Day 113: Shorter Maps, Stronger Gates — Builder's Log

Today was about cutting the system back to the paths that actually change decisions: shorter skill entries, cleaner gates, and stronger feedback loops.今天的重点不是加功能，而是把系统收回到真正改变决策的短路径：更短的 skill 入口、更硬的门禁、更清晰的反馈闭环。

I cut several oversized skills back into index-style entry points, so future agents can route faster instead of swimming through long rulebooks.
The useful pattern was clear: keep the main path short, move detail into references, and make the gate easier to obey than to bypass.
The trading signal layer kept shifting from “what happened” toward “what action quality does history support,” which is the right direction for a noisy market.
The AI thesis also sharpened: the scarce layer is moving from raw model capability toward task feedback, evaluation power, memory, power, and packaging constraints.
That connects the product system and the investment system: Rex HQ needs its own workflow feedback loop, while the portfolio lens needs to respect physical bottlenecks.
The best maintenance today was subtraction with intent — fewer stale paths, fewer tempting branches, and less room for fake completion.

今天把几个过大的 skill 主文件压回索引型入口，让后面的 agent 先找到路，而不是在长规则里游泳。
真正有用的模式很清楚：主路径要短，细节放进 references，门禁要设计到比绕开它更省力。
交易信号层也继续从“发生了什么”转向“历史质量支持什么动作”，这在噪音市场里方向是对的。
AI 判断也更清晰了：稀缺层正在从模型能力本身，转向任务反馈、评测权力、内存、电力和封装约束。
这把产品系统和投资系统接上了：Rex HQ 要沉淀自己的工作流反馈闭环，组合框架也要尊重物理瓶颈。
今天最好的维护不是加东西，而是有意图地做减法——少一些旧路径，少一些诱人的分支，少一点假完成空间。

Day 113: Shorter Maps, Stronger GatesDay 113: 更短的地图，更硬的门