谷歌WikiSkill：持久化技能编辑历史提升Agent表现
Beautiful paper from Google.
Beautiful paper from Google.
来自谷歌的精美论文。
Skill-evolving agents rewrite a skill file each round, but the reasoning behind each edit gets scattered.
技能进化型代理每轮都会重写技能文件，但每次编辑背后的推理过程却散落各处。
Google found that keeping it in a separate, persistent wiki improves results across 5 benchmarks.
谷歌发现，将其保存在一个独立的、持久的维基中，能在5个基准测试上提升结果。
WikiSkill splits the agent workspace into 3 layers: immutable execution traces, an active skill file, and a wiki between them.
WikiSkill将代理工作区分为三层：不可变的执行轨迹、活跃的技能文件，以及介于两者之间的维基。
The wiki records failure patterns, and every past proposal with its accept-or-reject outcome.
维基记录了失败模式，以及每个过往提案及其接受或拒绝的结果。
A rejected skill edit disappears; the wiki's record of it does not.
被拒绝的技能编辑会消失；但维基中关于它的记录不会。
On Gemini-3.5-Flash, WikiSkill averaged 68.1% across the 5 benchmarks, against 56.1% for the strongest competing method and 49.5% with no skills.
在Gemini-3.5-Flash上，WikiSkill在5个基准测试中的平均得分为68.1%，而最强竞争方法的得分为56.1%，无技能时的得分为49.5%。
One boundary the paper names: skills are handed to the agent rather than retrieved, so skill selection is never part of the test.
论文指出的一个边界：技能是直接交给代理的，而非检索而来，因此技能选择从未成为测试的一部分。
If you are evolving agent skills, keep the rejection history somewhere the optimizer can read it.
如果你在进化代理技能，请将拒绝历史保存在优化器可读取的地方。
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