Repair Feature Flag Targeting Bugs

The problem

Feature flag inconsistencies often lead to broken rollouts, non-deterministic targeting, or race conditions during flag evaluation. Debugging why a specific user bypassed a rule or why a percentage rollout is skewed across different SDKs is time-consuming and error-prone.

What it does

• Normalizes evaluation inputs to ensure stable subject and bucket keys across components.
• Fixes broken rule precedence and percentage hashing logic for deterministic rollouts.
• Ensures parity between multi-SDK implementations and offline cache behaviors.
• Establishes canonical contracts for identity, trust boundaries, and state ownership.
• Implements deterministic local fixtures and regression tests to verify targeting logic.

Why this beats prompting it yourself

General LLMs often suggest broad retries or weakening validation to "fix" targeting bugs, which can cause production outages. This skill strictly preserves rollout percentages and kill-switch policies while enforcing validation before state mutation, preventing the introduction of new security or consistency risks.

Use cases

• Fixing drifting percentage rollouts where users get different results on different devices.
• Repairing race conditions in concurrent flag evaluations or state transitions.
• Aligning rule precedence logic between a backend service and a frontend SDK.
• Adding redacted observability to flag evaluation paths for easier production debugging.

Known limitations

Cannot contact live production services or read actual customer secrets. It operates strictly on repository files and local test fixtures.