Preface¶

When writing agent workflows in DSH, a common approach is to “read a book” within a casual request: the user sends a sentence like “Help me tell you about a book,” and the model returns a summary of varying length. The quality depends on the prompt, and it is unclear exactly what the book is about.

dsh-plugin-book-deconstruct solidifies this into a plugin. It acts as a “book deconstruction” tool in the conversation, invoked by the agent, with a fixed output of two parts: a clear explanation of the entire book’s plot, and a piece of deep thought. For DSH developers who need to perform book deconstruction, summarization, or content organization, this is a solution that can be directly integrated into a profile, rather than a prompt that needs to be rewritten every time.

Below, we introduce the origin of this plugin, its installation method, and usage.

Plugin Information¶

The plugin repository is zmz123612/dsh-plugin-book-deconstruct, maintained by zmz123612, published on the community directory SkillHub, and categorized as Workflow. Judging from the installation path and README, it is a DeepSeek Harness Host plugin: after installing it into a profile, it appears in the plugin list and can be called in any conversation without needing to be reloaded in every session.

A few points in the package info are worth noting:

1. The name declared in package.json is dsh-plugin-book-deconstruct, and the version is 1.3.1.
2. The license is MIT, and the author field lists Kang.
3. The engines restrict dsh: ">=0.2.0", indicating it depends on a relatively new DSH runtime.
4. The repository description states: “Book Deconstruction: /dsh-plugin-book-deconstruct slash + dsh-plugin-book-deconstruct tool for full book walkthroughs.”
5. The directory site and GitHub show 1 star and 1 fork.

Core Features¶

The plugin provides two entry points:

1. Chat natural language trigger, for example: “Help me deconstruct the book [Title] by author [Author]”.
2. Slash command trigger, for example: /dsh-plugin-book-deconstruct [Book Title] | [Author].

During actual execution, the agent calls the tool named dsh-plugin-book-deconstruct, passing the book title and author to the plugin.

The output is fixed as two parts:

1. Plot Summary – Explains the whole book clearly, not a one-sentence review, and not fluff.
2. A piece of deep thought – Written as a single paragraph, not broken into multiple paragraphs or lists.

These two outputs constitute the entire behavior of the plugin; there are no extra parameter switches. The README includes two screenshots of usage methods, corresponding to the two trigger paths mentioned above.

Installation and Activation¶

Install from npm; the README recommends this command:

dsh plugin --profile desktop add dsh-plugin-book-deconstruct

Install from GitHub; the form given in the README is:

dsh plugin --profile desktop add github:OWNER/dsh-plugin-book-deconstruct

After installation, you must fully exit and reopen DeepSeek Harness for the plugin to appear in the list. This step is easily overlooked; simply reloading the window may not take effect.

Optional Configuration¶

In the profile’s cordis.patch.yml, override by line id dsh-plugin-book-deconstruct. The configuration items are provider and model:

- id: dsh-plugin-book-deconstruct
  name: dsh-plugin-book-deconstruct
  config:
    provider: ''   # Paired with model; empty uses the default model
    model: ''

The provider must be filled in paired with model; leaving both empty uses the default model.

Typical Usage¶

After installation, simply send the request directly in the conversation. The following two methods can trigger the plugin.

Method 1, natural language:

帮我解读书籍 活着 作者余华

Method 2, slash command:

/dsh-plugin-book-deconstruct 书名 | 作者

For example, entering it in the second form:

/dsh-plugin-book-deconstruct 沙丘 | 弗兰克·赫伯特

After calling, the plugin will first output the story plot of the entire book, and then output a piece of thought standing alone as a paragraph.

Applicable Scenarios and Precautions¶

The suitable scenarios are roughly of two types. One is content organization and book summarization, where you need to clearly explain the main line of a book rather than glossing over it in one sentence. The other is DSH agent development, where you need to hang the book deconstruction capability as a tool in the profile to allow different sessions to reuse the same set of interpretation logic.

Points to note:

1. The plugin runs with the permissions of the current dsh process. Before installing, you should check the source code and license, then decide whether to integrate it into a production profile.
2. Stars and forks are only 1, indicating low community verification; it is recommended to run it through locally first before promoting it.
3. There are very few configuration options, and there is no switch for output length. If you want to control the level of detail in the plot section, you need to append requirements yourself in the conversation.

Summary¶

The value of this plugin does not lie in how complex it is, but in solidifying “reading a book” into two fixed output structures: plot plus thought. Combined with profile installation and slash commands, it can be reused in any session. The installation command and configuration items come from the README, while version numbers, licenses, and other information come from package.json.

Directory page: https://www.skillhub.cn/plugins/zmz123612/dsh-plugin-book-deconstruct
GitHub: https://github.com/zmz123612/dsh-plugin-book-deconstruct