dsh-scientific

DeepSeek Harness 的科研计算插件套件 —— 让 DSH agent 直接做文献调研、分子对接、结构可视化、分子动力学。

中文 | English

---

这是什么

dsh-scientific 是面向 DeepSeek Harness (DSH) 的科研计算插件 monorepo。把文献调研 → 靶点/配体准备 → 对接筛选 → MD 验证 → 组会汇报这条全链路，统一到 DSH 的对话框里，让 agent 用 DSH 工具 + skills 按科研方法论一气呵成。

它不是另一个 AI 助手 —— 是 DSH 的插件包，不开新窗口、不存新数据；DSH 在哪，它就在哪。

由合成生物学研究者在真实科研工作流中打磨。

仓库状态

模块	状态	说明
dsh-zotero	✅ 完成	Zotero 文献库检索/全文/附件/笔记，9 个工具（独立仓库）
plugins/vina	🔲 骨架	AutoDock Vina 分子对接（规划中）
plugins/pymol	🔲 骨架	PyMOL 无头脚本（规划中）
plugins/chimerax	🔲 骨架	ChimeraX 无头脚本（规划中）
plugins/gromacs	🔲 骨架	GROMACS MD 流程（规划中）
skills/	✅ 5 个	literature-review / batch-docking / docking-analysis / molecular-dynamics / journal-club
workflows/	✅ 3 个	literature-research / literature-to-slides / docking-workflow
examples/	✅ 2 个	vina-docking / gromacs-md 配置示例

dsh-zotero 已独立成仓库（2026-08 拆分）：Hongcheng-LI/dsh-zotero。本仓库的 plugins/ 不再包含 zotero，skills / workflows 里引用 zotero_* 工具时需先单独安装该插件。

骨架插件包结构齐全、装上不弄崩启动；通过 DSH plugin-check 审计。等核心算法 + UI 敲定后再注册工具。

三层架构

┌────────────────────────────────────────────┐
│  Workflows（人读手册，多插件端到端）        │
│  workflows/literature-research.md          │
│  workflows/docking-workflow.md             │
│  workflows/literature-to-slides.md         │
└──────────────────┬─────────────────────────┘
                   │ 引用
┌──────────────────▼─────────────────────────┐
│  Skills（模型读方法论，按 description 触发）│
│  skills/literature-review/                 │
│  skills/batch-docking/                     │
│  skills/docking-analysis/                  │
│  skills/molecular-dynamics/                │
│  skills/journal-club/                      │
└──────────────────┬─────────────────────────┘
                   │ 引用具体工具
┌──────────────────▼─────────────────────────┐
│  Plugins（DSH 工具层）                      │
│  dsh-zotero      (独立仓库, 9 tools, ready) │
│  plugins/vina/     (skeleton)               │
│  plugins/pymol/    (skeleton)               │
│  plugins/chimerax/ (skeleton)               │
│  plugins/gromacs/  (skeleton)               │
└────────────────────────────────────────────┘

• Plugins = "能调什么"（DSH 工具）
• Skills = "怎么干"（方法论，模型按需触发）
• Workflows = "组合起来怎么用"（人读操作手册，含检查点）

快速开始

1. 装 DSH（WorkBuddy 用户可跳过）

npm install -g @deepseek-ai/dsh

2. 克隆并构建插件

git clone https://github.com/Hongcheng-LI/dsh-scientific
cd dsh-scientific
npm install && npm run build

3. 安装你需要的插件

Zotero 插件是独立仓库，直接一行装：

dsh plugin --profile web add github:Hongcheng-LI/dsh-zotero

骨架插件（vina / pymol / chimerax / gromacs）仍在 plugins/ 下，按目录打包安装（目前不注册工具，安装无副作用）。本仓库是 monorepo，根目录不是插件包。

# 重启 DSH
dsh --profile web

4. 试试看

DSH 对话框里说：

在我的 Zotero 里搜一下 2023 年以后的 CRISPR 相关论文，挑两篇读全文。

→ 触发 zotero_search → zotero_item → zotero_fulltext，配合 skills/literature-review/ 工作流。

插件清单

插件	工具数	状态	前置软件	文档
dsh-zotero	9	✅ 完成（独立仓库）	Zotero 7+	github.com/Hongcheng-LI/dsh-zotero
dsh-vina	0	🔲 骨架	AutoDock Vina	README
dsh-pymol	0	🔲 骨架	PyMOL	README
dsh-chimerax	0	🔲 骨架	ChimeraX	README
dsh-gromacs	0	🔲 骨架	GROMACS	README

Skills & Workflows

• Skills（模型按 description 触发）
  • literature-review — 系统性文献综述
  • batch-docking — 批量对接编排
  • docking-analysis — 对接打分表解读
  • molecular-dynamics — MD 流程与平衡判据
  • journal-club — 组会汇报生成
• Workflows（人读手册）
  • literature-research.md — 从选题到综述
  • literature-to-slides.md — 文献到组会 PPT
  • docking-workflow.md — 对接全流程

开发

git clone https://github.com/Hongcheng-LI/dsh-scientific
cd dsh-scientific
npm install
npm run build   # 构建全部插件
npm test        # 跑全部测试

• 写新插件 → docs/plugin-development.md
• 写新 skill → docs/skill-development.md
• 入门 → docs/getting-started.md

