LLM-based agents have moved automated program repair (APR) from fixed-context patch generation to interactive repository-level repair. However, existing agentic APR systems still struggle to use execution evidence to guide localization, patch generation, and validation. We propose EviACT (Evidence-to-Action), an agentic APR framework that coordinates three evidence-driven guardrails across repair stages. The retrieval scaffold grounds repair context, the compile gate filters invalid edits, and the test-driven gate checks target-test recovery before full regression. Across four benchmarks, EviACT improves resolve rate over the strongest reported comparable baselines by 1.6-6.0 percentage points and shows 70.1-88.6% lower reported per-bug API cost where baseline costs are available. Ablations and diagnostics suggest that these gains are associated with the coordinated evidence-to-action chain, making agentic APR more effective and efficient.

翻译：基于大语言模型的智能体已将自动程序修复（APR）从固定上下文的补丁生成推向交互式仓库级修复。然而，现有智能体 APR 系统仍难以利用执行证据来指导定位、补丁生成与验证。我们提出 EviACT（证据到行动），一种协调三类证据驱动护栏贯穿修复阶段的智能体 APR 框架：检索脚手架为修复上下文奠定基础，编译门控过滤无效编辑，测试驱动门控在完整回归测试前检查目标测试恢复。在四个基准测试上，EviACT 比已报告的最强可比基线将解决率提升 1.6-6.0 个百分点，并在存在基线成本的场景中，将每个缺陷的 API 成本降低 70.1-88.6%。消融实验与诊断表明，这些改进与协作性的证据到行动链条相关，使智能体 APR 更加高效与经济。