Preface

DeepSeek Harness (DSH) is a plugin-based agent development platform. When large models are used for academic research or document writing, they sometimes fabricate references, authors, or DOIs, which is known as “hallucination.” The DeepSeek Harness academic literature plugin is designed to address this issue. It integrates into DSH as a pure utility plugin, responsible for retrieving data from real sources and generating citation formats, thereby preventing the model from fabricating reference information by design.

Core Capabilities

The plugin provides the following core capabilities:

  1. Multi-source retrieval: Supports two public data sources: arXiv and Semantic Scholar. Retrieval supports filtering by keywords, authors, and year ranges, and allows sorting by relevance, date, or citation count.
  2. Single-paper details: Full paper metadata can be retrieved by paperId, DOI, or arXiv ID, including the abstract, author list, journal/conference name, citation count, and PDF link.
  3. Citation format generation: Deterministically generates citations locally in GB/T 7714, APA, and BibTeX formats, without relying on large models.
  4. Local reference library management: Supports adding retrieved references to the current session’s reference library. The reference library supports session isolation and automatic deduplication (based on DOI/arXiv ID/title), as well as deletion, clearing, and batch export.
  5. Side panel and anti-hallucination: Provides a visual side panel (with light/dark theme support) for viewing history and managing the reference library. The System Prompt strictly requires that academic references must be retrieved through tools and forbids fabrication.

Installation and Enablement

Before installation, ensure that the DeepSeek Harness environment is installed, Node.js 22+ is used, and the Web profile is enabled.

Installation Command

Run the following command to add the plugin to the Web profile:

dsh plugin --profile web add dsh-plugin-academic-paper

Restart to Take Effect

After installation, restart the Web profile to activate the plugin:

dsh web

Verification

After restarting, a “📚 References” capsule entry will appear in the bottom-right corner of the DeepSeek Harness web page. Click it to expand the side reference library panel; at this point, the model can use the academic_* tool series.

Configuration

The plugin supports configuration, and changes usually take effect via hot reload without requiring a restart. The main configuration items are as follows:

  • enable: Whether to enable the plugin (default is true).
  • default_source: Default data source; options include arxiv, semantic_scholar, or both.
  • max_results: Maximum number of results returned per retrieval (1–50, default is 10).
  • default_cite_format: Default citation format; options include gb_t_7714, apa, or bibtex.
  • request_timeout: API request timeout in seconds (default is 15).
  • max_library_size: Maximum reference library size per session (default is 500).

Typical Usage

The features above can be invoked through natural language interaction with the model:

  1. Retrieve references:
    > “Help me search for papers about transformer, within the last 5 years, from arXiv.”
  2. Retrieve details:
    > “Retrieve the detailed information for Vaswani’s 2017 paper, Attention Is All You Need.”
  3. Generate citations:
    > “Add this paper to the reference library and then generate a citation in GB/T 7714 format.”
  4. Export the reference library:
    > “Export the reference library as BibTeX and Markdown.”

Notes

  • Pure utility plugin: This plugin does not provide paid full-text download services; it returns only metadata and public abstracts.
  • Data isolation: The reference library, retrieval history, and cache are stored in memory only, isolated by session. They are automatically cleared when the plugin is uninstalled or when Harness is restarted, with no disk residue.
  • License: Released under the MIT License.

Conclusion

By integrating public APIs and performing local deterministic processing, the plugin provides DSH with reliable academic reference retrieval and citation management capabilities. For developers or researchers who need a rigorous citation context, this is a necessary auxiliary tool.

  • GitHub: https://github.com/zhaoxuejie/dsh-plugin-academic-paper