你刚刷到这条消息，本来准备顺手划走，但又怕自己错过了真正会影响下一步判断的那一点。
最容易做错的，是addyosmani / agent-skills；代价往往是如果只盯表面热闹，你很容易在错误方向上花掉时间、预算和注意力。；我先给一个保守判断：没写进规则文件的经验，对 AI 等于不存在。。
My take: CLAUDE.md can be more valuable than a model upgrade. If a team rule is not written into the rules file, it does not exist for AI.
What stood out to me in addyosmani/agent-skills is that rules files like CLAUDE.md or AGENTS.md are treated as layer 1 context, meaning the most important persistent input. In plain English, this is the file that tells the agent how your team actually works. That is a very different bet from chasing every new model release.
The same guidance also warns that once you dump more than 5,000 lines of unrelated context into a session, the agent starts losing focus, and it recommends keeping a task under 2,000 relevant lines. The point is hard to miss: the fix is usually not more context. It is written-down judgment.
This is not an argument that model upgrades do not matter. It is a narrower one: a tool update is worth your attention only if it changes your next decision. For most teams, the next decision is not "which model do we switch to?
" It is "which rules do we keep re-explaining in chat instead of writing once?"
真正该讨论的是：addyosmani / agent-skills