_version_ 1866913090293989376
author Meng, Fanqing
Du, Lingxiao
Wu, Zijian
Chen, Guanzheng
Liu, Xiangyan
Liao, Jiaqi
Jiang, Chonghe
Wan, Zhenglin
Gu, Jiawei
Zhou, Pengfei
Huang, Rui
Zhao, Ziqi
Ding, Shengyuan
Yu, Ailing
Peng, Bo
Xia, Bowei
Sun, Hao
Liang, Haotian
Xie, Ji
Chen, Jiajun
Song, Jiajun
Yang, Liu
Xu, Ming
Qiu, Qionglin
Fu, Runhao
Zhai, Shengfang
Wang, Shijian
Ma, Tengfei
Wu, Tianyi
Jin, Weiyang
Wang, Yan
Dai, Yang
Lai, Yao
Shu, Youwei
Liu, Yue
Hao, Yunzhuo
Niu, Yuwei
Huang, Jinkai
Zhuo, Jiayuan
Shen, Zhennan
Wu, Linyu
Yao, Hannah
Chen, Charles
Xie, Cihang
Zhou, Yuyin
Zhang, Jiaheng
Zheng, Zeyu
Hu, Mengkang
Shieh, Michael Qizhe
author_facet Meng, Fanqing
Du, Lingxiao
Wu, Zijian
Chen, Guanzheng
Liu, Xiangyan
Liao, Jiaqi
Jiang, Chonghe
Wan, Zhenglin
Gu, Jiawei
Zhou, Pengfei
Huang, Rui
Zhao, Ziqi
Ding, Shengyuan
Yu, Ailing
Peng, Bo
Xia, Bowei
Sun, Hao
Liang, Haotian
Xie, Ji
Chen, Jiajun
Song, Jiajun
Yang, Liu
Xu, Ming
Qiu, Qionglin
Fu, Runhao
Zhai, Shengfang
Wang, Shijian
Ma, Tengfei
Wu, Tianyi
Jin, Weiyang
Wang, Yan
Dai, Yang
Lai, Yao
Shu, Youwei
Liu, Yue
Hao, Yunzhuo
Niu, Yuwei
Huang, Jinkai
Zhuo, Jiayuan
Shen, Zhennan
Wu, Linyu
Yao, Hannah
Chen, Charles
Xie, Cihang
Zhou, Yuyin
Zhang, Jiaheng
Zheng, Zeyu
Hu, Mengkang
Shieh, Michael Qizhe
contents Language-model agents are increasingly used as persistent coworkers that assist users across multiple working days. During such workflows, the surrounding environment may change independently of the agent: new emails arrive, calendar entries shift, knowledge-base records are updated, and evidence appears across images, scanned PDFs, audio, video, and spreadsheets. Existing benchmarks do not adequately evaluate this setting because they typically run within a single static episode and remain largely text-centric. We introduce \bench{}, a benchmark for coworker agents built around multi-turn multi-day tasks, a stateful sandboxed service environment whose state evolves between turns, and rule-based verification. The current release contains 100 tasks across 13 professional scenarios, executed against five stateful sandboxed services (filesystem, email, calendar, knowledge base, spreadsheet) and scored by 1537 deterministic Python checkers over post-execution service state; no LLM-as-judge is invoked during scoring. We benchmark seven frontier agent systems. The strongest model reaches 75.8 weighted score, but the best strict Task Success is only 20.0\%, indicating that partial progress is common while complete end-to-end workflow completion remains rare. Turn-level analysis shows that performance drops after the first exogenous environment update, highlighting adaptation to changing state as a key open challenge. We release the benchmark, evaluation harness, and construction pipeline to support reproducible coworker-agent evaluation.
format Preprint
id arxiv_https___arxiv_org_abs_2604_23781
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ClawMark: A Living-World Benchmark for Multi-Turn, Multi-Day, Multimodal Coworker Agents
Meng, Fanqing
Du, Lingxiao
Wu, Zijian
Chen, Guanzheng
Liu, Xiangyan
Liao, Jiaqi
Jiang, Chonghe
Wan, Zhenglin
Gu, Jiawei
Zhou, Pengfei
Huang, Rui
Zhao, Ziqi
Ding, Shengyuan
Yu, Ailing
Peng, Bo
Xia, Bowei
Sun, Hao
Liang, Haotian
Xie, Ji
Chen, Jiajun
Song, Jiajun
Yang, Liu
Xu, Ming
Qiu, Qionglin
Fu, Runhao
Zhai, Shengfang
Wang, Shijian
Ma, Tengfei
Wu, Tianyi
Jin, Weiyang
Wang, Yan
Dai, Yang
Lai, Yao
Shu, Youwei
Liu, Yue
Hao, Yunzhuo
Niu, Yuwei
Huang, Jinkai
Zhuo, Jiayuan
Shen, Zhennan
Wu, Linyu
Yao, Hannah
Chen, Charles
Xie, Cihang
Zhou, Yuyin
Zhang, Jiaheng
Zheng, Zeyu
Hu, Mengkang
Shieh, Michael Qizhe
Computer Vision and Pattern Recognition
Software Engineering
Language-model agents are increasingly used as persistent coworkers that assist users across multiple working days. During such workflows, the surrounding environment may change independently of the agent: new emails arrive, calendar entries shift, knowledge-base records are updated, and evidence appears across images, scanned PDFs, audio, video, and spreadsheets. Existing benchmarks do not adequately evaluate this setting because they typically run within a single static episode and remain largely text-centric. We introduce \bench{}, a benchmark for coworker agents built around multi-turn multi-day tasks, a stateful sandboxed service environment whose state evolves between turns, and rule-based verification. The current release contains 100 tasks across 13 professional scenarios, executed against five stateful sandboxed services (filesystem, email, calendar, knowledge base, spreadsheet) and scored by 1537 deterministic Python checkers over post-execution service state; no LLM-as-judge is invoked during scoring. We benchmark seven frontier agent systems. The strongest model reaches 75.8 weighted score, but the best strict Task Success is only 20.0\%, indicating that partial progress is common while complete end-to-end workflow completion remains rare. Turn-level analysis shows that performance drops after the first exogenous environment update, highlighting adaptation to changing state as a key open challenge. We release the benchmark, evaluation harness, and construction pipeline to support reproducible coworker-agent evaluation.
title ClawMark: A Living-World Benchmark for Multi-Turn, Multi-Day, Multimodal Coworker Agents
topic Computer Vision and Pattern Recognition
Software Engineering
url https://arxiv.org/abs/2604.23781