StagedWorkspace: A Versioned Workspace for Knowledge-Work Agents
- ID: db38d993
- 原文链接: https://arxiv.org/abs/2608.18050
- PDF: https://arxiv.org/pdf/2608.18050v1
- 作者: Yining Hua, Hongbin Na, Yifan Zhou, Akshay Kalose, Cyrus Ayubcha, Levi Lian
- 日期: 2026-08-18
- 更新: 2026-08-18
- 分类: cs.AI
- 来源类型: arxiv
- 标签: agents, knowledge-work, versioning, workspace, arxiv
- 质量评分: 4/5
- 抓取时间: 2026-08-20T05:27:37Z
中文导读
知识工作型智能体会持续产出和修改代码库文档表格幻灯片等持久数字工件，但它们检索的解析视图编辑的原生文件评审的变更与最终提交的工件可能指向同一工作对象的不同版本论文将其形式化为工作区状态契约：每个视图都应显式绑定到演进中工作区状态的某个版本编码智能体已借助代码仓库契约部分满足这一需求，论文为更广泛的知识工作智能体提出了带版本管理的 StagedWorkspace 方案
为什么值得关注
把知识工作智能体的工作区形式化为状态契约，提出带版本管理的 StagedWorkspace，让每个视图显式绑定工作区版本
Grounded note: per the abstract, dual parsed/native access improved OfficeQA Pass@1 by 8.3-12.1 points and APEX mean rubric score by 4.7-9.2 points in fixed-harness ablations; SW-AGENT reached 63.9% on OfficeQA with Gemini 3.1 Pro vs a published same-model score of 29.3%. Directly relevant to document/office agent design.
关键信息
- 论文标题: StagedWorkspace: A Versioned Workspace for Knowledge-Work Agents
- 作者: Yining Hua, Hongbin Na, Yifan Zhou, Akshay Kalose, Cyrus Ayubcha, Levi Lian
- arXiv: https://arxiv.org/abs/2608.18050
- 发布时间: 2026-08-18
- arXiv 分类: cs.AI
- Comment: Under Review
- 关联标签: agents, knowledge-work, versioning, workspace, arxiv
English Abstract
AI agents increasingly perform knowledge work (i.e., produce and modify persistent digital artifacts such as code repositories, documents, spreadsheets, slides, reports), yet the parsed views they search, the native files they edit, the changes they review, and the artifacts they submit can refer to different versions of the same work product. We formulate this as a workspace-state contract: every view should be explicitly tied to a version of the evolving workspace state. Coding agents partly address this need through repository contracts for search, diffs, and tests, whereas an analogous contract is less explicit for PDFs, spreadsheets, slides, notebooks, and mixed-format project folders. We propose StagedWorkspace, a versioned workspace for knowledge-work agents. The workspace binds parsed records and review diffs to content hashes of the native files as they change. In fixed-harness ablations on OfficeQA Pro and APEX-Agents, dual parsed/native access has the highest point estimate for every tested model; relative to the more limiting single view, it improves OfficeQA Pass@1 by 8.3-12.1 points and APEX mean rubric score by 4.7-9.2 points. SW-AGENT scores 63.9% with Gemini 3.1 Pro on OfficeQA and 42.1 with GPT-5.4 Nano on APEX, compared with published same-model scores of 29.3% and 25.5, respectively. A paired review-axis ablation on 57 file-editing tasks further finds higher observed scores when diffs are visible. These results identify workspace state as an experimental variable in knowledge-work agents and motivate benchmarks that score evidence, staged edits, and submitted artifacts as explicit state transitions.
English Summary
AI agents increasingly perform knowledge work (i.e., produce and modify persistent digital artifacts such as code repositories, documents, spreadsheets, slides, reports), yet the parsed views they search, the native files they edit, the changes they review, and the artifacts they submit can refer to different versions of the same work product. We formulate this as a workspace-state contract: every view should be explicitly tied to a version of the evolving workspace state. Coding agents partly address this need through repository contracts for search, diffs, and tests, whereas an analogous contract is less explicit for PDFs, spreadsheets, slides, notebooks, and mixed-format project folders. We propose StagedWorkspace, a versioned workspace for knowledge-work agents. The workspace binds parsed records and review diffs to content hashes of the native files as they change....
Obsidian Notes
- 内容由
opencli arxiv paper拉取 arXiv 元数据与摘要生成。 - 中文导读与价值判断均锚定在条目已有摘要、论文摘要、作者、日期与分类信息上；未补充论文摘要之外的实验细节。