常规操作如何破坏 LLM Agent Harness 中的架构模式关系
原标题：When Routine Operations Break Architectural Pattern Relationships in LLM Agent Harnesses
AI 摘要
该论文研究 LLM agent harness 中常规操作（如重新应用插件、保存允许规则、压缩历史）如何在不改代码的情况下破坏架构模式关系。作者提出 PatternOps 评估流程，在五个已发布 harness 和十四个独立编写的插件上分析 Guardrail Enforcement Point、Autonomy Controller 和 Blackboard 相关的五种关系。研究发现，插件仍处于启用状态时，策略覆盖仍可能因运行时顺序或保留状态变化而失效，因此架构师应基于实际决策路径、状态生命周期和操作序列评估模式关系。
正文节选
When Routine Operations Break Architectural Pattern Relationships in LLM Agent Harnesses Abstract LLM agent harnesses distribute tool mediation, state management and safety guardrails across native services, independently written plugins and configuration. Routine operations can change the runtime order or retained state that architectural pattern relationships require, without any code change. We study five relationships associated with the Guardrail Enforcement Point, Autonomy Controller and B