LLM Agent Trajectory Synthesis

LLM: Large Language Model

This field studies how to create useful sequences of agent actions, tool calls, observations, and decisions when real interaction data is scarce. Common approaches generate tasks and executable environments, collect teacher or simulated-agent rollouts, and verify them with state checks, constraints, or outcome rubrics; the resulting data is used for supervised fine-tuning or reinforcement learning. Researchers also reuse failed trajectories through relabeling, recovery examples, and skill extraction, especially for web, GUI, terminal, API, and long-horizon tasks. A central challenge is balancing scale and diversity with reliable, learner-relevant supervision that transfers to live or unseen environments rather than teaching brittle benchmark-specific behavior.

Latest papers

69 papers17 papers in the last four weeks, up 70% on the four weeks before. 0.1% of all new papers.

FlashScript: Compiling Recurring Agent Workflows into Reusable Neuro-Symbolic Programs
Oct 24, 2026Sadra Sabouri, Hosein Hasanbeig, Adam Sage +6LLM Agent Workflow Optimization LLM-Based Program Synthesis
University of Southern California, Los Angeles, CA · Microsoft, Redmond, WA
ReGE: Stronger Agents through Experience Synthesis
Oct 24, 2026Jiarong Liang, Yubo Wang, Zijie Liu +3LLM Agent Self-Improvement LLM Agent Trajectory Synthesis
University of Waterloo · University of North Carolina at Chapel Hill · University of British Columbia
Trajectory2Gym: Decision-Grounded Compositional Interaction Synthesis for Tool-Use Agents
Oct 24, 2026Yihao Hu, Zhihao Wen, Yitong Lin +5Tool-Using Agents LLM Agent Trajectory Synthesis
Westlake University (westlake.edu) · Ant International (ant-intl.com)
DynaWeave: Learning to Adapt through Dynamic Scenario Synthesis
Oct 21, 2026Xuan Yang, Chunxiao Liu, Hongsheng Xin +2LLM Agent Training Synthetic Data Generation for LLM Agents
Department of Data Science, City University of Hong Kong. · Hong Kong Institute of AI for Science, City University of Hong Kong. · Independent Researcher. +1
GRADUS: Discovering and Recovering Agent Lessons from the Environment's Latent Syllabus
Oct 20, 2026Qing Ping, Wei Yang, Panpan Xu +4LLM Agent Training LLM Agent Trajectory Synthesis
AWS AI · University of Southern California · BMW Group Research and Technology
DataPraxis: Scaling Analytics Tasks via Verified Workspace Evolution
Oct 17, 2026Nan Jia, Wenxuan Shi, Jingqing Ruan +11Data Analysis Agents Synthetic Data Generation for LLM Agents
Huazhong University of Science and Technology · Institute of Computing Technology, Chinese Academy of Sciences · Meituan +3
SWE-Stream: Scaling Executable Training Data for SWE Agents via Progressive Construction and Elastic Streaming
Oct 14, 2026Chengyang Liu, Wen Yang, Shifeng Chen +5Software Engineering Agents LLM Agent Training
Xiamen University · Unisound · Institute of Automation, Chinese Academy of Sciences
Interlock: From Local Correctness to Task Completion in Long-Horizon Agents
Oct 14, 2026Linji Hao, Yanji He, Zihan Wang +6Long-Horizon Agent Tasks LLM Agent Trajectory Synthesis
Peking University · The Hong Kong University of Science and Technology (Guangzhou) · Tsinghua University +1
Encounter-Driven City-Scale Human Mobility Generation with Large Language Models
Oct 13, 2026Chao Li, Renhe Jiang, Chuan Xiao +1Human Behavior Simulation LLM Agent Trajectory Synthesis
The Hong Kong University of Science and Technology · The University of Tokyo · The University of Osaka
WorkBorne: Discovering Agentic Tasks from Real-World Artifact Transitions
Oct 10, 2026Junyi Zhou, Yufei Wang, Qiyuan Zhang +7LLM Agent Trajectory Synthesis Long-Horizon Agent Tasks
City University of Hong Kong · WeChat AI, Tencent · MBZUAI
ClawLoop: Self-Improving Agents via Harness-Native Training Loop
Oct 9, 2026Hansi Yang, Linji Hao, Hongyan Hao +4LLM Agent Training Self-Improving Agents
Meituan, China · Peking University
Scaling Agentic Data for Long-Horizon Terminal Intelligence
Oct 9, 2026Shun Zou, Ziyu Ma, Yi Zou +6Synthetic Data Generation for LLM Agents Long-Horizon LLM Agents
University of Science and Technology of China · AMAP, Alibaba Group · Southern University of Science and Technology
Know What You Don't Know: Agentic Data Synthesis by Evolving Generation Space for Closed-Loop Learning
Oct 8, 2026Wei Yang, Qing Ping, Youzhi Luo +4Synthetic Data Generation LLM Agent Trajectory Synthesis
University of Southern California · AWS AI
ToolRACER: A Robust Agentic Conversation Emulation Resource for Agent Training and Evaluation
Oct 6, 2026Arkajyoti Chakraborty, Aryan Tayal, Ishika Agarwal +5Tool-Using Agents AI Agent Benchmarks
Uniphore · University of Illinois Urbana-Champaign
GAMEGO: Training Game-Dev Agents with Synthetic Trajectories Anchored in Real-World Assets
Oct 2, 2026Haoyue Yang, Jingyao Li, Zhengfan Wu +3Agentic Code Generation Synthetic Data Generation for Language Models
Institute of Automation, Chinese Academy of Sciences · Baidu Inc., Beijing, China
Scaling Trajectories for Complex Tasks through Recursive Self-Rewrite
Oct 2, 2026Zongxia Li, Yucheng Shi, Zhongzhi Li +7LLM Agent Trajectory Synthesis Self-Training for Language Models
Tencent HY LLM Frontier · University of Maryland, College Park · University of Georgia +4
SkillGym: Training Skill-Use Agents with Automatic Verifiable Environment Generation
Sep 29, 2026Renxi Wang, Mingshan Hee, Fajri Koto +2Supervised Fine-Tuning LLM Agent Skill Learning
Mohamed bin Zayed University of Artificial Intelligence
Skill2Env: Capability-Oriented Environment Synthesis from Skills for General Agents
Sep 27, 2026Weiyi Xu, Xiaowen Yang, Wen Da +7Tool-Using Agents LLM Agent Trajectory Synthesis
AllSpark Team
ExpVoyager: Direct Experience Navigation for Dynamic Agent Skill Synthesis
Sep 26, 2026Kwangwook Seo, Dongha LeeLLM Agent Skill Learning Continual Learning for LLM Agents
Yonsei University
A Wrong Turn Does Not Ruin the Journey: Deviation-Guided Skill Self-Evolution for LLM Agents
Sep 24, 2026Yichun Feng, Jiawei Wang, Haozhe SunLLM Agent Self-Improvement LLM Agent Skill Learning
University of Chinese Academy of Sciences · University of Science and Technology of China · Meituan
MimicAgent: Quadruped Skills via Prompt-to-Trajectory Generation
Sep 21, 2026Lucky Kant Nayak, Narayanan Palghat Parameswaran, Neehar Peri +1Reinforcement Learning Trajectory Generation
Carnegie Mellon University
SAP: State-Guided Data Synthesis with Argument Provenance for Multi-Turn Tool Use
Sep 5, 2026Zichen Tian, Jinpeng Chen, Cheng Gong +2Synthetic Data Generation for Language Models LLM Agent Trajectory Synthesis
Independent Researcher · Huawei Research
Constructing and Evaluating Clinical Reasoning Trajectories for Medical Agent
Sep 4, 2026Yunqi Zhu, Wen Tang, Jinghao Niu +2Reasoning Evaluation Clinical Reasoning
Guangzhou University, Guangzhou, China · Institute of Automation, Chinese Academy of Sciences, Beijing, China
EnvCraft: Synthesizing Executable Environments in Agentic RL for Claw-like Agent
Sep 4, 2026Yirong Zeng, Shen You, Jinhang Feng +8LLM Agent Training Agentic RL
Harbin Institute of Technology, SCIR Lab · Huawei Technologies Co., Ltd · Peking University +1
Terminal-Universe: Turning Agent Trajectories into Scalable Terminal Environments
Sep 3, 2026Jie Wu, Zhenru Zhang, Beichen Zhang +11Software Engineering Agents LLM Agent Trajectory Synthesis
1Qwen Team, Alibaba Group · 2Tsinghua University
From Final Artifacts to Trajectories: Retrospective Process Supervision for Evidence-Grounded Long-Form Generation
Aug 31, 2026Junjie Huang, Jiarui Qin, Di Yin +4LLM Agent Trajectory Synthesis Process Supervision
Shanghai Jiao Tong University · Tencent Youtu Lab
SimCRAFT: Distilling Remote Sensing Agents via Synthetic Trajectories and Contextual Retrieval-Augmented Fine-Tuning
Aug 31, 2026Haoran Wang, Jing Yao, Xu Yang +4Fine-Tuning Remote Sensing
State Key Laboratory of Remote Sensing and Digital Earth, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100101, China · University of Chinese Academy of Sciences, Beijing 100049, China · The Hong Kong University of Science and Technology (Guangzhou), Guangzhou 511453, China +1
What Makes Good Agentic Data? An ACE Lens on Data Generation for LLM Agents
Aug 27, 2026Xingshan Zeng, Zishan Xu, Boju Zhang +11Synthetic Data Generation for LLM Agents Synthetic Data Generation
Huawei Technologies Co., Ltd · Shanghai Jiao Tong University · Northwestern University, Chicago, IL +2
BrowserForge: Scaling Web Episode via Parallel Browser Sandboxes
Aug 25, 2026Fei Tang, Huawen Shen, Zhiqiong Lu +7Web Agents LLM Agent Trajectory Synthesis
Zhejiang University · LLM Department, Tencent
PROOF-Gen: From Optimized Data to Better Distillation
Aug 24, 2026Anh Ta, Junjie Zhu, Shahin ShayandehLLM Agent Training LLM Agent Trajectory Synthesis
Apple

