Systematically improve code through structured analysis-mutation-evaluation loops. Adapted from ALMA (Automated meta-Learning of Memory designs for Agentic systems). Use when iterating on code quality, optimizing implementations, debugging persistent issues, or evolving a design through multiple improvement cycles. Replaces ad-hoc "try and fix" with disciplined reflection, variant tracking, and principled selection of what to change next.

通过结构化的分析-变异-评估循环系统性地优化代码。源自ALMA（面向智能系统的自动化元学习内存设计）。适用于迭代代码质量、优化实现、调试顽固问题或通过多次改进循环演进设计。取代了随意的“尝试修复”模式，转而采用有纪律的反思、变体追踪和基于原则的下一步变更选择。