dsh-plugin-fusion: Fusion-Mode Generation Tool for DeepSeek Harness

To address the issue of unstable single-generation quality for models in DeepSeek Harness, the community has released the `dsh-plugin-fusion` plugin. By using a “multiple sampling + self-verification” mechanism—first generating multiple candidate answers and then having the model itself evaluate and select the best one—the plugin improves the reliability and consistency of results. Its core features include: calling the model to obtain five candidate answers at different temperatures and completing the evaluation; returning the selected answer, the candidate set, and the reasoning process; and supporting zero dependencies. After installation, the service must be restarted for the changes to take effect. During use, simply invoke the `fusion_generate` tool directly, although it is limited by the `count` parameter (1–10). This tool is suitable for scenarios that prioritize high quality and can accommodate the associated reasoning cost.