Salesforce AI Research Introduces OpenSkill for LLM Agent Self-Evolution

Salesforce AI Research posted a reply sharing OpenSkill: Open-World Self-Evolution for LLM Agents. The announcement presents a framework that lets LLM agents develop skills after deployment in open settings. It lists authors Zhiling Yan, Hanrong Zhang, Wei Liang, Yuxuan Zhang, Yutong Dai, Lifang He, Philip S. Yu, Xiang Li and others. The linked arXiv entry states that prior self-evolving methods depend on curated skills, successful trajectories or verifier signals, while the new approach targets real-world adaptation without those assumptions.