Thanks! I have it working pretty well in some of my repos (so far). I'm trying to make it easier to install and also ensure there is a reliable backup so updates to the npm install don't nuke your agents memory and coordinate documents. I was inspired by how openclaw uses memory and cross agent dat...

And using nostr for coordination with a simple script and throwing away the keys makes the model inference cheaper and more readable to avoid task collisions because the meta prompt logic can be much smaller and easier to reason about.