Oct 24, 2026J/K move · Enter open · S save

FlashScript: Compiling Recurring Agent Workflows into Reusable Neuro-Symbolic Programs

Sadra Sabouri, Hosein Hasanbeig, Adam Sage, Shreya Sinha+5

University of Southern California, Los Angeles, CA · Microsoft, Redmond, WA

LLM Agent Workflow Optimization LLM-Based Program Synthesis Program Synthesis LLM Agent Trajectory Synthesis

Open paper AskPDF arXiv

Many agent tasks recur with new data or small changes, yet agents regenerate their reasoning and actions on each run, producing different results. Distilling a prior run into a skill helps the agent avoid earlier mistakes, but it still re-runs the full process each time, so runs can vary and incur that cost. We compile a successful agent run into a reusable neuro-symbolic program (NSP) with a set of guidelines we call FlashScript. Actions and data transformations become code, instance-specific values become parameters, and decisions that need model judgment become LLM calls. We identify three layers where these calls belong and three techniques that replace most of them with code. The compilation does not remove nondeterminism; it localizes it, so the deterministic program and its neural calls can be reviewed and tested separately. Across 120 WorkBench analytics tasks with Claude Haiku 4.5, the NSPs were correct and made identical tool calls on every run, using about 70 times fewer input tokens and a third of the runtime of a skill-guided agent whose runs were inconsistent. In a composite version of the tasks, the skill-guided agent's accuracy fell to 23%, and its output inconsistency increased, while the NSPs remained correct and identical across all runs. The NSPs successfully generalized to all 108 unseen instances of the same templates. Our findings suggest that NSPs from agent trajectories can be considered cumulative software assets for end-users to test, maintain, and reuse in their recurring workflows.