feature-flag-management（功能标志管理）

Feature Flag Management Skill

Overview

Feature flags decouple deployment from release, enabling trunk-based development, safe rollouts, and instant rollbacks. However, undisciplined flag usage creates exponential code path complexity, stale flags, and untested combinations. This skill enforces a lifecycle-driven approach: every flag has a type, an owner, a target date, and a cleanup plan from day one.

When to Use

• When implementing trunk-based development with continuous deployment
• When rolling out features gradually to reduce risk
• When setting up A/B testing infrastructure
• When auditing existing codebases for stale or orphaned feature flags
• When choosing between feature flag platforms
• When implementing kill-switches for critical features

Iron Laws

1. ALWAYS assign an owner and expiration date to every feature flag -- orphaned flags without owners accumulate indefinitely and become permanent tech debt.
2. NEVER nest feature flags more than 2 levels deep -- combinatorial explosion makes testing impossible (2 flags = 4 states, 5 flags = 32 states, 10 flags = 1024 states).
3. ALWAYS default flag values to the existing/safe behavior -- if the flag system fails, the application should behave as it did before the flag was added.
4. NEVER use feature flags as a substitute for configuration management -- flags are temporary toggles for release control, not permanent application settings.
5. ALWAYS clean up flags within 30 days of full rollout -- stale flags in code increase cognitive load, slow onboarding, and hide dead code paths.

Anti-Patterns

Workflow

Step 1: Flag Classification

Classify every flag before creation:

Step 2: Implementation Pattern

// OpenFeature SDK pattern (vendor-neutral)
import { OpenFeature } from '@openfeature/server-sdk';

const client = OpenFeature.getClient();

// Typed flag evaluation with safe default
const showNewUI = await client.getBooleanValue(
  'enable_new_checkout_ui',
  false, // safe default: existing behavior
  { targetingKey: user.id, attributes: { plan: user.plan } }
);

if (showNewUI) {
  renderNewCheckout();
} else {
  renderLegacyCheckout();
}

Step 3: Gradual Rollout Strategy

Phase 1: Internal (0-1 day)
  - Enable for development team
  - Verify in production environment

Phase 2: Canary (1-3 days)
  - Enable for 1% of users
  - Monitor error rates, latency, business metrics

Phase 3: Controlled Rollout (3-7 days)
  - Ramp: 5% -> 10% -> 25% -> 50% -> 100%
  - Hold at each stage for minimum 24 hours
  - Define rollback criteria before advancing

Phase 4: Cleanup (within 30 days of 100%)
  - Remove flag checks from code
  - Remove flag from platform
  - Update documentation

Step 4: Flag-Aware Testing

// Test both flag states in CI
describe('Checkout Flow', () => {
  describe('with new_checkout_ui enabled', () => {
    beforeEach(() => {
      flagProvider.setOverride('enable_new_checkout_ui', true);
    });

    it('should render new checkout components', () => {
      // test new path
    });
  });

  describe('with new_checkout_ui disabled', () => {
    beforeEach(() => {
      flagProvider.setOverride('enable_new_checkout_ui', false);
    });

    it('should render legacy checkout components', () => {
      // test legacy path
    });
  });
});

Step 5: Stale Flag Detection

# Find flags older than 30 days that are fully rolled out
# Custom script pattern for codebase scanning
grep -rn 'isEnabled\|getBooleanValue\|getFlag' src/ | \
  awk -F"'" '{print $2}' | \
  sort -u > active_flags.txt

# Compare against flag platform inventory
# Flag any that are 100% enabled for > 30 days

Step 6: Cleanup Checklist

For each flag being retired:

• Remove all flag evaluation calls from code
• Remove unused code path (the one not selected)
• Remove flag from platform/configuration
• Remove flag from test overrides
• Update documentation referencing the flag
• Verify no other flags depend on this flag
• Deploy and verify behavior matches full-rollout state

Complementary Skills

Memory Protocol (MANDATORY)

Before starting:

Read .claude/context/memory/learnings.md for prior feature flag patterns and platform-specific decisions.

After completing:

• New pattern -> .claude/context/memory/learnings.md
• Issue found -> .claude/context/memory/issues.md
• Decision made -> .claude/context/memory/decisions.md

ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.