Future-proofing is a very wise idea :). As you noted, both xp_regread and xp_instance_regread are also undocumented. And http://social.msdn.microsoft.com/Forums/sqlserver/en-US/b83dd2c1-afde-4342-835f-c1debd73d9ba/xpregread explains your concern (plus, it offers you an alternative).
Your trace and your run of sp_helptext 'sp_get_sqlagent_properties' are a good start. The next thing to do is run sp_helptext 'sp_helptext', and note its reference to sys.syscomments. BOL sys.syscomments topic redirects you to sys.sql_modules, and that points to the next step. Unfortunately for your needs, just one row (for 'sp_get_sqlagent_properties') will be returned by running USE msdb; SELECT object_name(object_id) FROM sys.sql_modules WHERE definition LIKE '%sp_get_sqlagent_properties%'. I thus assume you are out of luck - there appears to be no alternative, publicly documented, module (sproc). My assumption could be wrong :).
I deduce that xp_reg% calls exist for client (SMO, SSMS, etc.) needs, such as setting/getting agent properties. More importantly (for your needs), your sp_helptext run also reveals SSMS (a client) is using a registry store (i.e. not a SQL store). Unfortunately, I must deduce (based upon an absence of proof from a library search) that those keys (and their values) are also not documented...
The above appears to put you in a pickle. You could decide "if we are going to rely upon undocumented registry keys, we might as well rely on the undocumented calls to read them", but I won't recommend that:). You could also file a feature request at https://connect.microsoft.com/ (your need is clear), but because your need concerns a client-side feature request, I do not recommend holding your breath while waiting for a fix :).
Perhaps it is time to step back and take a look at the bigger picture:
How often can that key be changed, and how often will this process poll for that change?
Email uses a mail primitive. Sender: "Dear recipient, did you get my mail?" Recipient: "Dear sender, did you send me mail?" Disabling an email profile is not the only reason for an email failure.
Would a different approach be more useful, when compared to periodically checking a key?
One approach would be to periodically send "KeepAlive" email. If the "KeepAlive" email isn't periodically received, maybe that key was tweaked, maybe the Post Office is on a holiday, or maybe something else (equally bad) happened. Plus, this approach should be fully supported, documented, and be future-proof. Who knows how and what keys will be used in the next version of SQL Server Agent?
The first bullet isn't addressed (neither would it be addressed by periodically checking a key), and perhaps you have additional needs (worth mentioning on MS connect).