Exploring the Relationship Between Code Metrics and the Ability of Large Language Models to Solve Code Issues

Large Language Models (LLMs) are increasingly integrated into software engineering workflows, assisting with tasks such as code generation, unit testing, and bug resolution. Despite their growing use, limited research has examined how code-level metrics such as Halstead complexity, maintainability index, and source lines of code correlate with issue resolution rates for LLM-powered agents when modifying existing codebases. The paper bridges the well established field of code metrics with the emerging domain of AI software engineering agents by exploring correlations between these metrics and the success of LLM driven agents in resolving GitHub issues. The key contribution is a quantitative analysis of the relationship between code metrics and LLM agent issue resolution success. Preliminary results show varying degrees of biserial correlation across models, with coefficients reaching $\mathbf{0. 4}$ for maintainability index and issue resolution. These findings suggest that code metrics can inform the evaluation of LLM coding agents and offer insights into improving models and finding indicators of issue resolution success. This research underscores the importance of maintainable, high-quality code not only for human developers but also for LLM-driven agents. Our hypothesis is that LLMs can generate correct patches or semi-automated fixes more successfully when the underlying codebase already exhibits strong maintainability and code quality metrics. Furthermore, as LLM-generated code becomes part of long-lived software products-reviewed, integrated, and extended by both humans and potentially future LLM agents-its maintainability and quality remain critical.