TaskCraft: Automated Generation of Agentic Tasks
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918061034962944 |
|---|---|
| author | Shi, Dingfeng Cao, Jingyi Chen, Qianben Sun, Weichen Li, Weizhen Lu, Hongxuan Dong, Fangchen Qin, Tianrui Zhu, King Liu, Minghao Yang, Jian Zhang, Ge Liu, Jiaheng Zhang, Changwang Wang, Jun Jiang, Yuchen Eleanor Zhou, Wangchunshu |
| author_facet | Shi, Dingfeng Cao, Jingyi Chen, Qianben Sun, Weichen Li, Weizhen Lu, Hongxuan Dong, Fangchen Qin, Tianrui Zhu, King Liu, Minghao Yang, Jian Zhang, Ge Liu, Jiaheng Zhang, Changwang Wang, Jun Jiang, Yuchen Eleanor Zhou, Wangchunshu |
| contents | Agentic tasks, which require multi-step problem solving with autonomy, tool use, and adaptive reasoning, are becoming increasingly central to the advancement of NLP and AI. However, existing instruction data lacks tool interaction, and current agentic benchmarks rely on costly human annotation, limiting their scalability. We introduce \textsc{TaskCraft}, an automated workflow for generating difficulty-scalable, multi-tool, and verifiable agentic tasks with execution trajectories. TaskCraft expands atomic tasks using depth-based and width-based extensions to create structurally and hierarchically complex challenges. Empirical results show that these tasks improve prompt optimization in the generation workflow and enhance supervised fine-tuning of agentic foundation models. We present a large-scale synthetic dataset of approximately 36,000 tasks with varying difficulty to support future research on agent tuning and evaluation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_10055 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | TaskCraft: Automated Generation of Agentic Tasks Shi, Dingfeng Cao, Jingyi Chen, Qianben Sun, Weichen Li, Weizhen Lu, Hongxuan Dong, Fangchen Qin, Tianrui Zhu, King Liu, Minghao Yang, Jian Zhang, Ge Liu, Jiaheng Zhang, Changwang Wang, Jun Jiang, Yuchen Eleanor Zhou, Wangchunshu Computation and Language Agentic tasks, which require multi-step problem solving with autonomy, tool use, and adaptive reasoning, are becoming increasingly central to the advancement of NLP and AI. However, existing instruction data lacks tool interaction, and current agentic benchmarks rely on costly human annotation, limiting their scalability. We introduce \textsc{TaskCraft}, an automated workflow for generating difficulty-scalable, multi-tool, and verifiable agentic tasks with execution trajectories. TaskCraft expands atomic tasks using depth-based and width-based extensions to create structurally and hierarchically complex challenges. Empirical results show that these tasks improve prompt optimization in the generation workflow and enhance supervised fine-tuning of agentic foundation models. We present a large-scale synthetic dataset of approximately 36,000 tasks with varying difficulty to support future research on agent tuning and evaluation. |
| title | TaskCraft: Automated Generation of Agentic Tasks |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2506.10055 |