Newest papers

Teaching Code Agents to Ask Without Leaking Private Code

Wenjia Jiang, Jieke Shi, Zhensu Sun, Chengran Yang, Shidong Pan, Peixin Zhang, Yifan Jia, James Hoang, Xiwei Xu, Zhenchang Xing, David Lo, Zhou Yang · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.24824 · cited by 0 works (OpenAlex) · open-access PDF

French

Les agents de codage local peuvent étendre leur portée en consultant un modèle de nuage plus fort. De telles consultations, cependant, exposent souvent les identifiants privés, les chemins de fichiers et le code source du dépôt. Le masquage à l'hôte cache les identifiants reconnus mais laisse le code environnant et la phrase de l'agent inchangé, tandis que la formation pour moins de divulgation risque de supprimer la consultation ou de produire des questions trop vagues pour répondre. Nous proposons une abstraction vérifiée par l'utilité (UVA), qui enseigne à l'agent à décrire les entités privées par leur rôle et leur type tout en conservant les détails techniques d'un expert. Des trajectoires de réparation résolues, nous réécrivons

Machine translation (Meta M2M100); the English original follows.

Abstract

Local coding agents can extend their reach by consulting a stronger cloud model. Such consultations, however, often expose the repository's private identifiers, file paths, and source code. Host-side masking hides recognized identifiers but leaves the surrounding code and the agent's phrasing unchanged, while training for less disclosure risks suppressing consultation or producing questions too vague to answer. We propose utility-verified abstraction (UVA), which teaches the agent to describe private entities by their role and type while retaining the technical details an expert needs. From solved repair trajectories, we rewrite each question into this form and keep the rewrite only if it discloses fewer private entities and the repair, replayed from the same execution state with the repair trajectories. Across two benchmarks, the trained agent consults more often than before training. On repository-level issue repair, it solves 37.6% of issues, against 25.3% for the same model working alone, a gain from the training as a whole rather than from consultation alone. Its questions disclose 3.7-fold fewer private entities and 1.4-fold less verbatim code than the untrained agent's questions to the same expert.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Identifier, Coding (social sciences), Ask price, Computer security, Code (set theory)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange