dsh-session-audit

After running a DeepSeek Harness Agent session, you might wonder: what exactly happened? How long did it take? Which steps went smoothly, and which tools failed or were called repeatedly? Were verification commands like npm test or cargo build executed? While DSH logs are comprehensive, the raw data can be hard to interpret at a glance. The dsh-session-audit plugin solves this by folding the persistent event log of a single session into a clear audit report, giving you a transparent view of the Agent's execution journey.

The plugin's core strength lies in its deterministic execution analysis. Instead of offering subjective scores, it presents objective facts: session duration, counts of turns and steps, success/failure distributions of tool calls, and rule-based audit signals—such as consecutive failures, high failure rates, repeated tool invocations, or frequent reads of the same file. A notable feature is verification command detection: it can catch the actual outcomes of test, build, lint, and similar commands from shell calls, letting you see whether code verification steps completed successfully. All analysis happens locally with no network requests, ensuring full privacy control.

dsh-session-audit is ideal for users who need to dive deep into Agent execution details. Whether you're a developer diagnosing performance bottlenecks or debugging failure scenarios, a team manager auditing tool efficiency and code verification practices, or simply curious about how a complex session works internally, this plugin provides actionable insights. It outputs reports in three formats (text, Markdown, JSON) and supports auditing both current and historical sessions, allowing you to flexibly obtain analysis based on your needs.