Introduction

Under the “everything is a plugin” philosophy of DeepSeek Harness (DSH), a common requirement is to connect papers, literature context, code implementation, and research questions. academic-research-graph is a DSH plugin package. Its goal is not to output another citation list, but to generate an expandable research graph starting from a paper, a research concept, a benchmark, or a GitHub repository.

The following introduces the plugin’s positioning, core capabilities, installation methods, and several directly reusable request examples.

What is it

The repository is referred to as watericetangcw/academic-research-graph in the following text, while the package name in package.json is dsh-academic-research-graph.

It is maintained by watericetangcw and is released under the MIT License. This repository can also be used as a native DeepSeek Harness profile bundle, with the bundle format targeting @deepseek-ai/dsh 0.1.0-rc.5.

The core problem it addresses is: given a research entry, instead of simply listing relevant papers, it explicitly visualizes relationships such as “who proposed key ideas,” “which works truly extended them,” “which works questioned them,” “what the code actually does,” and “which papers truly used the benchmark,” while allowing further research to continue from existing nodes.

Core Features

Generate an Explorable Paper Graph

The generated graph page can display:

  • Clickable paper nodes and relationship edges;
  • preprint, conference, journal, and publication status;
  • sourced top-venue badges;
  • readable paper guides and metadata links;
  • highlighting main lineage and controversy;
  • optional representative figures for key papers;
  • Chinese, English, or bilingual content;
  • The latest expansion section is highlighted only after a genuine incremental update occurs.

Explain Papers, Context, and Disagreements

This plugin explains key nodes to those who are familiar with the field but have not yet opened the PDF.

It highlights the main lineage and controversy points, and can also identify genuine disagreements, including:

  • criticism;
  • contradictory findings;
  • failed reproductions;
  • incompatible assumptions.

Regarding benchmark-related issues, it also distinguishes between two types of papers: those that merely cite benchmarks and those that actually evaluate on the benchmark, load its data, or use its official code.

Connect Papers and Code

It supports starting from a GitHub repository:

  1. Parse the associated paper;
  2. Inspect the implementation on the pinned commit;
  3. Compare the code implementation with the methods described in the paper.

In other words, it not only visualizes literature relationships but also allows for a comparison of the “methods claimed in the paper” and the “implementation in the repository” within the same research context.

Support Incremental Expansion

You can select any existing node in the graph to start a new research branch, merge the reviewed results back into the original graph, and avoid rebuilding from scratch.

Output Language

Supports the following output modes:

output_language: zh-CN
output_language: en
output_language: bilingual

Installation and Usage

First, ensure your environment meets the following requirements:

  • Node.js ^22.19.0 or >=24.0.0;
  • pnpm available in PATH; corepack enable can provide pnpm.

If you do not yet have pnpm, you can enable corepack first:

corepack enable

Then install the plugin into the target profile. The commands below use the default web profile:

npx @deepseek-ai/dsh plugin --profile web add github:watericetangcw/academic-research-graph

If you need to pin to v0.2.4, you can append the version after the GitHub package spec:

npx @deepseek-ai/dsh plugin --profile web add github:watericetangcw/academic-research-graph#v0.2.4

If the dsh command is already installed on your machine, you can use dsh instead of npx @deepseek-ai/dsh:

dsh plugin --profile web add github:watericetangcw/academic-research-graph

Note: Plugins are profile-scoped. If you are actually running headless or another profile, you need to replace web with the corresponding profile name in the command.

After launching Harness, you can type /academic-research-graph and follow it with your research request. For example:

/academic-research-graph Research arXiv:2406.05720. Trace its important origins and meaningful follow-up work, identify major disagreements, and verify which papers actually used its benchmark. Produce the graph in English.

Harness may also automatically load this Skill when a request matches the plugin’s description.

Typical Usage

The following requests can be used directly as examples or rewritten according to your own research questions.

1. Trace the Context Starting from a Specified Paper

Research arXiv:2406.05720. Trace its important origins and meaningful follow-up work, identify major disagreements, and verify which papers actually used its benchmark. Produce the graph in English.

This request focuses on four things:

  1. The paper’s important origins;
  2. Meaningful follow-up work;
  3. Major disagreements;
  4. Which papers actually used its benchmark.

2. Build a Chinese Research Graph for a Research Concept

Build a Chinese research graph for Vision Transformer. Explain every core paper for a field-familiar reader who has not read it, and highlight the main lineage from Transformer to ViT and its most important descendants.

This request is suitable for those already familiar with the Transformer field but who want to quickly grasp the context of Vision Transformer. It requires explaining core papers and highlighting the main lineage from Transformer to ViT and its important subsequent branches.

3. Compare Papers and Code Starting from a GitHub Repository

Start from this GitHub repository, identify the associated paper, compare the implementation with the paper, and map its origins, follow-up work, and criticism.

This request uses the code repository as the entry point, identifies the associated paper, compares the implementation with the paper description, and supplements the information with origins, follow-up work, and criticism.

4. Continue Research from an Existing Node

Continue from the DeiT node in the current graph. Investigate the data-efficiency branch and merge the reviewed results back into the graph.

This request assumes there is already a DeiT node in the current graph. It will start from that node to investigate the data-efficiency branch in depth and merge the reviewed results back into the original graph.

Use Cases and Notes

Suitable for the following scenarios:

  • Want to quickly understand a paper’s origins, follow-ups, and disagreements;
  • Want to distinguish between “citing a benchmark” and “actually using a benchmark”;
  • Want to trace back from a GitHub repository to the paper and compare the paper’s methods with the code implementation;
  • Want to continue expanding a specific branch on an existing research graph;
  • Need Chinese, English, or bilingual output.

Note when using:

  • The plugin runs under the permissions of the current dsh process; you should check the source code and the MIT License before installing;
  • DSH plugins are profile-scoped; do not assume the web profile is suitable for your runtime environment;
  • The bundle format targets @deepseek-ai/dsh 0.1.0-rc.5; the environment must meet the Node.js and pnpm requirements;
  • To pin a version, you can use #v0.2.4;
  • The skillhub.cn plugin directory mentioned in the text is an independent site and should not be understood as the official app store for DeepSeek or Hypersphere.

Links

  • GitHub: https://github.com/watericetangcw/academic-research-graph
  • Plugin Directory Page: https://www.skillhub.cn/plugins/watericetangcw/academic-research-graph