arXiv:2608.05144 · 中英对照阅读

Argus：面向长程推理的通用智能体运行时

Argus: A General-Purpose Agentic Runtime for Long-Horizon Reasoning

Boxiu Li, Zimo Wen, Yijia Fan, Junxiang Lei, Sufeng Guo, Jiaao Wu, Ruize Tang, Mukai Li, Yifei Shen, Xiaoyu Chen, Wanbo Zhang, Runjing Gu, Yifei Gao, Yuheng Wu, Xuyao Huang, Zelong Zhao, Jiachen Zhang, Shibo Hu, Hangxi Guo, Yilin Chen, Yuzhe Zhang, Fan Yang, Chuan Wen, Xian Zhang, Xuanhe Zhou, Zhijie Deng

摘要

Argus 技术报告 · 2026 年 8 月

ARGUS TECHNICAL REPORT · AUG 2026

术语表

Argus: Argus
agentic runtime: 智能体运行时
long-horizon reasoning: 长程推理
verification-guided persistence: 验证引导的持久化
pivoting: 转向
runtime self-evolution: 运行时自进化
persistent state: 持久状态
operational objective: 操作目标
verification criteria: 验证标准
standing intent: 持续意图
goal drift: 目标漂移
mission: 任务单元
Manager: 管理者
Planner: 规划者
Engineer: 工程师
Reviewer: 审查者
self-review: 自审查
control plane: 控制平面
execution plane: 执行平面
records plane: 记录平面
working contract: 工作契约
evidence-backed refinement: 基于证据的细化
verification-gated: 验证门控
fixed-model: 固定模型
typed trajectory: 类型化轨迹
reviewed compression: 审查压缩
compounding: 复合积累
SWE-Bench Pro: SWE-Bench Pro
AARRI-Bench: AARRI-Bench
Skill/Wiki evolution: 技能/知识库演化

原文 arXiv:2608.05144