Introduction

The philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” Conducting academic literature research is cumbersome and prone to omissions, from searching and verification to synthesis. dsh-scopus-searcher is an academic literature research plugin for DSH. It includes two Agent presets, scopus-scholar (search-driven) and scopus-knowledge (knowledge-driven), aiming to provide an automated solution from literature discovery to bidirectional verification.

Core Capabilities

This plugin provides the following core features:

  1. scopus-scholar (search-driven mode): Supports Scopus discovery and OpenAlex enrichment, executing a closed loop of search → synthesis → claim → bidirectional verification → next step.
  2. scopus-knowledge (knowledge-driven mode): Enforces a code-based pipeline and consults an independent “scientific advisor” persona for verification.
  3. scopus-search (skill): Exact-phrase search and abstract retrieval.
  4. openalex_tool (tool): Batch enrichment of authors, citation counts, abstract reconstruction, and bidirectional citation tracking.
  5. Bidirectional verification: Claims must be attached with an evidence chain and confidence level.

Preset One: scopus-scholar (Search-Driven Mode)

Search-driven mode turns the model into a research assistant that can retrieve information, distill scientific claims, and perform bidirectional verification. Its core loop is: search → synthesis → claim → bidirectional verification → next step.

In this mode, the model performs the following operations:
* Multi-angle parallel retrieval: Uses Scopus and OpenAlex data to perform combined searches across synonyms and mechanism terms.
* Separation of claims and evidence: Distinguishes “proof” from “assertion,” requiring every claim to be attached with an evidence chain (paper/DOI/data point) and a confidence level.
* Bidirectional citation tracking: Uses OpenAlex tools to perform bidirectional citation network tracking without being restricted by the 40-citation limit.

Preset Two: scopus-knowledge (Knowledge-Driven Mode)

Knowledge-driven mode upgrades the research workflow from “prompt-level conventions” to a “code-enforced state-machine pipeline.” It enforces the flow THINK → VERIFY → SYNTHESIZE → SEARCH → DECIDE → REPORT through the knowledge-pipeline plugin.

Key features include:
* Scientific advisor persona: The main agent creates an independent “scientific advisor” session for consultation. This advisor persona is disabled from all tool usage at the code level and works solely from model memory.
* Hard gates: The pipeline has hard gates; for example, retrieval tools are prohibited during the THINK stage, and the advisor must be consulted first.
* Liveness heartbeat: The advisor session uses a 60-second heartbeat to monitor liveness and prevent deadlocks.

Technical Implementation

  • Dual-source division: Scopus handles discovery and abstract retrieval; OpenAlex handles metadata enrichment (addressing issues such as truncated Scopus abstracts and missing citation counts).
  • Pipeline: A code-level state machine with hard gates (for example, retrieval is prohibited during the THINK stage).

Installation and Configuration

Before use, configure the environment variable SCOPUS_API_KEY. The plugin is distributed as an npm package and includes two presets and shared skill files.

Typical Usage

  1. Select the scopus-scholar preset in DSH for literature retrieval and bidirectional verification.
  2. Select the scopus-knowledge preset in DSH for in-depth research driven by scientific curiosity.
  3. The pipeline automatically creates an independent scientific advisor session and maintains the heartbeat.

Applicable Scenarios and Notes

  • Applicable scenarios: Academic literature research, review writing, and claim verification that requires a strict evidence chain.
  • Notes:
    • License: MIT.
    • In knowledge-driven mode, the scientific advisor persona has no tool-calling permissions.
    • The pipeline has hard gates (for example, retrieval is prohibited during the THINK stage).
    • Dual-source division: Scopus handles discovery, and OpenAlex handles enrichment.