Praxa：面向受治理 AI 智能体执行的证据约束 harness
原标题：From Proposal to Verified Effect: Praxa, an Evidence-Bound Harness for Governed AI Agent Execution
AI 摘要
该论文提出 Praxa，一个面向 LLM 智能体执行的证据约束型 harness，通过确定性准入、代理执行、外部回读、对账与审核晋升，显式区分「提议、授权、派发、外部验证、对账、晋升」六种状态。作者报告四条证据线：仓库本地审计通过 1,027/1,027 单元测试与 89/89 Workerd 测试；Terminal-Bench 试点中基线与可靠性层各通过 17/36 严格试验，可靠性层多耗 37.49% 输入与 50.73% 输出 token，故不支持其更优；协调代理对比中候选方案 token 少 37.11%、成本低 33.84%、步骤少 11.63%，但未证明质量或生产行为提升。论文强调当前证据不足以确立对抗安全、生产安全或用户收益。
正文节选
From Proposal to Verified Effect: Praxa, an Evidence-Bound Harness for Governed AI Agent Execution An Engineering Validation, Descriptive-Pilot, and Benchmark-Governance Study Preprint — intended for arXiv submission (cs.AI / cs.SE)) Abstract Large-language-model agents can propose and execute actions, but proposal, authority, dispatch, verified external effect, and serving promotion are different claims. We present Praxa, an agent harness that represents these states explicitly through determin