dsh-outcome-loop

In DeepSeek Harness (DSH), relying solely on "the model says it's done" is dangerous. dsh-outcome-loop is designed to bridge the gap between model outputs and actual execution results. It transforms every DSH session's goals, constraints, acceptance criteria, execution evidence, and final outcome into an auditable task record, ensuring tasks are genuinely completed rather than just perceived as done.

The plugin's core strength lies in its mechanical evidence-based acceptance mechanism. It uses hard metrics like tests, builds, and exit codes to determine task status. By default, it incurs zero extra model cost, avoiding any additional LLM calls or prompt injection. All data is stored locally in your DSH backend, ensuring privacy and user ownership. It also clearly distinguishes between success, failure, unknown, and user acceptance, preventing unknown states from being silently converted to success.

This plugin is ideal for developers and teams who require strict quality assurance, cost control, and full autonomy over their task history. It empowers users to truly own their data, with the ability to inspect, delete, or export records via a safe preview and redaction gate, providing a transparent, reliable, and low-cost solution for managing task outcomes in complex workflows.