dsh-error-audit

When running automated tasks under DeepSeek Harness, errors arrive fast and scatter quickly across multiple event streams. Reconstructing the full context of a single incident—error code, triggering session, the AI action in flight, the user's original prompt—often means stitching together fragments from several log sources by hand, which is slow and prone to missing critical details.

dsh-error-audit turns that process into a single write. It listens to agent/error, agent/request-error, and tools/result events in real time and, the instant an error occurs, persists a complete record: timestamp, error code, session ID, workspace path, the user's verbatim input, and the AI action—none of it truncated. For error-level events it also proactively injects a notice into the affected session (live for running tasks, wake for idle ones), so the AI can immediately call the built-in read_error_logs tool to pull up the full report and close the loop on its own.

If you are building production-grade agentic workflows with DeepSeek Harness and want every anomaly to leave a traceable, filterable, restart-proof audit trail, this plugin is the lightest starting point: it works out of the box with zero configuration, and your first complete error report lands the moment the first error fires.