DeepSeek Harness (DSH) uses a plugin-based architecture. Output style control typically relies on in-session skills or manual injection. In scenarios that require consistently maintaining a specific output structure (such as ADHD-friendly mode), frequent manual maintenance can be costly. The dsh-i-have-adhd plugin is introduced below; it injects a global systemPrompt via the Host-only method, making it persistently effective.

Plugin Positioning

This is a DeepSeek Harness client plugin maintained by NecromanAlbert and released under the MIT license. It provides a persistent ADHD-friendly output system prompt, adapted from the rules in ayghri/i-have-adhd.

Core Features

  1. Host-only architecture: Runs only as a Host and does not depend on client-side model loading.
  2. Global injection: Injects systemPrompt for all sessions under the configuration file.
  3. Persistent enablement: Takes effect after installation and does not require explicit model-side loading.
  4. Rule adaptation: Based on an MIT-licensed ADHD output rule library.

Installation and Enablement

Install it using the official command:

dsh plugin --profile desktop add github:NecromanAlbert/dsh-i-have-adhd

After installation, restart Desktop. New sessions take effect immediately, while existing sessions take effect on the next model request.

Typical Usage

Add the plugin to a composition row in the DSH configuration file and adjust parameters through the config field:

- id: dsh-i-have-adhd
  name: dsh-i-have-adhd
  config:
    enabled: true
    order: 8
    extra: ''
  • enabled: Set to false to disable that section.
  • order: Controls the injection order of systemPrompt.
  • extra: Used to append specific lines.

Applicable Scenarios and Notes

  • Applicable scenarios: Developers who need to consistently maintain an ADHD-friendly output style (for example, action-first, numbered steps, and restating task progress).
  • Notes:
    1. Non-official product: It is a community plugin, not an official DeepSeek AI product.
    2. Non-Skill: It is not a skill that the model must load, nor a housekeeper identity fiber.
    3. Technical terms: The output content keeps technical terms accurate and does not use the original language.
    4. Environment requirements: A Node.js 22+ environment is required.

By using persistent injection, this plugin reduces the maintenance cost of ADHD-friendly output style. For more details, visit GitHub or the community catalog.