DeepSeek Harness (DSH) uses a plugin-based architecture. When using the DSH Agent, developers sometimes encounter issues where the output is too verbose, steps are unclear, and actionable instructions are hard to locate quickly. The dsh-i-have-adhd plugin aims to address this pain point by using specific rules to reshape the output style, making it better suited for quick execution.

Plugin Overview

This plugin is named dsh-i-have-adhd, maintained by GongYuanCaiJi, and released under the MIT open-source license. It is ported from the upstream project i-have-adhd and adapted to the DeepSeek Harness plugin interface.

Core Features

  1. Action first: Prioritize output commands, paths, or code snippets; put explanatory text later.
  2. Number multi-step tasks: Break tasks into numbered actions with the minimum number of steps.
  3. End with one concrete next step: Provide an action completable within two minutes at the end of the response.
  4. Suppress tangents: Finish the current issue first; handle secondary topics separately.
  5. Restate state each turn: Clarify current progress, avoiding making the reader track it themselves.
  6. Specific time estimates: Provide minute-level time estimates.
  7. Make completion visible: Directly show currently runnable content, rather than burying it in a summary.
  8. Full 10 rules: Includes the complete rule set.

Installation and Activation

Run the following command in the terminal to install:

dsh plugin --profile <你的 profile> add github:GongYuanCaiJi/dsh-i-have-adhd

After installation, enter the following command in a DSH session to enable ADHD mode:

/i-have-adhd

Usage

  • Enable: Enter /i-have-adhd; the rules will be injected into the current session.
  • Disable: Enter stop adhd mode or normal mode.

The rules apply to each subsequent turn in the session and remain effective even if the topic changes.

Notes

  • The DSH side does not have a SessionStart mechanism equivalent to the upstream ~/.claude/.i-have-adhd-always; this plugin only provides on-demand activation.
  • The upstream scripts/check_pi_extension.py and pi-load-check.yml are verification scripts for the pi context; they cannot run because the package name was changed to dsh-i-have-adhd, so they are kept as-is without modification.

Summary

After installation and activation, the Agent’s output will be more focused on actions and progress, making it suitable for scenarios that require rapid task execution. See GitHub for the project.