Tag
DocSkill提出一种带技能编排策略的智能体框架，用于多语言多跳文档问答，通过MCTS离线探索轨迹并结合SFT/DPO训练策略，同时发布包含60K QA对、覆盖10种语言的MMHop-DocBench数据集。
OpenSquilla is an open-source project that enables self-organizing skill orchestration for agents via MetaSkill 3.0, combined with intelligent routing to reduce token costs. The author integrated it into WeSight, demonstrating how a single sentence can convert a WeChat public account article into a Xiaohongshu post, showcasing the potential for agents to self-assemble workflows.
AgentSkillOS is an open-source framework that enables developers to build AI agents by retrieving and orchestrating pipelines from over 200,000 available skills.
SCOPE is a specification-guided framework for text-to-image generation that tracks semantic commitments to better fulfill complex visual intents. It introduces the Gen-Arena benchmark and demonstrates strong performance on complex generation tasks.