MemGovern: Enhancing Code Agents through Learning from Governed Human Experiences

2026-01-11

arXiv.org

AI Summary

This work addresses the limitation of existing autonomous software engineering agents, which operate under a “closed-world” assumption and struggle to effectively leverage open human experience from platforms like GitHub. To overcome this, the authors propose MemGovern, a framework that employs an experience governance mechanism to transform raw GitHub data into structured experience cards. Coupled with a logic-driven agent retrieval strategy, MemGovern enables, for the first time, the construction of fragmented open-source community knowledge into a plug-and-play memory infrastructure. Evaluated on SWE-bench Verified using 135,000 governed experience cards, the approach demonstrates a 4.65% absolute improvement in problem resolution rate, substantiating its effectiveness and practical utility in real-world software engineering tasks.

Technology Category

• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Multiagent Systems: Distributed Problem Solving
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

While autonomous software engineering (SWE) agents are reshaping programming paradigms, they currently suffer from a"closed-world"limitation: they attempt to fix bugs from scratch or solely using local context, ignoring the immense historical human experience available on platforms like GitHub. Accessing this open-world experience is hindered by the unstructured and fragmented nature of real-world issue-tracking data. In this paper, we introduce MemGovern, a framework designed to govern and transform raw GitHub data into actionable experiential memory for agents. MemGovern employs experience governance to convert human experience into agent-friendly experience cards and introduces an agentic experience search strategy that enables logic-driven retrieval of human expertise. By producing 135K governed experience cards, MemGovern achieves a significant performance boost, improving resolution rates on the SWE-bench Verified by 4.65%. As a plug-in approach, MemGovern provides a solution for agent-friendly memory infrastructure.

Problem

• autonomous software engineering
• closed-world limitation
• human experience
• GitHub data
• issue-tracking data

Innovation

• experience governance
• code agents
• experiential memory
• agentic retrieval
• software engineering automation