TRAJDEBUG: Tracing Error Lifecycle to Identify Critical Failures in Long-Horizon Agent Trajectories

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Yunjia Qi, Zehua Yin, Xintong Shi, Hao Peng, Songyuanyi Lu, Yixian Liu, Richeng Xuan, Yuhong Liu, Zhichao Hu, Xiaozhi Wang, Lei Hou, Bin Xu, Juanzi Li
• Venue: arXiv: cs.AI
• Published: 2026-08-06
• Collected: 2026-08-07
• URL: http://arxiv.org/abs/2608.06346v1
• PDF: https://arxiv.org/pdf/2608.06346v1
• DOI: UNKNOWN

Plain-Language Brief

In this study, the authors propose a framework named TrajDebug to identify critical errors in long-horizon agent trajectories, which are complex systems that can perform tasks but often suffer from cascading errors that are difficult to debug. The framework addresses two main challenges: identifying individual errors in long trajectories and distinguishing between local errors that contribute to the final failure. It uses multi-granularity history compression and evidence-based error identification, and it also traces each error's resolution status and terminal impact. The authors have created a benchmark of 486 failed trajectories from Tau2Bench and SWE-Bench Pro for testing, and their experiments show that TrajDebug outperforms existing baselines. They plan to release the codes and data for further research.

What the Document Says

• The authors propose a framework called TrajDebug for identifying critical errors in long-horizon agent trajectories.
• Long trajectories make it difficult to identify individual errors due to scattered evidence across distant instructions, observations, and prior context.
• Failed trajectories often contain multiple local errors with different downstream effects, only some of which remain responsible for the final failure.
• TrajDebug addresses these challenges using multi-granularity history compression and evidence-based error identification.
• The framework also traces each error's resolution status and terminal impact.
• A benchmark of 486 manually annotated failed trajectories from Tau2Bench and SWE-Bench Pro has been created for testing.
• Experiments show that TrajDebug outperforms existing baselines.
• The authors plan to release the codes and data for further research.

Key Concepts Mentioned

• long-horizon agent trajectories
• cascading errors
• critical error detection
• error-lifecycle tracing framework
• TrajDebug
• multi-granularity history compression
• evidence-based error identification
• benchmark
• Tau2Bench
• SWE-Bench Pro

Claims Made by the Document

• TrajDebug addresses the challenges of long-trajectory error discovery and critical attribution.
• TrajDebug outperforms existing baselines.

Evidence Used

• The authors propose a framework called TrajDebug
• Experiments across diverse agent benchmarks show that TrajDebug achieves the best overall performance over existing baselines

Methods / Approach

• multi-granularity history compression
• evidence-based error identification
• tracing each error's resolution status and terminal impact

Limitations Stated by the Authors

Not stated in source

Limitations Noted by the Agent

Agent interpretation — not from the source.

The study focuses on the proposed framework TrajDebug. Further research may be needed to explore other approaches for error detection and debugging in long-horizon agent trajectories.

Open Questions

• How does TrajDebug compare with other state-of-the-art methods in various scenarios?
• What are the potential improvements or modifications to TrajDebug?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The study provides a framework for error detection and debugging, but it doesn't discuss its comparison with other methods or future improvements. However, it mentions that experiments show better performance than existing baselines.

Raw Abstract / Excerpt

LLM-based agentic systems have shown remarkable capabilities in complex domains, while suffering from cascading errors and difficulty in debugging. Critical error detection aims to locate the earliest error step in a failed trajectory that is responsible for the final failure. However, progress faces two main challenges. First, long trajectories make it difficult to identify individual errors, since the evidence for judging a step may be scattered across distant instructions, observations, and prior context. Second, failed trajectories often contain multiple local errors with different downstream effects, only some of which remain responsible for the final failure. In this work, we propose TrajDebug, an error-lifecycle tracing framework that addresses long-trajectory error discovery with multi-granularity history compression and evidence-based error identification, and supports critical attribution by tracing each error's resolution status and terminal impact. We further construct TrajErrBench, a benchmark of 486 manually annotated failed trajectories from Tau2Bench and SWE-Bench Pro, covering realistic tool-use and coding scenarios. Experiments across diverse agent benchmarks show that TrajDebug achieves the best overall performance over existing baselines, and application studies further demonstrate that its diagnoses provide actionable feedback for improving downstream agent success. We will release the codes and data to facilitate further research.