dsh-arena: A Local Experiment and Evaluation Workbench Plugin for DeepSeek Harness

dsh-arena is a local experimentation and evaluation workbench plugin for DeepSeek Harness (DSH), designed to address the issues of stability comparison among multiple candidate profiles under the same task, as well as the trade-offs among quality, duration, and cost. The tool is licensed under MIT, version 0.2.0, and supports local installation rather than npm publication. Its core features include executing deterministic hard threshold judgments (failing if not met), summarizing winners via Pareto ranking, and providing a real-time web matrix interface to view running status and metrics. The tool incorporates local atomic persistence, concurrency control, and timeout abort functionality, and automatically sanitizes credentials before exporting reports. All operations are performed locally and offline, with no remote uploads. Installation requires building and testing first, then adding via the `pnpm dsh plugin` command, requiring Node