A few days ago, the agent wallet product from vibe entered the polishing phase before its launch. This stage is both about refining details and polishing thought processes. Compared to the former, the latter requires more mental effort, as it gradually touches upon the core aspects of agent natural language programming. What does user-generated agent mean in a philosophical sense? This determines how we constrain its construction process and behavior. I believe that everyone involved in agent development will have questions about this issue. Regardless of how we define it, whether we call it context engineering or prompt engineering, all concepts point to the need for a methodology, as this constraint space is crucial, and it is not even a mathematical problem. I have a vague feeling that in this era of rapid action and code delivery, the biggest moat for online software is its understanding of the natural language constraint space. Because creating constraints is not knowledge, but the wisdom of a profound understanding of natural language. We can continuously explore the path that leads to this wisdom through experimentation. From this perspective, I use cc's plan mode to help me write various skills to analyze the effects of constraints, but this is merely a way of finding a path in the dark, like a blind person touching an elephant, and not a road leading to wisdom. However, I am happy to see such profound changes occurring in programming decades after the internet emerged. If you are also writing agents but are unclear about what I have recorded here, come back and take a look after some time; I believe you will resonate with it.