Cameri on Nostr: Having to switch contexts pushed me in this direction. I have a list of 20 items in ...

Having to switch contexts pushed me in this direction. I have a list of 20 items in my todo and I could do only 1 at a time. Of course I can use subagents which most harnesses already support.

I could have multiple installations/config folders I maintain manually but that just adds overhead and a perpetual config drift.

A subagent isn’t a full fledged agent, and if you want to resume its session later to do one more thing you can’t: you start from scratch again.

Then there’s the need for isolation: git worktrees aren’t true isolation, and harnesses for the most part reuse the same skills plugins etc and if you change the config folder you now have to manually authenticate everything for each new agent.