审计

每个插件用 @deepseek-ai/dsh-plugin-check 体检。当前成绩：

$ cd dsh-zotero && npm run audit   # 独立仓库里跑
verdict: PASS
checks: 18 pass / 0 fail / 0 warn / 1 skip

其他 4 个骨架插件等填 apply() 后补检。zotero 的审计记录已随仓库迁移：github.com/Hongcheng-LI/dsh-zotero/tools/。

路线图

• zotero 插件完成（含 plugin-check 审计通过）
• vina 工具（Vina 版本探测 + 对接打分表解析）
• pymol / chimerax 工具（无头脚本 + 图像导出）
• gromacs 工具（子命令白名单 + stdin 应答交互选择）
• synbio 插件（Europe PMC / KEGG / RCSB PDB，国内网络直连可达）
• docs/ARCHITECTURE.md（详细分层设计）
• CONTRIBUTING.md（贡献指南）

贡献

Issue / PR 欢迎。开发规范见 docs/plugin-development.md。

License

MIT —— 详见 LICENSE。

引用

如果本仓库对你的研究有帮助，请引用：

@software{li2026dshscientific,
  title  = {dsh-scientific: DeepSeek Harness plugin suite for scientific computing},
  author = {Li, Hongcheng},
  year   = {2026},
  url    = {https://github.com/Hongcheng-LI/dsh-scientific}
}

---

What is this

dsh-scientific is a plugin suite for DeepSeek Harness (DSH) covering literature management, molecular docking, structure visualization and molecular dynamics — built by synthetic biology researchers for real lab workflows.

End-to-end pipeline: literature review → target/ligand prep → docking screen → MD validation → journal club.

It is not another AI assistant — it's a plugin pack for DSH; no separate window, no separate data store.

Repository status

Module	Status	Notes
dsh-zotero	✅ Done	Zotero library search/fulltext/attachment/note, 9 tools (standalone repo)
plugins/vina	🔲 Skeleton	AutoDock Vina molecular docking (planned)
plugins/pymol	🔲 Skeleton	PyMOL headless scripts (planned)
plugins/chimerax	🔲 Skeleton	ChimeraX headless scripts (planned)
plugins/gromacs	🔲 Skeleton	GROMACS MD flow (planned)
skills/	✅ 5	literature-review, batch-docking, docking-analysis, molecular-dynamics, journal-club
workflows/	✅ 3	literature-research, literature-to-slides, docking-workflow
examples/	✅ 2	vina-docking, gromacs-md

dsh-zotero split into its own repo (2026-08): Hongcheng-LI/dsh-zotero. This repo's plugins/ no longer contains zotero; install it separately for the zotero_* tools referenced by skills/workflows.

Skeleton plugins pass the DSH plugin-check audit; structure is complete, no tools registered yet, safe to install.

Three-layer architecture

Workflows (human-readable manuals, end-to-end)
    ↓ references
Skills    (model-readable methodology, triggered by description)
    ↓ references
Plugins   (DSH tool layer)

• Plugins = what you can call (DSH tools)
• Skills = how to do it (methodology the model triggers by need)
• Workflows = how to combine them (human-readable playbooks with checkpoints)

Quick start

1. Install DSH (skip if using WorkBuddy)

npm install -g @deepseek-ai/dsh

2. Clone and build

git clone https://github.com/Hongcheng-LI/dsh-scientific
cd dsh-scientific
npm install && npm run build

3. Install plugins

The Zotero plugin lives in its own repo — one-line install:

dsh plugin --profile web add github:Hongcheng-LI/dsh-zotero

dsh --profile web

Skeleton plugins (vina/pymol/chimerax/gromacs) are under plugins/; install per directory via npm pack + dsh plugin add. They register no tools yet; installing them has no effect.

4. Try it

In the DSH chat:

Search my Zotero for CRISPR papers since 2023, pick two and read the full text.

→ triggers zotero_search → zotero_item → zotero_fulltext, guided by skills/literature-review/.

Plugin catalog

Plugin	Tools	Status	Prerequisite	Docs
dsh-zotero	9	✅ done (standalone)	Zotero 7+	github.com/Hongcheng-LI/dsh-zotero
dsh-vina	0	🔲 skeleton	AutoDock Vina	README
dsh-pymol	0	🔲 skeleton	PyMOL	README
dsh-chimerax	0	🔲 skeleton	ChimeraX	README
dsh-gromacs	0	🔲 skeleton	GROMACS	README

Development

git clone https://github.com/Hongcheng-LI/dsh-scientific
cd dsh-scientific
npm install
npm run build
npm test

• Plugin dev → docs/plugin-development.md
• Skill dev → docs/skill-development.md
• Getting started → docs/getting-started.md

Audit

Each plugin is checked by @deepseek-ai/dsh-plugin-check. Current scores:

$ cd dsh-zotero && npm run audit   # in the standalone repo
verdict: PASS
checks: 18 pass / 0 fail / 0 warn / 1 skip

The 4 skeleton plugins will be checked after they fill in apply().

License

MIT — see LICENSE.

Citation

@software{li2026dshscientific,
  title  = {dsh-scientific: DeepSeek Harness plugin suite for scientific computing},
  author = {Li, Hongcheng},
  year   = {2026},
  url    = {https://github.com/Hongcheng-LI/dsh-scientific}
}