Introduction

The DSH plugin ecosystem emphasizes “Everything is a plugin.” For Chinese DSH users, a common requirement is to include study materials, text transcriptions, and audio transcriptions in the current session for review, rather than setting up a separate, complex external system.

dsh-project-knowledge-review is a standalone local knowledge review plugin for Chinese DSH users. It saves text, OCR, and ASR materials to a local knowledge base, prioritizing the retrieval of local evidence before answering; in strict mode, it explicitly refuses to answer if evidence is insufficient.

What is this

The basic information of this plugin is as follows.

  • Plugin Name: dsh-project-knowledge-review
  • Positioning: Standalone local knowledge review plugin for Chinese DSH users
  • Maintainer: HeWhenJay
  • License: MIT
  • Ecosystem Position: A community project in the DSH plugin ecosystem. The community directory is an independent site with no official affiliation with DeepSeek / HF-01.

Core Features

This plugin mainly provides the following capabilities:

  • Zero-configuration local knowledge base, automatically creates v2 storage on first use
  • Evidence-first Q&A: Retrieves local evidence first, then answers based on evidence
  • In strict mode, explicitly refuses to answer if evidence is insufficient
  • Automatically generates Markdown summaries after text, OCR, and ASR are imported
  • Filters import metadata, RAG templates, and pure timestamps
  • Two-level classification: System initial classification and user-adjustable secondary classification
  • Sidebar knowledge base workspace for viewing detailed summaries and original content
  • Safe Markdown preview, supports rendered view and Markdown source code view, does not execute HTML or scripts
  • Optional OCR/ASR, disabled by default
  • Requires explicit enabling and DSH credential configuration when OCR/ASR is enabled
  • Lightweight JSONL index, single document source files, cursor pagination, and on-demand source reading
  • Automatically migrates v1 old libraries to v2 and retains original backups
  • The plugin only provides evidence retrieval and system constraints; the final answer is generated by the current session model

Installation and Usage

First, install the plugin. The installation command is as follows:

dsh plugin --profile web add dsh-project-knowledge-review

After installation, restart the current DSH Web process and refresh the page.

On first use, the plugin automatically creates v2 storage. The default local storage path is:

~/.dsh/project-knowledge-review/knowledge.json

By default, no database, vector model, account login, or project token is required.

If using OCR or ASR, it needs to be explicitly enabled. Steps are as follows:

  1. Open “Settings → Knowledge Review”
  2. Enable the corresponding service
  3. Configure Base URL and Model Name
  4. Save the API Key to the DSH credential store

The API Key is not written to the plugin settings JSON and will not be echoed back.

Typical Usage

Adding Text Materials

In a DSH conversation, paste the text directly and explain the information to be added to the knowledge base:

Add the following content to the knowledge base, the title is..., and the source is...:

(Paste text you have the right to use)

This step adds the text materials you provided to the local knowledge base.

Review Based on Knowledge Base

Ask questions directly in the conversation:

Explain... based on my knowledge base

The plugin will first retrieve local evidence, then let the current session model answer based on the evidence.

Query Knowledge Base Overview

You can ask the following questions:

How many materials are in my knowledge base?
Where are the knowledge base materials stored?
Is the knowledge base shared with the current project?

These queries are used to check the number of materials, storage location, and sharing scope.

Browsing Materials in Sidebar

Click “Knowledge Base” in the sidebar to search local materials and browse by system classification or personal classification.

While reading, you can switch between the following view groups:

  • Summary / Original Content
  • Rendered / Markdown Source Code

Safe Markdown preview does not execute HTML or scripts.

Switching Answer Strategies

In “Settings → Knowledge Review”, you can switch answer strategies:

  • reference: Allows the current session model to supplement general knowledge, but requires distinguishing between knowledge base evidence and model supplements
  • strict: Explicitly refuses to answer when there is insufficient evidence

Use Cases and Notes

This plugin is suitable for these scenarios:

  • You want to keep Chinese study materials, notes, text transcriptions, or audio transcriptions in a local knowledge base for review
  • You want to prioritize using local evidence before answering
  • You want to avoid the current session model drawing conclusions based on insufficient evidence in strict mode

Pre-use warnings:

  • The plugin runs with the permissions of the current DSH process; check source code and MIT license before installing
  • Only import materials you wrote, have public authorization for, or have confirmed rights to use
  • The plugin does not connect to, install, or identify any external projects; sharedWithCurrentProject is always false
  • OCR/ASR is disabled by default; when enabled, it only accepts public HTTP(S) image URLs or public directly downloadable audio URLs
  • OCR/ASR will reject URLs with user info, non-HTTP(S) protocols, localhost addresses, and private network addresses
  • Safe Markdown preview does not execute HTML or scripts
  • v1 old libraries will be automatically migrated to v2 and retain original backups

Conclusion

dsh-project-knowledge-review provides a relatively lightweight local knowledge review path: local knowledge base, evidence-first Q&A, Markdown preview, and secondary classification.

Related entry points are as follows:

  • Plugin directory page: https://www.skillhub.cn/plugins/HeWhenJay/dsh-project-knowledge-review
  • GitHub: https://github.com/HeWhenJay/dsh-project-knowledge-review