Agent Skills为何有效：将混乱经验蒸馏为可执行流程
Agent skills work for a very specific reason: they turn messy past experience in…
做Agent开发的同学必看，这篇用数据拆解了“技能”生效的本质是程序化锚定而非知识补充，直接指导我们如何设计更好的经验蒸馏流程。
Agent skills work for a very specific reason: they turn messy past experience into a clean procedure the agent can follow.
智能体技能的作用非常明确：它们将杂乱无章的过往经验转化为智能体可以遵循的清晰流程。
The researchers gave agents the same past trajectories in 2 forms: Workflow Memory, which keeps more execution detail, and a distilled SKILL.md.
研究人员以两种形式向智能体提供了相同的过往轨迹：保留更多执行细节的工作流记忆（Workflow Memory），以及经过提炼的 SKILL.md。
The skill version performed 6.06 percentage points better than Workflow Memory.
技能版本的性能比工作流记忆高出 6.06 个百分点。
Because the agent was not getting more experience. It was getting the same experience packaged better.
因为智能体并没有获得更多经验，而是获得了包装得更好的相同经验。
Their trajectory analysis makes the mechanism clearer: 65.7% of skill cases worked through procedural anchoring, while only 4.5% worked by supplying missing knowledge.
他们的轨迹分析使机制更加清晰：65.7% 的技能案例通过程序锚定起作用，而仅有 4.5% 是通过补充缺失知识起作用的。
So skills mainly help with execution: what to do first, which tools to use, what to verify, and which mistakes to avoid.
因此，技能主要在执行层面提供帮助：先做什么、使用哪些工具、验证什么以及避免哪些错误。
This also explains the failure mode. A skill can still hurt when it is used in the wrong situation or followed too rigidly.
这也解释了失败模式。当技能在错误的情境下使用或过于僵化地遵循时，它仍可能造成损害。
Overall takeaway, self-improving agents need better distillation and application of experience, not just bigger memory libraries.
总体启示是，自我改进的智能体需要更好地提炼和应用经验，而不仅仅是更大的记忆库。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力