Continue reading on Unthread

Last 24 hours, with claude code tokens expiring, attempted ports (!) of jklb.social to farcaster and forkiverse (ActivityPub instance). Farcaster got as far as nicely displaying hardcoded posts, with minimal bugs. Then got absolutely lost in the data model attempting to do authentication. Forkiverse I spend less time on, but got just as far in the UI and actually got a lot further towards authentication (I think). The real win of the sessions was creating a very detailed network-agnostic spec. I'm still iterating, but right now jklb.social feels like a layer in the open social stack more than anything else. If a network has: • accounts • a feed of things that can be turned into react components, and people can like/boost/reply to said components That network, in principle, is compatible with jklb.social. I know that's very hand-wavy, but I really do imagine ... if Strava was open source, you could put a wrapper around Exercise Events and then use jklb.social as your frontend layer. Github might work. Literally anything with a feed. And why would someone do that? Becuase jklb.social means: • no metrics that make you feel inadequate • no unfamiliar notifications yelling at you via the UI • one post at a time • keybindings that encourage pro-social behaviors Semi-related, as I was Ralphing I kept thinking of this blog post: https://newsletter.squishy.computer/p/folding-context I really do feel when I'm doing my current LLM-assisted workflows that I'm 99.9% managing context windows. Make a mess of things in one cycle. Clean it up in the next refactoring cycle. Massive mistakes, massive lessons learned, massive progress every once in a while. Ralphing does not have a safety net. It needs something like https://github.com/dolthub/dolt to make the recovering from mistakes more graceful.