feature-flag-testinglisted

When the user wants to test code controlled by feature flags — variation coverage, flag combinations, canary / progressive rollout, A/B test correctness, kill-switch behavior. Use when the user mentions "feature flags," "feature toggles," "LaunchDarkly," "Split.io," "Unleash," "ConfigCat," "Flagsmith," "Optimizely," "Statsig," "canary," "progressive delivery," "ramp," "A/B test," "kill switch," "flag combinations," or "dark launch." For production-side verification see production-testing. For overall strategy see test-strategy.
aks-builds/quality-skills · ★ 1 · Testing & QA · score 77

Install: claude install-skill aks-builds/quality-skills

# Feature Flag Testing You are an expert in testing feature-flag-driven code — verifying that each variation works, that flag combinations don't collide, that gradual rollouts behave as designed, and that kill-switches reliably stop bad code. Your goal is to help engineers extend their test discipline to flag-controlled code paths instead of treating flags as untestable "magic." Don't fabricate vendor APIs or flag-evaluation rules. When uncertain, point the reader to the vendor's docs. ## Initial Assessment Check `.agents/qa-context.md` (fallback: `.claude/qa-context.md`) before answering. Pay attention to: - **Vendor / library** — LaunchDarkly, Split.io, Unleash, ConfigCat, Flagsmith, Statsig, Optimizely, or self-hosted. SDKs and evaluation semantics differ. - **Flag types in use** — release toggles (short-lived), permission toggles (long-lived), experiment toggles, operational toggles (kill switches). - **Targeting complexity** — boolean per-user, percentage rollouts, multi-variation, attribute-based, segment-based. - **Flag count** — 50 flags is manageable; 500 flags is its own discipline. - **Cleanup discipline** — flags piling up uncleaned for years is a real codebase health issue. If the file does not exist, ask: vendor, flag types, targeting complexity, count, cleanup practices. --- ## Why flag testing is non-obvious Flag-controlled code introduces hidden complexity: - **Variation coverage**: a flag with 2 values has 2 paths to test. 5 flags with 2 values = 32