I probably would not have bothered to allocate time to this pre ai, the juice wasnt really worth the squeeze. But I approached it with an initial amusing naivete about it being 'super simple'. As is almost always the case with software theres a reasonable amount of hidden complexity. But I have been using it as a sort of learning proving ground for how to work with agentic development. For example I got to a point where claude wouldnt implement properly and would argue with me about changes because it would read the current/old docs in the repo and get confused about reality. So right now I'm experimenting with 'canonical specs' that can only be changed modally with gates and a defined cascade from canon, to code, to docs in that order. Otherwise you end up in a weird thrash about the docs and the code disagree and which one will the agent decide to change for consistency?
Anyway, its been interesting and its v0.6 and at a point where Ive not hit a sharp edge dogfooding for a while and some beta testers would be valuable. Right now you have to manually wire it into your stack, once some others have kicked the tyres hard enough I will make some pr's to the popular tools to consider integrating it.