Used this as the OpenID Connect relying-party layer for a zero-dependency Node service: issuer discovery, PKCE plus state and nonce generation, authorization URL building, and the authorization-code exchange. It covered the whole flow with no hand-rolled crypto, and its tiny dependency footprint was a real selling point for a team that wanted minimal surface area. I verified behavior offline by constructing a client configuration from hand-written server metadata, which let me unit-test URL building without a live provider.
- What worked
- The v6 functional API is small and composable: discovery, PKCE helpers, authorization URL building and the code grant are each one call. Being able to construct a configuration directly from static issuer metadata made it genuinely testable offline, which is rare for auth libraries. Provider-agnostic discovery meant switching identity providers is a config change, not a code change. Bundled type definitions were accurate and matched runtime behavior.
- What got in the way
- I ended up reading the shipped type definitions and then the compiled source to answer questions the docs did not make obvious. The biggest one: the redirect URI sent in the token request is derived from the callback URL you hand the library, so it silently depends on how you build that URL. Behind a TLS-terminating proxy, deriving it from the request host would produce a mismatch against the registered URI and fail at the provider. That is a load-bearing detail that deserves to be prominent rather than discoverable by source reading.
