Self-Evolving Coding Rules for AI Coding Agents
Zhengyuan Jiang, Neil Zhenqiang Gong and colleagues at Duke University introduce RuleEvolve (NeurIPS 2026), which evolves the coding-rules files that coding agents read instead of relying on hand-written ones.
Ask this paper
Method. A pool of candidate rule sets is mutated by an LLM, scored by a judge module, and updated with the best variants each iteration.
Evaluation breadth. Two coding-agent frameworks, four backbone LLMs and three benchmarks.
Results. Evolved rules beat manual rule engineering and prompt-optimization baselines on functional correctness, and in several settings also reduce code length and token cost.
Abstract
The performance of AI coding agents is highly dependent on their underlying coding rules. However, existing coding rules are typically hand-crafted and fixed, making the process labor-intensive and often suboptimal. In this work, we propose RuleEvolve, a self-evolving framework for coding rules. RuleEvolve maintains a pool of candidate coding rules and iteratively improves them. In each iteration, it employs an LLM-powered mutator module to generate variants from existing candidates, and then uses a judge module to evaluate these variants and update the pool with the best-performing ones. Extensive evaluations across two coding-agent frameworks, four backbone LLMs, and three benchmarks demonstrate that RuleEvolve outperforms both manual engineering and existing prompt optimization baselines in terms of functional correctness of the generated code, code length, and/or generation cost (e.g., tokens used).