Wangchunshu
Abstract
TaskCraft automates the generation of scalable, multi-tool, and complex agentic tasks to enhance prompt optimization and fine-tuning of agentic models.
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 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.
Community
Automating Agentic Task Generation!
Existing instruction datasets lack essential information on tool usage and environment interaction, while resources like GAIA and BrowserComp depend heavily on manual annotation and remain limited in scale.
Introducing TaskCraft: an automated workflow for generating multi-tool, difficulty-scalable agentic tasks with verifiable execution trajectories. Starting with simple, easily verifiable atomic tasks, we progressively enhance complexity using depth- and width-based extensions—creating structured, hierarchically challenging problems. Our incremental validation strategies ensure efficiency and reliability, while also enabling the generation of tasks that exceed the generation agent’s capabilities.
With our generated dataset, agents achieve significant performance gains via Prompt Learning, while Agent Foundation Models benefit from SFT.
We’ve built a large-scale dataset of 36,000 agentic tasks, providing a robust foundation for systematic tuning and evaluation of AI agents.
Paper: https://arxiv.org/abs/2506.10055
Code & Data: https://github.com/OPPO-PersonalAI/TaskCraft
It's one of the first attempt to synthesize agentic tasks for agent learning and evaluation automatically.
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- OWL: Optimized Workforce Learning for General Multi-Agent Assistance in Real-World Task Automation (2025)
- WebDancer: Towards Autonomous Information Seeking Agency (2025)
- Agentic Reasoning and Tool Integration for LLMs via Reinforcement Learning (2025)
- MaskSearch: A Universal Pre-Training Framework to Enhance Agentic Search Capability (2025)
- ProRefine: Inference-time Prompt Refinement with Textual Feedback (2025)
- Advancing Tool-Augmented Large Language Models via Meta-Verification and Reflection Learning (2025)
- Enhancing LLM-Based Agents via Global Planning and Hierarchical Execution (2025)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment:
@librarian-bot
recommend
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper