Amidst the National Day festivities, DeepSeek, a prominent player in the domestic large language model (LLM) arena, rolled out an enhanced iteration of its agent framework, DeepSeek Harness. This latest version boasts improved plugin and cross-platform functionalities, along with the introduction of an experimental compatibility layer tailored for Claude Code Mods. In theory, once adapted, certain extension features crafted using Claude Code will be capable of operating within this framework. Framework lead Cui Tianyi elucidated that the primary objective behind developing the compatibility layer was to ascertain whether the extensibility of Claude Code Mods falls within the architectural purview of DSH's 'everything is a plugin' paradigm. At present, not all Claude Code Mods can operate flawlessly, but technical advancements are anticipated to make seamless integration a reality in the near future. Cui underscored that the 'everything is a plugin' principle has been integral to the framework's ethos from its inception, with open-sourcing remaining a steadfast commitment for DeepSeek. The team eschews unoriginal replication, aiming instead to propel AGI development from the framework level through innovation, fostering globally recognized, cutting-edge advancements. Public records indicate that DeepSeek's Agent Harness team was established in March of this year. The team's helm, Cui Tianyi, is a post-90s alumnus of Zhejiang University's Computer Science program. Prior to joining DeepSeek in March, he accrued nine years of experience at a leading quantitative institution. Harness represents a software support framework constructed around large language models, acting as the comprehensive operating system necessary for LLMs to function effectively as AI Agents. Previously, the DeepSeek Harness v0.1 developer preview version was open-sourced under the MIT license, amassing over 70,000 stars on GitHub within a mere 24 hours of its debut.