Thanks for sharing your rationale! I tend to agree -- the practical alternative for Linux anyways is some kind of key agent, and it's unclear to me (having written a few password managers at this point, including an age-based one[1]) that the attacker model for an agent is any different. The only advantage is the "public" channel one I mentioned, and that can of course be done out of band.
And since I forgot to say it earlier: excellent work!