dsh-mcmp: DeepSeek Harness Mathematical Modeling Competition Paper Pipeline Plugin

dsh-mcmp is a persistent Cordis plugin based on DeepSeek Harness, designed to provide a fully automated pipeline for writing mathematical modeling competition papers. The plugin decomposes the paper generation process into five stages—initialization, EDA, modeling, writing, and typesetting—comprising a total of twenty-two sub-stages, ensuring that each step yields verifiable artifacts. Its core mechanisms include dual questioning from the central console, P0-P3 rating handling, exponential backoff retry, and forced fallback mechanisms. Regardless of whether the pipeline completes normally, encounters errors, or is aborted, a Final_Paper.md file is ultimately generated, along with stage artifacts, rollback records (ROLLBACK_LESSONS.yaml), file indexes (FILE_INDEX.yaml), and runtime logs. The plugin supports resuming from breakpoints, real-time progress monitoring, and manual intervention, via pnpm.