English

CODESTRUCT: Code Agents over Structured Action Spaces

Artificial Intelligence 2026-04-17 v3 Software Engineering

Abstract

LLM-based code agents treat repositories as unstructured text, applying edits through brittle string matching that frequently fails due to formatting drift or ambiguous patterns. We propose reframing the codebase as a structured action space where agents operate on named AST entities rather than text spans. Our framework, CODESTRUCT, provides readCode for retrieving complete syntactic units and editCode for applying syntax-validated transformations to semantic program elements. Evaluated on SWE-Bench Verified across six LLMs, CODESTRUCT improves Pass@1 accuracy by 1.2-5.0% while reducing token consumption by 12-38% for most models. Models that frequently fail to produce valid patches under text-based interfaces benefit most: GPT-5-nano improves by 20.8% as empty-patch failures drop from 46.6% to 7.2%. On CodeAssistBench, we observe consistent accuracy gains (+0.8-4.4%) with cost reductions up to 33%. Our results show that structure-aware interfaces offer a more reliable foundation for code agents.

Keywords

Cite

@article{arxiv.2604.05407,
  title  = {CODESTRUCT: Code Agents over Structured Action Spaces},
  author = {Myeongsoo Kim and Joe Hsu and Dingmin Wang and Shweta Garg and Varun Kumar and Murali Krishna Ramanathan},
  journal= {arXiv preprint arXiv:2604.05407},
  year   = {2026}
}

Comments

Accepted at ACL 2026 main conference

R2 v1 2026-07-01T11:56:36.956Z