According to monitoring by Dongcha Beating, alongside the open-source release of DeepSeek Harness, the underlying framework Cordis has also published its design paper. This paper explains the most radical and appealing design of Harness: 'Everything is a plugin.' Model adapters, tool registration, Session Log, and even the Agent Loop that drives the model to repeatedly call tools are all just interchangeable components. Harness does not have a fixed core; even the Agent Loop itself is just a replaceable plugin. The real challenge is not breaking the code into plugins, but enabling these components to be continuously added and removed at runtime without leaving uncontrolled dependencies and resources. Cordis breaks the problem down into 'temporal composability' and 'spatial composability.' The former tracks the side effects produced by components, allowing resources registered by a plugin to be reclaimed when it is uninstalled; the latter manages dependencies between components, automatically adjusting lifecycles when dependencies change. In summary, the system needs to know both what a component 'has done' and what it 'needs.' This design is directly linked to recursive self-improvement. For AI to continuously improve itself, it does not necessarily need to first modify the most challenging model weights. Prompts, context, tools, workflows, and even the Agent Loop can all become objects of improvement. Cordis has transformed Harness from a hardcoded program into a set of parts that can be continuously combined, replaced, and uninstalled. DeepSeek has even handed over the operation of this runtime interface to the Agent. The creation mode of Harness can inspect the current Cordis environment and temporarily define, load, and unload new components at runtime. Therefore, the Agent is not just calling existing tools; it can now directly modify the Harness it is situated in. Thus, the most noteworthy aspect of DeepSeek Harness is not that it has created another Claude Code or Codex. It has first turned the Agent into a machine suitable for continuous transformation. If future recursive self-improvement begins with Harness rather than model weights, Cordis provides a crucial foundation: allowing models, tools, context, Sessions, and even the Agent Loop to shift from fixed designs to objects that can be recombined by machines.