RefinementGuard:面向隐藏 Agent 工具执行的语义验证框架
RefinementGuard: Counterexample-Directed Semantic Qualification for Hidden Agent-Tool Execution
AI 导读
arXiv 论文提出 RefinementGuard,一种针对 Agent 工具调用隐藏并行图的翻译验证与 fail-closed 准入机制。在 100 个非等价故障中,基于义务的父语义验证拒绝 49/100,同时接受全部 60/60 个有效变体,而精确相等验证会连有效变体一起拒绝。
正文 · 原文
Abstract:Tool runtimes can lower an agent-selected operation into a hidden parallel graph, reducing latency and context while risking semantic mistranslation. We introduce RefinementGuard: translation validation for hidden, effectful agent-tool lowerings, coupled to artifact-bound, fail-closed runtime admission. Qualification compares a candidate graph with an independent reference on isolated state using explicit parent-operation obligations. We evaluate two repository-level contracts, bug localization and patch validation, on controlled repositories. Among 100 activated, non-equivalent development/qualification faults adjudicated by one human, parent-semantic validation rejects 49/100 (operator-cluster bootstrap 95% CI [33.7%, 64.8%]). The objective is selective admission, not maximum mutation rejection: exact equality rejects 100/100 faults but also all 60 valid variants, whereas obligation-based validation accepts 60/60; its 51 residual misses identify concrete specification gaps. On a frozen disjoint input split, the held-out B2-B3 result is localization-only because 21 patch mutants fail structurally; B3 rejects 38/56 localization defects. In a controlled asynchronous-delay regime, the localization serial/guarded median-latency ratio is 4.50x, and hidden execution exposes 74.3% less parent-visible context than visible execution. Patch validation shows no demonstrated speedup; zero-delay runs expose overhead. End-to-end qualification thus detects lower-layer misses while preserving localization's parallelism.
| Subjects: | Artificial Intelligence (cs.AI); Software Engineering (cs.SE) |
| Cite as: | arXiv:2611.23611 [cs.AI] |
| (or arXiv:2611.23611v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.23611 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Adhyayan Veer Singh [view email]
[v1]
Tue, 17 Nov 2026 23:04:28 UTC (78 KB)
来源:arXiv cs.AI · arxiv.org