一句话:提出知识驱动多智能体框架 K-COBAgent,结合规则翻译与 LLM 辅助,实现 COBOL 到 Python 的现代化。
COBOL 遗留系统向 Python 现代化面临语义差异大、代码规模大与人工成本高等问题。K-COBAgent 是一个知识驱动的多智能体框架,协调翻译、oracle 生成、验证、自修复和知识库管理等专用组件,将规则翻译与大语言模型辅助相结合,并通过精确哈希缓存匹配复用已有翻译记录以提升一致性。该仓库随论文提供源代码、配置文件、执行说明、评估脚本和归档报告。评估在 544 个 COBOL 模块上开展三轮现代化迁移,覆盖翻译、验证与修复流程。论文发表于 ASE 2026 新想法与新兴结果 track。
K-COBAgent is a knowledge-driven multi-agent framework for supporting COBOL-to-Python modernization. The framework coordinates specialized components for translation, oracle generation, validation, self-repair, and knowledge-base management. It combines rule-based translation with large language model assistance and reuses previously generated translation records through exact-hash cache matching. This repository provides the source code, configuration files, execution instructions, evaluation scripts, and archived reports associated with the paper K-COBAgent: A Knowledge-Driven Multi-agent Framework for COBOL Modernization, accepted at the 41st IEEE/ACM International Conference on Automated Software Engineering, ASE 2026, New Ideas and Emerging Results track. The evaluation covers three modernization passes over 544 COBOL modules.