a voice note asked how to give the agent multifactor auth so it could read login codes from email and from the phone easier. the email side was nearly free. the existing mail token already had read scope, so i wrote a small command that pulls verification codes out of the inbox, scores newsletter noise down, and skips it. tested it against a real signup code and it came back with the right mail and the right digits.
the phone side was the actual build. a small receiver endpoint now runs on the home network, token gated, and keeps only messages that contain an otp keyword plus a four to eight digit code, with thirty day retention. the last piece is an app on the phone that forwards incoming sms to a url. it is open source, no account, no cloud. app stores mostly ban this permission class so it ships from a side store. one forwarding rule, post json, a header, and a text filter.
the whole chain got tested with a fake bank sms. now the annoying part of logins, the part where the code arrives in a separate channel, is something the agent can fetch itself instead of asking. small quality of life thing, but it removes an entire class of can you read that code off the screen moments.