一句话：提出 PROBE，将失败运行遥测转为结构化诊断与有界恢复指导，提升软件工程智能体的失败后恢复能力。
软件工程智能体失败后的恢复仍昂贵、人工且临时。现有系统只暴露轨迹或生成反馈，未将异构运行时证据转化为有依据、有边界的恢复指导。本文提出 PROBE，一个失败锚定的结构化恢复框架，含遥测层、诊断层与指导门：保留细粒度运行时信号，融合跨信号证据形成有依据诊断，并仅在证据充分、可操作且限于智能体行为范围时生成恢复指导。在仓库级软件修复、企业工作流恢复与 AIOps 服务缓解的 257 个未解决案例上，PROBE 达到 65.37% Top-1 诊断准确率与 21.79% 恢复率，分别超最强基线 43.58 和 12.45 个百分点；微软 IcM 原型表明其可作为非侵入式旁路接入现有诊断流程。
Software engineering agents are increasingly deployed in evaluable engineering environments, yet post-failure recovery remains costly, manual, and largely ad hoc. Existing systems expose traces for inspection or generate follow-up feedback, but they do not turn heterogeneous runtime evidence into grounded, bounded recovery guidance for a subsequent attempt. We present PROBE, a failure-anchored framework for structured recovery in software engineering agents. PROBE organizes failed-run telemetry into structured evidence, structured diagnosis, and bounded recovery guidance, realized by a Telemetry Layer, a Diagnosis Layer, and a Guidance Gate. The Telemetry Layer preserves fine-grained runtime signals, the Diagnosis Layer fuses cross-signal evidence into a grounded diagnosis, and the Guidance Gate constructs diagnosis-derived recovery guidance only when it is evidence-grounded, actionable, and within the scope of agent-side behavior. We evaluate PROBE across three software engineering settings spanning repository-level software repair, enterprise workflow recovery, and AIOps service mitigation. On 257 initially unresolved cases, PROBE achieves 65.37% Top-1 diagnosis accuracy and a 21.79% recovery rate, outperforming the strongest non-PROBE baseline by 43.58 and 12.45 percentage points, respectively. The results reveal a substantial diagnosis-recovery gap: accurate diagnosis is necessary but not sufficient unless it can be translated into bounded guidance that the subsequent attempt can execute and verify. Beyond controlled evaluation, a Microsoft IcM prototype shows that PROBE can attach as a non-intrusive side channel to existing service-diagnosis agent workflows without changing the agent policy, toolset, or execution budget. These results suggest that telemetry-grounded, failure-anchored recovery is a practical path to improving post-failure recoverability of software engineering agents under realistic engineering constraints.
@inproceedings{abs-2605-08717,
title = {Debugging the Debuggers: Failure-Anchored Structured Recovery for Software Engineering Agents},
author = {Chenyu Zhao and Shenglin Zhang and Yihang Lin and Wenwei Gu and Zhimin Chen and Yongqian Sun and Dan Pei and Chetan Bansal and Saravan Rajmohan and Minghua Ma},
booktitle = {ASE},
year = {2026},
doi = {10.1145/3832783.3834492}
}