dsh-trace-insight

When building complex Agent workflows in DeepSeek Harness (DSH), developers often struggle with the challenge of hidden hundreds of steps within long conversations. Amidst massive trace data, it is difficult to intuitively grasp the Agent's decision-making logic or precisely pinpoint whether failures stem from model judgment errors, tool invocation issues, or environmental configuration problems.

Trace Insight is designed specifically to address this pain point. As a read-only review plugin for DSH, it transforms dense raw events into a clear analysis timeline. By using local rule analysis, it automatically identifies critical nodes such as tool failures, non-progressive loops, and path guessing. Furthermore, it utilizes a standalone model to deeply analyze the root causes and risks behind decisions without interfering with the main model's operation.

This plugin is ideal for developers and researchers who need a deep understanding of Agent behavior. Whether you are optimizing an Agent's execution strategy or seeking concrete original evidence for complex tasks, Trace Insight provides powerful assistance. It helps you clarify the Agent's "thought path," making every review targeted and effective.