DinoDS/dino-data-workflow-routing-preview

这个数据集名为Dino Data工作流路由预览，是一个专注于工作流路由的预览数据集，包含六个Dino Data能力切片：连接器意图检测、连接器动作映射、深度链接动作映射、文档导出规范、zip打包规范和深度链接意图检测。数据集的目标是训练或检查助手在工作流感知任务处理方面的行为，包括检测请求应路由到哪个动作或产品工作流，以及形成支持下游执行的结构化输出而不泄露内部控制逻辑。数据集包含120行数据，分为训练集（108行）、验证集（6行）和测试集（6行），每行都是一个扁平化的助手训练示例，包含路由和任务元数据。数据集可用于工作流意图检测实验、工具和连接器路由行为研究、结构化输出微调等。

This dataset is a focused workflow-routing preview built from six Dino Data capability slices: connector intent detection, connector action mapping, deeplink action mapping, document export specification, zip packaging specification, and deeplink intent detection. The goal is to train or inspect assistant behavior around workflow-aware task handling: detecting when a request should route into an action or product workflow, and shaping structured outputs that support downstream execution without leaking internal control logic. The dataset contains 120 rows, divided into training (108 rows), validation (6 rows), and test (6 rows) sets. Each row is a flattened assistant-training example with routing and task metadata. The dataset can be used for workflow-intent detection experiments, tool and connector routing behavior studies, structured output fine-tuning for downstream product actions, etc.