Introduction

The plugin ecosystem of DeepSeek Harness (DSH) allows developers to extend functionality through prompt injection. When building chatbots, plain-text responses can sometimes feel too rigid. The dsh-kaomoji plugin injects rules into the system prompt so that the model can automatically copy Japanese kaomoji based on the emotional tone of the conversation, enriching the tone of responses without modifying code or adding extra API calls.

Plugin Overview

This is a DSH plugin maintained by TianJie52009. It uses a pure-prompt approach: it does not truncate the stream, does not make additional model calls, and is friendly to KV caching. The word library is carefully selected from the Japanese kaomoji site kaomojiya.org and includes 12 built-in emotion categories.

Core Features

  • Pure-prompt approach: Uses DSH’s systemPrompt service to inject rules before the model generates output, without truncating the stream and without adding extra API calls.
  • Whitelist protection against garbled output: The model can only copy kaomoji from the built-in word library and will not fabricate garbled characters from memory.
  • Emotion categories: Includes 12 built-in emotion categories (happy, heart-fluttering, sad, crying hard, angry, surprised, confused, shy, playful, encouraging, thankful, apologetic).
  • Frequency and placement control:
    • Mode: Supports auto (smart triggering, default), frequent (include in every response), and off (disabled).
    • Placement: Supports inline (after the sentence that best matches the emotion, default) or end (fixed at the end of the response).
  • Visual configuration: Provides a Web configuration card under “Settings → General Settings”; changes take effect immediately without a restart.
  • Zero dependencies: Uses only DSH’s built-in services and the Node standard library, with no third-party runtime dependencies.

Installation and Activation

Install using the dsh CLI:

dsh plugin --profile web add dsh-kaomoji

After installation, restart the Web Host or wait for hot reload to take effect. A “Kaomoji” configuration card will appear at the bottom of the Web “Settings → General Settings” page.

Configuration and Usage

The plugin works with its default configuration (mode: auto). Configuration is divided into two layers: the user layer (General Settings card) and the deployment default layer (cordis.patch.yml).

The user settings file is located at ~/.dsh/dsh-kaomoji.json. Configuration can be modified remotely via the POST /dsh-kaomoji-settings endpoint.

The following is a configuration example for cordis.patch.yml:

- id: dsh-kaomoji
  config:
    mode: frequent     # off | auto | frequent
    placement: end     # inline | end
    customPrompt: "正式场景克制一点，优先使用「感谢 / 鼓励」类颜文字"

Example of the effect in frequent + end mode:

当然可以，我帮你把正则改成忽略大小写再跑一次测试。
测试已经全绿了 (๑•̀ㅂ•́)و✧

Compatibility and Notes

  • Version compatibility: Compatible with dsh 0.1.x and 0.2.x.
  • Coexistence: Can coexist with existing plugins such as dsh-emoji; the prompt segment name is different (dsh-kaomoji:guidance @ 176).
  • Code source: The plugin code and documentation were generated automatically by OpenAI Codex and have not been manually reviewed line by line. It is recommended to read the source code before use.
  • Security period: Within 24 hours of npm publication, if the target pnpm setup has enabled new package security periods, direct installation by bare name may not be possible. It is recommended to use direct installation from GitHub.

Summary

dsh-kaomoji provides a low-cost way to increase the friendliness of conversations. By restricting the model to copying characters only from a whitelist, it ensures reliable display while avoiding additional resource consumption.