TraceView: Interactive Visualization of Agentic Program Repair Trajectories

TraceView is an interactive visualization tool that helps developers understand and diagnose how LLM-based automated program repair agents work through their reasoning processes. By organizing agent trajectories into visual graphs with labeled components, the tool addresses a critical gap in debugging agent failures and improving repair outcomes.

TraceView addresses a fundamental challenge in AI-assisted software development: the opacity of agent decision-making. When LLM-based repair agents fail, developers have limited visibility into why, making it difficult to improve systems or prevent similar failures. This tool transforms raw agent logs into interpretable visual narratives, decomposing complex repair attempts into discrete Thought-Action-Result components that can be analyzed semantically.

The broader context reflects growing adoption of autonomous agents in software engineering workflows. As organizations increasingly deploy LLM-based tools for bug fixing and code generation, understanding agent behavior becomes critical for trust and safety. Current systems often produce black-box outputs where only final success or failure is visible, leaving developers unable to identify where reasoning becomes circular, actions become misaligned with objectives, or feedback loops fail.

TraceView's impact spans both research and industry. For researchers, it provides structured methodology for analyzing agent trajectories at scale, enabling systematic studies of failure modes. For enterprises deploying APR systems, it reduces debugging overhead and improves confidence in autonomous repair tools. The user study validation demonstrating improved comprehension suggests practical utility beyond academic interest.

The open-source release and documentation availability indicate community-focused development. Future development will likely focus on automated anomaly detection within trajectories, integration with popular APR frameworks, and scaling visualization techniques for increasingly complex agent behaviors. Organizations evaluating LLM-based code repair tools should monitor how interpretability tools like TraceView influence adoption decisions across software development teams.

• →TraceView makes LLM agent repair trajectories interpretable by organizing them into visual graphs with semantic components
• →The tool directly addresses the challenge of diagnosing why autonomous agents fail at code repair tasks
• →User studies confirm improved comprehension of agent behavior when using the visualization approach
• →Open-source availability enables broader adoption across research and enterprise software development
• →Interpretability tools are becoming essential infrastructure for deploying autonomous AI agents in production systems