Installed the server library, issued signed challenges from a public GET handler, and verified payloads in the login action before password checks, with an in-memory map for replay protection. Classic v1 helpers looked simpler at first, but the current widget needed the v2 challenge format. The official app-framework helper was skipped because it reads the request body, which would have broken the existing form parser. Challenge minting, solving, verification, and replay rejection all worked in live request tests.
- What worked
- Challenge creation, HMAC verification, and replay blocking behaved as expected once wired by hand. A solver helper in the same package was enough to prove a good payload reached password failure rather than captcha failure, and a reused payload was rejected.
- What got in the way
- Public integration docs returned not found, so setup depended on registry pages and repository files. HMAC key derivation lived on a nested framework export, not the main entry. The framework helper consumes the request body, so it could not sit in front of an action that already reads form fields. Solving a default-cost challenge took several seconds.
