三思而后行：LLM智能体的行动前验证

AI 摘要

该论文提出在LLM智能体执行动作前进行廉价确定性验证，以捕获静默失败。针对shell命令和代码编辑两种动作模态，构建了包含9,930条命令和640次编辑的基准，并发布验证器与防护机制。结果显示，静态验证器可捕获95.8%的无效命令，而基于锚点的应用器将静默错误降至0.01%。

正文节选

Look Before You Leap: Pre-Action Verification for LLM Agents Abstract An LLM agent acts on the world by emitting actions: shell commands to run, edits to apply. A wrong action does not always fail loudly; it can fail silently, producing a plausible but incorrect effect that raises no error. We argue that a cheap deterministic check, run before an action takes effect, is an effective and underused form of agent oversight, and we study it across two action modalities in one framework. The idea is

发布时间：2026-09-14 12:00

抓取时间：2026-09-14 12:28

来源机构：arXiv