What problem does it solve? Rolling out features safely requires deciding who sees what, consistently, on every request. This Skill documents how to define flags as JSON with variants, targeting rules, and percentage weights, then evaluate them deterministically so the same user lands in the same arm everywhere, while distinguishing a switched-off flag from a broken flag system. ## Core Features & Use Cases - Typed flag definitions: Define flags as JSON (FlagSet, FlagDefinition, TargetingRule, Condition, Split, segments) validated by Standard Schemas, so editors reject exactly what the engine would refuse. - Deterministic evaluation: parseFlagSet, evaluate, evaluateAll, and createEngine resolve flags as pure synchronous functions over a loaded snapshot, returning reasons like TARGETING_MATCH, TYPE_MISMATCH, or TARGETING_KEY_MISSING. - Pluggable stores: Keep definitions in Cloudflare KV (WorkerKVFlagStore), memory (InMemoryFlagStore), or any custom backend behind a single read method, verified with the shipped conformance suite. - Use Case: Roll out a new checkout flow to 10% of users on Cloudflare Workers: store the flag set in KV, call engine.load() inside waitUntil, and evaluate per request with stable subject hashing. ## Quick Start Ask the AI to create a flags engine with an in-memory store defining a boolean flag, load it, and evaluate it with a default value to see the resolution reason.
