Sorry for resurrecting something from so far upthread, but the whole thing is a pretty good read and after finishing reading through it all I wanted to weigh in on this Armor thing in particular.
Armor gives "you" +2 dice to resist Matrix damage, so exactly how that would interact with an unattended agent is up to the GM. I'd say no since the agent's persona is actually distinct from any persona derived from the cyberdeck, but there's some grey there. If you were running Armor and an agent out of one deck, I wouldn't give the agent the +2 dice then either, so I don't see a need to do so when you're not there....
Starting with the situation of one decker running one deck and and one agent, I don't see why a matrix attack on the Agent wouldn't benefit from the Armor program. For one thing, many programs simply modify ASDF stats and the agent would certainly benefit from the deck running a better firewall or whatnot. The agent is also using the same same matrix connection as you so it doesn't make sense for something like signal scrub to only reduce noise for one persona either. Both of these programs also use the "you" pronoun.
But the best argument for the Armor program specifically applying to an attack on an Agent is this: The agent doesn't really have a condition monitor of his own. Sure, his persona will
defend against a Data Spike with (Agent Rating + Deckstat), but when it comes to damage resistance (where the armor program benefit applies), it really is "your" condition monitor that's under attack (the decker's. Or the deck itself if you prefer). You'll be rolling Device Rating + Firewall to resist damage, neither of which are unique to the Agent. It some ways it's like creating a second face for people to punch you in. Not that a second face matters terribly much in the non-euclidean geometry of the matrix where it's just as easy for 20 people to simultaneously punch you in the face as it is 2 people,... but you get my point.
We might make an argument that the offensive programs like Biofeedback and Hammer only apply to the decker (I'm undecided here), but I think that for damage resistance programs (Armor, Biofeedback filter, Shell, Defuse) the benefits of those programs apply at the point of the damage resistance test, which is unified for any number of persona running on the deck, not distinct and different pools like the defense tests would be for different personas.
The more tricky questions are about marks ON the agent and if/how they affect the decker persona. If an agent is marked, isn't the Deck that formed that persona also marked? (thereby allowing snoop and spoof and other trixy schemes) So if the deck is marked, isn't the decker persona now also marked? That would definitely make him non-hidden... a rude surprise if he things he's stealthing along quietly. It's tricky to be sure and there are important consequences to whatever conclusions you come up with.
In the end though, the conclusion Kincaid makes remains the same, Agents are useful tools but they introduce certain additional vulnerabilities and running a second deck for the sake of having an(other) agent is really, really inefficient and potentially counterproductive.