Context
PR #21 removed the use of the client_logins table which tracked per-client first logins to determine whether to send a "Welcome" or "Sign-in" OTP email. The welcome/sign-in distinction currently falls back to checking whether the user has a PDS account at all (isNewUser = !did), which means a returning user logging into a new client app gets "Sign-in" instead of "Welcome".
The per-client consent data now lives in @atproto/oauth-provider's authorized_client table on pds-core. The auth-service could query this via a new internal API endpoint to restore the per-client welcome email behavior.
Proposed approach
- Add
/_internal/has-authorized-client?email=...&client_id=... endpoint on pds-core
- Auth-service calls it during OTP send to determine
isNewUser per-client
- Remove the unused
hasClientLogin/recordClientLogin methods and client_logins table from EpdsDb
Why not just keep client_logins?
The table is no longer written to by any code path. Consent tracking moved upstream. Keeping use of a dead table and re-adding writers would duplicate state that the upstream provider already manages.
🤖 Generated with Claude Code
Context
PR #21 removed the use of the
client_loginstable which tracked per-client first logins to determine whether to send a "Welcome" or "Sign-in" OTP email. The welcome/sign-in distinction currently falls back to checking whether the user has a PDS account at all (isNewUser = !did), which means a returning user logging into a new client app gets "Sign-in" instead of "Welcome".The per-client consent data now lives in
@atproto/oauth-provider'sauthorized_clienttable on pds-core. The auth-service could query this via a new internal API endpoint to restore the per-client welcome email behavior.Proposed approach
/_internal/has-authorized-client?email=...&client_id=...endpoint on pds-coreisNewUserper-clienthasClientLogin/recordClientLoginmethods andclient_loginstable fromEpdsDbWhy not just keep client_logins?
The table is no longer written to by any code path. Consent tracking moved upstream. Keeping use of a dead table and re-adding writers would duplicate state that the upstream provider already manages.
🤖 Generated with Claude Code