DeepSeek Harness Plugin Hub

发布与管理完整 Harness Profiles，发现适合你的插件。

探索

插件目录环境预设文档中心动态

社区

发布插件联系我们报告问题

相关链接

Plugin Hub GitHubDeepSeek Harness 官方项目系统状态隐私说明
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

独立、非官方社区项目，与 DeepSeek 官方无隶属、授权或背书关系。

Dual Model Eval — DeepSeek Harness 插件（DSH Plugin）
DeepSeek Harness Plugin Hub
ProfilesPlugins分类动态文档登录管理 Profiles
ProfilesPlugins分类动态文档登录
← Plugins
D

dsh-dual-model-eval

Dual Model Eval

在 DeepSeek Harness 中使用隔离的 Git 工作树并排比较多个编码模型

插件会安装到这里；不确定时保持 web。

npx -y @deepseek-ai/dsh plugin --profile web add github:huangdaxianer/dsh-dual-model-eval#799b701efc099bff629dcc1441febec71e3adf75
README兼容性版本

兼容性与来源证明

Dual Model Eval 以 dsh-dual-model-eval 发布，当前版本为 0.1.1。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。

DSH 兼容范围
*
运行环境
web
发布来源
github
Registry 更新时间
2026/8/21

版本

0.1.1stable
2026/8/21

相关插件

正在加载相关插件…

最新版
0.1.1
DSH
*
HMR
重启进程
Tree shaking
未声明可安全裁剪
解包体积
未提供
文件数
未提供
Surface
web
许可证
MIT
发布源
github
GitHub
★ 0
周下载
0
查看源码 ↗
README Badge

点击下方 Badge 复制 Markdown，粘贴到 README 即可。

这是你的 Plugin？认领权益 · 优先安全扫描

验证 package.json 声明的 GitHub 仓库，即可管理这个公开页面。认领后，Hub 会优先安排当前版本的安全扫描，并在通过后公开展示结果。

认领这个 Plugin →
报告问题

相关插件

继续浏览 developer-tools 分类下经过校验的插件。

Web App@deepseek-ai/dsh-web-appdsh 浏览器界面捆绑包：位于 dsh-base 之上的 Web 补丁层，加上运行时粘合插件（提供前端 dist、Web 界面提示符、bash 运行时变量和 URL 行）Sdk Minimal@deepseek-ai/dsh-sdk-minimal独立的最小 SDK 配置包：JSON-RPC、一个 DeepSeek 适配器、持久化 Shell 和 JSONL 会话Sdk App@deepseek-ai/dsh-sdk-appdsh SDK 配置包：基于 dsh-base 提供 stdio JSON-RPC 服务和进程生命周期管理Subagent Codex@deepseek-ai/dsh-subagent-codex基于官方 app-server 协议的一次性 Codex 子代理提供程序

README

DeepSeek Harness Multi-Model Evaluation (dsh-dual-model-eval)

English | 简体中文

Compare multiple coding models side by side inside DeepSeek Harness. dsh-dual-model-eval is an installable multi-model and coding-agent evaluation plugin: one prompt runs concurrently across selected LLM routes, every model works in an isolated Git worktree, and the normal Chat tab streams tool trajectories and renders comparable, adoptable results.

Compatibility: the first release targets DeepSeek Harness 0.1.0-rc.7. DeepSeek Harness is currently a developer preview, so plugin APIs may change between release candidates.

Install

Install the pinned release into the built-in web profile:

npx @deepseek-ai/dsh@0.1.0-rc.7 plugin --profile web add github:huangdaxianer/dsh-dual-model-eval#v0.1.1

Restart the Harness web process after installation:

npx @deepseek-ai/dsh@0.1.0-rc.7 web

The repository contains committed, prebuilt lib/ artifacts. Installing from GitHub therefore does not require authorizing a dependency prepare script.

What it adds

  • A Comparison test switch inside the existing model selector.
  • Multi-select for two to four configured model routes.
  • Concurrent execution from one shared Git commit in isolated worktrees.
  • Live, independently expandable tool-call trajectories for every model.
  • Compact elapsed-time and tool-count summaries, with token, cache, TTFT, decoding, and tool details inside the expanded view.
  • Side-by-side final replies and line-change statistics (+added, -deleted, percentages, and changed-file count).
  • Changed-file preview with downloads for one edited file or the complete candidate worktree.
  • An Adopt this result action that commits the chosen patch locally and advances the shared baseline for later rounds.
  • Later rounds inherit the accepted requests, final responses, and commits, while unselected candidate responses stay out of the shared context.
  • A composer gate that prevents the next round until one completed candidate has been adopted.

Use

  1. Open the standard model menu and enable Comparison test.
  2. Select at least two model routes that are already configured in Harness.
  3. Submit one coding request. Result cards appear immediately and stream progress while both agents run.
  4. Inspect each answer and its expandable tool trajectory.
  5. Choose Adopt this result under the preferred candidate. The next comparison starts from that committed code baseline and receives the accepted conversation context.

Git and workspace behavior

  • Every candidate gets a new detached worktree created from the same base commit.
  • Existing Git repositories must be clean before a comparison starts.
  • A non-Git workspace is initialized automatically by default. The plugin stages files not ignored by .gitignore and creates a local baseline commit.
  • Adoption reconstructs the selected full patch, creates a local commit with command-scoped identity, and fast-forwards the source workspace.
  • The plugin never pushes to a remote and does not modify global Git identity.
  • Temporary worktrees are removed after a run by default. Bounded private evidence remains under ~/.dsh/dual-model-eval/.

Review generated commits and patches before publishing them. A completed orchestration run is evidence that execution settled, not a model-quality score.

Update or remove

Update to a newer tagged release:

npx @deepseek-ai/dsh@0.1.0-rc.7 plugin --profile web add github:huangdaxianer/dsh-dual-model-eval#v0.1.1

Remove the bundle:

npx @deepseek-ai/dsh@0.1.0-rc.7 plugin --profile web remove dsh-dual-model-eval

Restart the web profile after either operation.

Development

Requirements: Node.js 22.19+ (or 24+), pnpm, and Git.

pnpm install
pnpm run check
pnpm run pack:release

To validate a local checkout without changing your normal Harness home:

export DSH_PLUGIN_TEST_HOME="$(mktemp -d)"
DSH_HOME="$DSH_PLUGIN_TEST_HOME" npx @deepseek-ai/dsh@0.1.0-rc.7 plugin --profile web add ./dsh-dual-model-eval
DSH_HOME="$DSH_PLUGIN_TEST_HOME" npx @deepseek-ai/dsh@0.1.0-rc.7 --profile web --dump-config

Security

DeepSeek Harness plugins execute with the permissions of the local user. Inspect the source and pin a release tag or commit before installing any third-party plugin. This plugin can create Git worktrees and local commits inside the selected workspace; it does not push those commits.

License

MIT. This standalone package includes code derived from the MIT-licensed DeepSeek Harness project; see NOTICE.