Join us to explore [SkillOpt](https://arxiv.org/pdf/2605.23904), a framework that treats an agent’s skill as a trainable state rather than static prompt text. We’ll discuss how agent rollouts can drive bounded, validation-gated edits such as adding, removing, or replacing skill content, to improve reliability while keeping deployment-time inference unchanged. The session will cover the paper’s optimization strategy, safeguards against unproductive revisions, and the practical implications of transferable, self-evolving skills for agentic systems. Slides for past meetups posted: [Github](https://github.com/YanXuHappygela/LLM-reading-group/tree/main) Recordings posted at: [YanAITalk](https://www.youtube.com/@yanaitalk/videos) Feel free to reach out if you want to present at upcoming meetups! Note: You must have a Zoom account to login (free account is sufficient). Zoom Link will be posted to the event page one day before the meetup.