Introduction¶

The core idea of DeepSeek Harness (DSH) is “everything is a plugin.” For developers or content operators, building a complete content production pipeline (topic selection, titles, copywriting, scheduling) usually involves multiple steps and tools. The dsh-content-lab plugin integrates this workflow within the DSH environment and directly generates required content through local LLM calls.

Plugin Overview¶

dsh-content-lab (Content Lab) is a DeepSeek Harness plugin designed to provide tools for topic selection, title generation, copywriting, and content scheduling for self-media, AI side businesses, or content marketing. The plugin is maintained by user hanyi7867069-create and follows the MIT license. Its core value lies in zero external dependencies: all tools are registered internally within the plugin using native JSON-Schema and call the LLM directly for structured output.

Core Features¶

The plugin provides four main tools, covering the entire content production process:

Generate viral topics (content_topic_generator)
Specify a field and platform to generate a list of topic ideas, including angles, hooks, and rationale.
Rewrite titles (content_title_rewriter)
Supports multiple styles: suspense, numbers, emotion, contrast, and questions.
Generate complete copy (content_post_writer)
Supports generating content for platforms such as Xiaohongshu, WeChat Official Accounts, Douyin, and Bilibili.
Generate content calendar (content_calendar)
Generate a content schedule plan for N days.

Installation and Configuration¶

Installing the plugin requires DSH’s plugin command. After installation, it defaults to using the deepseek-official provider and the deepseek-v4-flash model.

The installation command is:

dsh plugin --profile web add dsh-content-lab

In the cordis.patch.yml file of the web profile, you can override the configuration using id: content-lab:

- id: content-lab
  name: dsh-content-lab
  config:
    provider: deepseek-official
    model: deepseek-v4-flash

Each tool function also supports a separate model parameter for overriding.

Typical Usage¶

The following section describes how to call each tool.

Generate Topics¶

Specify a field and platform to obtain structured topic plans.

Rewrite Titles¶

Specify the original title and rewriting style to generate optimized titles.

Generate Copy¶

Generate complete article content based on the platform type.

Generate a Calendar¶

Set the number of days to generate a content publication plan.

Notes¶

Model output stability: LLMs may sometimes output malformed JSON. The plugin includes built-in parseJsonLoose logic that returns the text as-is when parsing fails, instead of throwing an error directly.
Runtime permissions: The plugin runs with the permissions of the current DSH process. Please make sure to review the source code and license (MIT) before installation.
Technical architecture: The plugin does not depend on any @deepseek-ai/* external packages. It uses native ctx.llm.stream() and message construction, making it suitable for scenarios requiring deep customization or lightweight usage.

Conclusion¶

dsh-content-lab standardizes the content production pipeline and is suitable for users who want to use DeepSeek models for large-scale content planning and generation. Through structured output and loose parsing mechanisms, it lowers the barrier to content generation.

Plugin directory: dsh-content-lab
GitHub repository: dsh-content-lab

DeepSeek Harness Plugin: dsh-content-lab (Content Lab)