In the plugin ecosystem of DeepSeek Harness (DSH), building a traceable and structured research process is a common need. Traditional conversation logs often lack systematic management of evidence chains and hypothesis evolution. SciFork is a local, Git-native biomedical research graph plugin that maps research processes into a graph, helping developers organize the relationships among questions, hypotheses, and evidence.

Introduction

SciFork is a plugin designed for DeepSeek Harness (DSH). It transforms research questions into connected and inspectable research graphs, supports literature-evidence-based extension, and maintains local Git auditability for project files. The project is maintained by user zhang-bin-98 and is released under the MIT license.

Core Features

The plugin provides the following capabilities:

  • Graph-based research: Turns open-ended research questions into connected and inspectable research graphs.
  • Import literature evidence: Bring in literature evidence via PMID or DOI.
  • Separation of concerns: Separates a research team’s “results” from “interpretations” and “unverified hypotheses”.
  • View switching: Provides a main view for inspecting the entire project, or an evidence view to focus on direct assertions of specific entities.
  • Literature-driven expansion: Runs single-step literature-driven expansion from the current focus.
  • Local snapshots: Creates local Git snapshots containing only files managed by SciFork.

Installation and Enabling

Before installation, ensure your environment meets the following requirements:

  • DSH 0.1.1-rc.2 (SciFork 0.0.3 is tightly coupled to this version’s interface).
  • DSH Web mode is configured for local loopback access (127.0.0.1).

Run the installation command:

dsh plugin --profile web add git+https://github.com/zhang-bin-98/sci-fork.git

After installation completes, restart DSH.

Typical Usage

  • Start DSH Web mode:
    dsh --profile web
  • Initialize a research project in Chat:
    /research init
  • Validate the project state:
    /research validate

Notes

  • Version restriction: SciFork 0.0.3 is tightly coupled to the public interface of DSH 0.1.1-rc.2.
  • Runtime environment: The plugin is designed to run locally on the DSH loopback web server.
  • Data security: The plugin treats literature, PDFs, model outputs, and project Markdown as untrusted data. It does not upload projects or add cloud synchronization, and it does not store full abstracts or PDF content.
  • Privacy compliance: Before submitting or sharing a research project, check whether it contains PHI (Protected Health Information), PII (Personally Identifiable Information), or restricted-access data.

SciFork is suitable for developers who need to manage research workflows locally and ensure auditability of evidence chains. For more details and source code, see its GitHub repository.