AgenC //: 👾 coding agents are getting dumber every session and almost nobody is fixing it automatically. AgenC is. not claude code. not cursor. none of them. they get better through humans noticing patterns and filing issues. three sprints later someone patches a prompt. most of it never gets captured at all. what if the agent just watched its own runtime traces? it would catch things no human ever would. tool X called after tool Y always tanks output. context over 80k and reasoning falls apart on step 3. grok chokes on a specific schema shape every single time. run that across enough sessions and you learn something nobody else sees. exactly where the model slips and exactly how to route around it. those patterns turn into PRs. prompt patches, tool ordering rules, context strategies. merge them, ship them to every instance. so now you have two loops running:...