CodeTracer: Towards Traceable Agent States

TL;DR

CodeTracer is a novel tracing architecture that reconstructs agent state transitions and localizes failures in complex code agent workflows, improving debugging and failure analysis.

Contribution

It introduces a hierarchical trace reconstruction and failure localization method, along with a large benchmark dataset for evaluating code agent debugging tools.

Findings

CodeTracer outperforms direct prompting and lightweight baselines in failure detection.

Replaying diagnostic signals with CodeTracer recovers failed runs under matched budgets.

The approach enables systematic failure analysis in complex code agent workflows.

Abstract

Code agents are advancing rapidly, but debugging them is becoming increasingly difficult. As frameworks orchestrate parallel tool calls and multi-stage workflows over complex tasks, making the agent's state transitions and error propagation hard to observe. In these runs, an early misstep can trap the agent in unproductive loops or even cascade into fundamental errors, forming hidden error chains that make it hard to tell when the agent goes off track and why. Existing agent tracing analyses either focus on simple interaction or rely on small-scale manual inspection, which limits their scalability and usefulness for real coding workflows. We present CodeTracer, a tracing architecture that parses heterogeneous run artifacts through evolving extractors, reconstructs the full state transition history as a hierarchical trace tree with persistent memory, and performs failure onset localization to pinpoint the failure origin and its downstream chain. To enable systematic evaluation, we construct CodeTraceBench from a large collection of executed trajectories generated by four widely used code agent frameworks on diverse code tasks (e.g., bug fixing, refactoring, and terminal interaction), with supervision at both the stage and step levels for failure localization. Experiments show that CodeTracer substantially outperforms direct prompting and lightweight baselines, and that replaying its diagnostic signals consistently recovers originally failed runs under matched budgets. Our code and data are publicly available.