My agent stack: sharpening the wrong side of the knife

I built my own supervisors, watchers, and assorted glue to keep agents working autonomously. It didn’t work particularly well, so I kept tinkering with the machinery.

Then I came across poteto’s work on pstack and its poteto-mode skill. It reminded me of goal/loop mechanisms already available in the harnesses—the same tools I’d been busy inventing replacements for.

With all the kitchen talk around poteto, this felt appropriate: I’d been sharpening my own knife from the wrong side.

For the last day or two, I’ve been adapting pstack and poteto-mode to my workflow and working out how to use those existing mechanisms properly. Autonomy is still a work in progress. But at least I’m no longer treating every missing behavior as a reason to build another watcher.

Next time, check the fucking drawer first.