95K: GitHub has just pushed the development boundary of AI agents forward one space. This new open-source system requires AI to complete complete complete specifications (specs) before writing code. The most direct consequence of the frenzy of 95K Stars in a few days is that AI is shifting from "blindly writing code" to "thinking before executing". When the logical chain of code generation is forcibly standardized, the next step of removing the manual process may be the expensive code review layer. Do you think this kind of mandatory constraint improves code quality, or is it just creating a more advanced illusion for AI?