OpenFeature

Shipeasy ships OpenFeature providers (CNCF OpenFeature parity) so apps standardized on the OpenFeature API can plug Shipeasy in as the backing provider. The provider is a pure adapter over the SDK's local evaluation — no change to how flags resolve.

They live in their own package, so @shipeasy/sdk itself stays peer-free:

npm install @shipeasy/openfeature

@shipeasy/sdk is a peer, not a dependency: the provider wraps the engine your configure() call already built, so it must resolve to the same installed copy.

The provider class is named ShipeasyProvider.

Server provider

Construct new ShipeasyProvider() with no argument — the global form resolves the engine that configure({ apiKey }) already built. Always configure() first:

import { OpenFeature } from "@openfeature/server-sdk";
import { configure } from "@shipeasy/sdk/server";
import { ShipeasyProvider } from "@shipeasy/openfeature/server";

// 1. Configure the SDK once at app boot (server key).
configure({ apiKey: process.env.SHIPEASY_SERVER_KEY! });

// 2. Register the provider — no-arg form resolves the configured engine.
await OpenFeature.setProviderAndWait(new ShipeasyProvider());

// 3. Read through the OpenFeature client.
const ofClient = OpenFeature.getClient();
const on = await ofClient.getBooleanValue("new_checkout", false, { targetingKey: "u1" });

The provider's initialize() runs the SDK's one-shot rules fetch (the SDK fires Ready when the blob resolves) and onClose() tears the configuration down.

Type / reason mapping

• getBooleanValue → getFlagDetail — the gate's FlagReason maps to an OpenFeature reason (RULE_MATCH → TARGETING_MATCH, DEFAULT → DEFAULT, FLAG_NOT_FOUND → ERROR/FLAG_NOT_FOUND, CLIENT_NOT_READY → PROVIDER_NOT_READY).
• getStringValue / getNumberValue / getObjectValue → getConfig, with a TYPE_MISMATCH error code when the stored config value doesn't match the requested type.

The EvaluationContext.targetingKey becomes the bucketing user_id.

Tracking

The provider's OpenFeature track() forwards to the SDK's track() (no-op without a targetingKey), so conversions for experiment analysis flow through the same pipeline.