Introduction

In the DeepSeek Harness (DSH) ecosystem, plugins are the primary way to extend capabilities. dsh-research-check focuses on compliance checking for deliverables (papers, documents, data, bids). It converts requirement documents into executable specifications, verifies numeric consistency, audits metadata, and generates provenance ledgers. The plugin does not generate content; it only performs verification.

Core Features

  • Specification Construction: Converts requirement documents into executable specifications.
  • Compliance Checking: Checks deliverable compliance (page count, abstract, citations, identity information).
  • Numeric Consistency: Verifies numeric consistency (contradictions).
  • Provenance Ledger: Builds a numeric provenance ledger.
  • Metadata Audit: Audits identity information in PDF/OOXML metadata.
  • MCP Support: Supports the MCP protocol (Claude Code, Codex, Cursor).
  • Cross-scenario Rules: Supports rules for multiple scenarios, including academic, software, data, documentation, and bidding.

Installation and Enablement

Use the official command to install:

dsh plugin --profile web add dsh-research-check

After installation, the plugin is registered when the DSH profile starts; restart the profile.

Typical Usage

1. Build Specifications

Convert requirement documents into an executable checklist:

python python/spec_build.py --requirements 格式规范.doc --out specs/我的要求. --profile academic --name "竞赛论文格式规范"

2. Run Compliance Checking

Check deliverables item by item:

python python/check_spec.py --spec specs/我的要求. --root . --pdf 我的论文.pdf --files 交付文件1.xlsx 交付文件2.docx --archive 材料.zip

3. Manage Numeric Ledgers

Build a “number-source” ledger to track the origin and changes of key numbers:

# Train ledger: identify numbers from the manuscript
node lib/ledger-cli.js teach --ledger paper-ledger. --paper 论文.tex --unit-filter 万元 --min-abs 100

# Record a number and its source
node lib/ledger-cli.js add --ledger paper-ledger. --key q3.total_cost --value 1521.6 --unit 万元 --source code/q3.py --anchor 全年费用

# Verify ledger: check numeric consistency within the document
node lib/ledger-cli.js verify --ledger paper-ledger. --paper 论文.pdf --near

4. Metadata Audit

Check for identity information leakage and manifest consistency in PDFs and archives:

python python/audit_paper.py --paper 论文.pdf --archives 材料.zip

5. Access via MCP

If not using DSH, you can integrate other AI tools through the MCP protocol:

{
  "transport": "stdio",
  "serverName": "research-check",
  "command": "node",
  "args": ["<插件目录>/mcp/server.mjs"]
}

Environment Requirements and Dependencies

  • Python Version: Requires Python 3.10+.
  • Peer Dependencies:
    • @deepseek-ai/cordis
    • @deepseek-ai/dsh-tools
    • @deepseek-ai/schemastery
  • Execution Environment: All checks are completed locally; no data is collected.

Summary

dsh-research-check provides a complete workflow from “requirements” to “specifications” and then to “provenance ledgers”. It binds numbers to sources, converts rules into executable scripts, and is suitable for various deliverable scenarios such as papers, software, data, documentation, and bids.