swe轨迹分析
CodeTracer: Towards Traceable Agent States
https://arxiv.org/pdf/2604.11641v2
Understanding Software Engineering Agents: A Study of Thought-Action-Result Trajectories
https://arxiv.org/pdf/2506.18824
Understanding Software Engineering Agents Through the Lens of Traceability: An Empirical Study
https://arxiv.org/pdf/2506.08311
Understanding Code Agent Behaviour: An Empirical Study of Success and Failure Trajectories
https://arxiv.org/pdf/2511.00197
Process-Centric Analysis of Agentic Software Systems(轨迹->图)
https://arxiv.org/pdf/2512.02393
Beyond Resolution Rates: Behavioral Drivers of Coding Agent Success and Failure
https://arxiv.org/pdf/2604.02547
swe轨迹优化/增强
Trajectory-Informed Memory Generation for Self-Improving Agent Systems
https://arxiv.org/pdf/2603.10600
TRACE: Capability-Targeted Agentic Training
https://arxiv.org/pdf/2604.05336
CLEANER: Self-Purified Trajectories Boost Agentic Reinforcement Learning
https://openreview.net/pdf?id=bWPhJpPlnc
Yet Even Less Is Even Better For Agentic, Reasoning, and Coding LLMs
https://arxiv.org/pdf/2604.00824
infra优化
Heddle: A Distributed Orchestration System for Agentic RL Rollout
https://arxiv.org/pdf/2603.28101
Trajectory Balance with Asynchrony: Decoupling Exploration and Learning for Fast, Scalable LLM Post-Training
https://arxiv.org/pdf/2503.18929
SortedRL: Accelerating RL Training for LLMs through Online Length-Aware Scheduling
https://arxiv.org/pdf/2603.23414
thinking
What Do Agents Learn from Trajectory-SFT: Semantics or Interfaces?
https://arxiv.org/pdf/2602.01611
swe轨迹用于reward、guidance和target training
Agent-RLVR: Training Software Engineering Agents via Guidance and Environment Rewards
https://arxiv.org/pdf/2506.11425
swe/agent相关的PRM
When Agents go Astray: Course-Correcting SWE Agents with PRMs
https://arxiv.org/pdf/2509.02360
SWE-Shepherd: Advancing PRMs for Reinforcing Code Agents
https://arxiv.org/pdf/2604.10493
AgentPRM: Process Reward Models for LLM Agents via Step-Wise Promise and Progress
https://arxiv.org/pdf/2511.08325