aoisop/mawoagent is a local agent runtime that turns feedback into a skill and adopts it only when evidence and tests pass. There is no public latest release. The README describes a structure it calls a harness: the model produces only the task intent, while the runtime handles permission judgment, tool execution, state storage, and applying the result. The web workbench and command-line input use the same harness.
This structure does not treat a model's proposal as immediately executable technology; it focuses on adopting it as a skill through tests and evidence. The runtime includes tool execution and permission checks, session recovery, and long-term memory. It is Python-based, recommends Python 3.11 or 3.12, and pins dependencies with a lock file. The web pages are served from the same Python service with no separate package manager. The license is MIT, and the repository has 35 stars. The reading material is written in Chinese, so Korean readers need a translation.
That said, it is hard to judge the level of completeness from the published evidence alone. Most files in the repository are data files and generated technical documents, there are ten commits, and there are eighteen test-related files. The recorded quality score is 4.0 and the technical score is 5.0. These numbers and the file composition do not prove implementation stability or actual effect. So for now, it is more reasonable to see this as an early design that tries to control skill adoption based on evidence, rather than a verified self-improvement tool.