两篇新研究论文提出了改进 LLM 编码代理的控制和可靠性的方法。一篇论文介绍了一个确定性控制平面 Rel(AI)Build，旨在将代理配置作为供应链进行管理，强制执行分层权限，并检测提示漂移。另一篇论文提出了一个用于编排的贝叶斯控制器，将工具使用决策视为成本敏感的顺序假设检验，以管理不确定性并提高正确性评分，尤其是在验证成本高昂的情况下。 AI
影响 这些方法旨在提高 LLM 编码代理的可靠性和安全性，可能带来更值得信赖和高效的 AI 辅助软件开发。
排序理由 该集群包含两篇提交到 arXiv 的学术论文，详细介绍了关于 LLM 编码代理的新研究。
AI 生成摘要 · Google Gemini · 来自 5 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
arXiv:2606.26924v1 Announce Type: cross Abstract: LLM coding harnesses grant agents broad file and shell access, yet the configuration layer that steers them -- rules files, agent definitions, IDE-specific markdown -- is largely unmanaged. A prevalence study of 10,008 public GitH…
LLM coding agents make it tempting to automate empirical research by delegating experiments to them directly, but naive delegation does not scale to large projects: low-rate instruction lapses compound into broken, irreproducible artefacts. To address this problem, we present Gli…
LLM coding harnesses grant agents broad file and shell access, yet the configuration layer that steers them -- rules files, agent definitions, IDE-specific markdown -- is largely unmanaged. A prevalence study of 10,008 public GitHub repositories (n=6,145 agent config files) finds…
arXiv:2606.24453v1 Announce Type: new Abstract: Modern coding agents pair LLM generators with various tools, including cheap diagnostics and expensive verifiers. The tool-use decisions are typically governed by orchestrators that often use fixed rules and ignore uncertainty. We f…
Modern coding agents pair LLM generators with various tools, including cheap diagnostics and expensive verifiers. The tool-use decisions are typically governed by orchestrators that often use fixed rules and ignore uncertainty. We formulate orchestration as cost-sensitive sequent…