Skill: feature-flags

What I do

I provide expertise in decoupling deployment from release. I enable runtime control of feature availability, gradual rollouts (1% → 100%), A/B testing, and operational kill-switches without new code deployments.

When to use me

• Releasing new features gradually to mitigate risk
• A/B testing different implementations/UI variants
• Trunk-based development with incomplete features
• Emergency feature disablement (kill switches)
• User segment targeting (e.g., beta testers only)

Core principles

1. Decouple Deploy from Release — Deploy code continuously; release features through configuration.
2. Short-Lived by Default — Release flags are temporary; remove them once 100% rolled out to avoid debt.
3. Fail Safe — If flag evaluation fails, always default to the "safe" (usually legacy/disabled) state.
4. Gradual Rollout — Progressively increase exposure (1% → 5% → 25% → 100%) and monitor metrics.
5. Fast Toggle — Changes must take effect in seconds without application restart.

Patterns & examples

Release Toggle Pattern (Go):

if features.IsEnabled("new-checkout-flow") {
    return newCheckoutFlow(ctx, order)
}
return legacyCheckoutFlow(ctx, order)

Percentage-Based Rollout (Go):

func (f *FlagStore) IsEnabledForUser(flagName, userID string) bool {
    // ... hash userID and check against rollout percentage
    hash := hashString(userID)
    bucket := hash % 100
    return bucket < uint32(f.flags[flagName].RolloutPercentage)
}

Experiment Variants:

variant := features.GetVariant("button-color")
switch variant {
case "red":   return renderRedButton()
case "green": return renderGreenButton()
default:      return renderBlueButton()
}

Anti-patterns to avoid

• ❌ Flag Sprawl — Accumulating hundreds of old flags; implement a "cleanup" task after 100% rollout.
• ❌ Testing only one path — Always test both the flag-enabled AND the fallback path.
• ❌ 200 for everything — Ensure your flag system failures don't return 200 OK with broken UI.
• ❌ Ignoring Metrics — Increasing rollout percentage without checking error rates/latency.
• ❌ Hardcoding Defaults — Use a central configuration source rather than scattered hardcoded checks.

KB Reference

~/vaults/baphled/3. Resources/Knowledge Base/AI Development System/Skills/Delivery/Feature Flags.md

Related skills

• devops — Pipelines that deploy flagged code
• monitoring — Observability during gradual rollouts
• configuration-management — Managing secrets and environment-specific flags
• breaking-changes — Using flags to manage risky API or schema transitions