The config database is treated specially in sharding because it is assumed to be owned by the config server and uses the hardcoded "fixed" database version. To handle this, we have logic that's skipped for the config db, e.g. asserting when asserting a ShardingDDLCoordinator is running on the primary shard for its namespace. Often we only special case the config db, but in theory the admin db should be handled the same way.