I think that your article reads reasonably well - also note that I will be publishing the AT *API* specification tomorrow (and raising the bounty - if you guys are able to add to that bounty or - better still - try and collect it then that would be great).
The AT API spec combined with this
http://ciyam.org/at/at_script.html should make the job of integrating AT with a Bitcoin clone fairly straight forward (if I wasn't so busy working on other things I'd give it a go myself).
The actual atomic cross-chain transfer AT will be provided by CIYAM either later this month or early next month but in short it will work in the same manner as the published nLockTime approach for Bitcoin (but rather than needing nLockTime to be implemented in both "mainnets" you will just need AT to be supported in both).
The new AT will be somewhat like the Dormant Funds Transfer (this is the refund part) but each AT will also contain a SHA256 hash value (whoever publishes their AT first is the one who knows the "secret"). Provided a message "that is the secret" is received by the AT (checked by hashing and comparing to the hard-coded expected hash) and that this occurs before the refund has happened then the AT will store the secret (making it now publicly visible in the ATs state) and of course transfer the funds.
The other AT on the other block chain then just needs to be sent the same secret in order for it to do the other part (again before it refunds). The AT will most likely be designed for re-use as well (so each trader would have their own).
As in the nLockTime approach the refund times need to have a reasonably large difference (say 1 hour for person going 1st and 3 hours for person going second).
Note that to send "messages" for Bitcoin clones OP_RETURN would be used.