In the DeepSeek Harness (DSH) ecosystem, a common need is to perform a deep, figure-by-figure and experiment-by-experiment close reading of a single biomedical paper. Existing approaches often rely on manual organization or generic tools, which makes it difficult to guarantee the integrity of the logical chain and strict evidence boundaries. The dsh-plugin-deep-paper-reading plugin fills this gap by bundling a dedicated “Deep Paper Reading Mode” preset.

Plugin Overview

  • Name: dsh-plugin-deep-paper-reading
  • Owner: 1283011852
  • Category: Workflow
  • License: MIT
  • Core value: Structured close reading of a single biomedical paper, by figure and experiment.

Core Features

The preset provided by this plugin mainly has the following capabilities:

  1. Structured close reading: Structured close reading of a single biomedical paper by figure and experiment.
  2. Correspondence table generation: Generates a correspondence table between figure legends and the main text.
  3. Experiment breakdown: Breaks down each experiment into four elements (aim / design and tools / results / interpretation).
  4. Logical chain reconstruction: Reconstructs the logical chain and categorizes transitions.
  5. Evidence boundaries: Strict evidence boundaries (no fabrication of figure numbers, numeric values, P-values, etc.).
  6. Report output: Produces a report based on a template that can be used directly for journal club.
  7. Export support: Supports exporting to Word documents.

Installation and Activation

Method 1: CLI installation

If you have a local dsh CLI environment:

dsh plugin --profile web add github:1283011852/dsh-plugin-deep-paper-reading

After installation, restart DSH. You will then see Deep Paper Reading Mode in the preset selector.

Method 2: Desktop import

Desktop users are encouraged to directly download the .dshpreset file and import it in DSH Desktop. This method does not touch profile configuration at all and is therefore safer.

Usage

  1. Create a new session and select “Deep Paper Reading Mode” in the preset selector.
  2. Drop in the PDF or DOI.
  3. Enter the instruction “Read this paper in depth”.

Notes

  1. Dependency check: The local machine needs a toolchain such as Python and PyMuPDF to parse PDFs. If these are missing, the agent will flag the issue first and will not pretend to have read the full paper.
  2. Configuration override risk: When installing via Method 1, the plugin’s patch overrides the entire config of the agent-presets line. This means preset directories originally configured in roots may be replaced. If coexistence is required, manually merge roots in the profile patch.
  3. Applicable scenarios: This plugin is only intended for close reading of a single paper. It is not suited for multi-paper literature reviews, formal peer review, or literature searching.
  4. License note: The skill itself has not been modified and retains the original author’s MIT license.