In DeepSeek Harness development practices, invoking external tools is a key step in building intelligent agents. Directly prompting the model to read skill descriptions and execute shell commands is often inefficient and error-prone. To address this issue, developer Ln1m provides the dsh-tool-literature plugin, which encapsulates the literature search workflow as a structured tool that can be invoked directly by the model.

Plugin Overview

dsh-tool-literature is a model-callable literature search tool for DeepSeek Harness. It is built on the paper-search.py script and integrates OpenAlex citation-based ranking and arXiv relevance evaluation. The plugin belongs to the online tool category, aiming to replace the traditional flow of reading skill descriptions and executing shell commands with a single structured call.

Core Features

  1. Model-callable interface: Provides a structured literature_search tool for agents to invoke directly.
  2. Script encapsulation: Internally calls the paper-search.py script to handle search logic.
  3. Citation ranking: Uses the OpenAlex data source to rank search results by citation count.
  4. Relevance scoring: Combines arXiv relevance algorithms to optimize search results.

Installation and Dependencies

According to the provided materials, the plugin repository does not explicitly list standard installation commands. The plugin defines its dependencies and configuration through package.json.

The following key dependencies can be found in package.json:
* @deepseek-ai/cordis (version >= 4.0.1)
* @deepseek-ai/dsh-tools (version >= 0.1.0-rc.6)

Internally, the plugin locates the paper-search.py script path through the environment variable DSH_PAPER_SEARCH_SCRIPT (default is ~/Research-Tools/paper-search.py). Since this repository has been archived and merged into dsh-tool-suite, it is recommended to follow the merged repository directly for integration.

Typical Usage

With this plugin, the agent no longer needs to understand natural language descriptions of skill modules, nor does it need to manually construct shell commands. A single structured tool call is sufficient to complete a literature search. Internally, the plugin calls the configured paper-search.py script and combines the algorithms of OpenAlex and arXiv to rank the results.

Applicable Scenarios and Notes

  • Applicable scenarios: Research-oriented agents that require automated literature search, citation ranking, and relevance scoring.
  • Notes:
    1. This plugin repository has been archived; its contents have been merged into dsh-tool-suite, it is no longer updated, and only historical records are preserved.
    2. The plugin runs with the permissions of the current DSH process. Before installation, the source code and MIT license should be reviewed.

Summary

This tool simplifies how agents interact with literature search tools. For more details and source code, refer to the project homepage:
* GitHub repository