AI 导读
在本周的 🦄 AI That Works 节目中，@tbrownio 分享了他们的“复利式”流程：每当 agent 拉取请求需要人工介入时，就生成复盘报告。 每周结束时，他们会回顾这些复盘报告，逐步改进技能、任务提示词和 agent 记忆文件，以避免未来再次出现同样的问题。
正文
on this week’s 🦄 AI That Works, @tbrownio shares their “compounding” process to generate post-mortems whenever human steering is needed on an agent pull request.
At the end of every week they review the post-mortems to incrementally improve the skills, task prompts, and agent memory files to avoid that problem in the future
来源：@dexhorthy · x.com