dsh-model-arena：同一提示词多模型并排对比与评测
dsh-model-arena 是 hj01857655 发布的 DSH web 插件，对多个模型用相同 prompt 与参数批量运行，并排查看输出，支持行级 diff、自动与人工评分、排行榜及 benchmark 回归报告，并提供 dsh-model-arena CLI 与 Settings 中的 Model arena 面板。安装命令为 dsh plugin --profile web add dsh-model-arena；数据持久化在 .arena/。SkillHub 标注原仓库已归档并合并进 dsh-plugins monorepo，包仍从 npm 安装。使用前需知插件以 dsh 进程权限运行，并自行审阅 MIT 源码。
Read Moredsh-agent-arena: Multi-model coding comparison and deterministic evaluation in the DSH plugin
`dsh-agent-arena` is a DSH plugin maintained by LeemanCheung, designed to compare the coding results of 2 to 4 different models through isolated execution, unified verification, and an explicit application process. The plugin launches model competitions in independent Git worktrees, utilizing `ctx.subprocess.spawn` to execute conservative argv verification commands (e.g., `pnpm test`), ensuring the process is shell-independent and does not use an LLM judge. Scoring is based on the weight percentage of successful verification command execution, with ties broken by the contestant ID. Users can review a diff covering the status of all files and explicitly apply the winner's changes only after confirming that the HEAD version remains unchanged and the workspace is clean.
Read More