PAPER / ARXIV:2609.08355
Hongzheng Chai, Jiakun Li, Hongyue Yu, Yuan Yuan
RESUMO
LLM-based agents solving repository-level code tasks rely on code search tools that typically return flat lists of isolated snippets, providing insufficient structure to distinguish a target function from similar alternatives in the same file. RepoNav reorganizes retrieved snippets into a file-centered navigation scaffold, guiding on-demand file-structure browsing so agents compare sibling symbols before selecting a target. Across models on LocBench, RepoNav improves function-level localization and narrows the file-to-function gap.
NO MESMO MAPA