking
4. Profile I/O with strace/perf trace for syscall analysis
5. Profile latency with distributed tracing (OpenTelemetry)
6. Identify bottleneck, formulate hypothesis, implement fix
7. Re-profile to verify improvement, repeat

Architecture Decision Trees

Flag Type Decision Tree

What is the purpose of the flag?
  ├── Release toggle → Remove after feature is stable (max 2 releases)
  ├── Experiment (A/B test) → Requires analytics exposure tracking, statistical significance
  └── Permission/Ops toggle → Long-lived, needs kill-switch capability
       Who controls the flag?
       ├── Developers → Code-level toggle, config file or env var
       ├── Product → LaunchDarkly/Flagsmith UI with targeting
       └── Users → Visible setting in preferences UI

Flag Evaluation Decision Tree

Is the flag evaluated server-side or client-side?
  ├── Server-side → Secure (user can't see), good for A/B tests
  └── Client-side → Visible in DevTools, good for UI experiments
       Does the flag affect page content or just behavior?
       ├── Content → Server-side eval to avoid FOUC
       └── Behavior → Client-side eval is fine
            Need real-time updates?
            ├── Yes → WebSocket/SSE connection for flag changes
            └── No  → Polling or fetch-on-load