Introduction

The plugin architecture of DeepSeek Harness (DSH) supports extending local retrieval capabilities. For developers who need to access classical texts (such as the Daozang) but want to keep operations completely offline, dao-zang-skill provides a complete solution. Without calling an external Embedding API or making network requests, it searches the Zhonghua Daozang and Zhengtong Daozang directly in a local database and supports extracting marked original text from source Markdown files.

Plugin Overview

This is a DSH plugin maintained by Godners-Code, designed to solve local retrieval and original-text traceability for classical texts. It has built-in text and semantic dual engines, can fully index 285,117 text chunks, and provides one-click environment self-check and workspace initialization.

Core Capabilities

  • Text Engine (default): Based on ChromaDB for full-text filtering, combined with TF/IDF ranking, with zero dependencies and fast startup.
  • Semantic Engine (optional): Runs the bge-m3 ONNX model locally to generate 1024-dimensional cosine vectors, consistent with the database vector dimensions, and supports semantic similarity search.

Original Text Location and Extraction

Using the --original parameter, the plugin can locate the specific position in the original .md file, return a text fragment marked with ⟦...⟧ and the line number, making it convenient for subsequent processing or manual proofreading.

File Filtering

Using the --source parameter, the search scope can be limited to files whose filenames contain specific keywords, improving retrieval efficiency.

Environment and Deployment

  • Self-check: Provides the check_env.py --selftest command to check environment configuration and run smoke tests.
  • One-click workspace setup: The setup_workspace.py script can download the dataset from HuggingFace and rebuild the local ChromaDB offline.

Installation and Activation

Run the following command in the terminal to install the plugin:

dsh plugin --profile web add dao-zang-skill

After installation, restart dsh web. In the chat interface input box, type / and select dao-zang, or instruct the assistant in the conversation with “use the dao-zang skill” to activate it.

Data Preparation

The plugin requires local data support. The workspace directory must contain ChromaDB/ (285,117 text chunks) and Markdowns/ (3,152 files).

Initialize the workspace:

python assets/dao-zang/scripts/setup_workspace.py --dir <workspace>

Use Cases and Notes

  • Offline scenarios: The plugin runs completely offline, suitable for intranet environments or scenarios with strict data privacy requirements.
  • Environment requirements: Ensure the runtime environment satisfies the dependency versions defined in package.json (e.g., Node.js >=20).
  • Data scale: The local ChromaDB data volume is relatively large (about 280,000 text chunks); first initialization or index rebuilding requires some disk space and time.

Summary

dao-zang-skill provides DSH users with complete local classical-text retrieval capabilities, supports multiple retrieval methods from keyword search to semantic search, and includes precise original-text traceability. For more details and source code, visit GitHub.