CodeRescue: Budget-Calibrated Recovery Routing for Coding Agents

Qijia He ,

Jiayi Cheng ,

Chenqian Le ,

Rui Wang ,

Xunmei Liu ,

Yixian Chen ,

Jie Mei ,

Zhihao Wang ,

Xupeng Chen ,

Yuhuan Chen ,

Tao Wang

Agent Autonomous Workflows AI Systems and Hardware CACRR

编码智能体日益在可执行环境中运行，此时一次失败的尝试所产生的并非单纯错误答案，而是具有实际指导意义的执行反馈。现有的成本感知型系统通常将此类失败视为级联决策问题：先调用成本较低的模型进行尝试，若失败则将难题升级交由更强大但更昂贵的模型处理。然而，在编程任务中，执行反馈本身还可用于引导低成本模型开展后续修复尝试，从而引出一个关键的预算约束型部署问题：智能体究竟应在何时投入更多低成本计算资源以尝试修复，又该在何时选择升级至更昂贵的模型？我们将这一失败后的决策建模为面向异构动作的“修复路由”（recovery routing）问题，并基于大量执行轨迹（execution rollouts）训练了一个监督式路由模型。为进一步提升该路由模型在不同预算约束下的泛化能力，我们引入了一层“共形风险控制”（Conformal Risk Control, CRC）机制——该机制可在部署阶段无需重新训练，仅通过动态选取成本惩罚系数，即可在数据满足交换性假设的前提下，对边际期望成本实现可控调节。在五个编程评测基准上留出的未见失败案例中，低成本修复与模型升级展现出互补的成功模式。经CRC校准的决策前沿（calibrated frontier）在性能上显著优于固定策略、仅依赖提示词的路由方法以及二元级联基线；在主实验设定（GPT-5.4-nano / GPT-5.4）下，某一个CRC校准前沿点的求解率已超越“始终升级”策略，同时其平均修复成本仅为后者的35%。相关代码开源地址为：https://github.com/Qijia-He/agent-budget-control。

Coding agents increasingly operate in executable environments where a failed attempt produces actionable feedback rather than merely an incorrect answer. Existing cost-aware systems typically treat such failures as cascade decisions: try a cheap model first, then escalate hard cases to a stronger and more expensive model. In coding, however, execution feedback can also make further cheap-model recovery worthwhile, raising a budgeted deployment question: when should an agent spend more cheap compute, and when should it escalate? We formulate this post-failure decision as recovery routing over heterogeneous actions and train a supervised router from execution rollouts. To make the same router usable under changing budgets, we add a Conformal Risk Control (CRC) layer that selects a deployment-time cost penalty without retraining and provides marginal expected-cost control under exchangeability. Across held-out failures from five coding benchmarks, cheap recovery and escalation exhibit complementary success patterns. The calibrated frontier improves over fixed actions, prompt-only routers, and a binary cascade baseline; in the main GPT-5.4-nano/GPT-5.4 setting, one CRC-calibrated frontier point exceeds always-escalate solve rate while using 35% of its mean recovery cost. Code is available at https://github.com/Qijia-He/agent-budget-control.

许愿