Preface

The extension approach of DeepSeek Harness (DSH) emphasizes plugin architecture, allowing assistants to call external capabilities. moononnn/DeepSeek-Harness-biaoqingbao is a community plugin that enables DSH assistants to pick stickers from a local gallery based on emotion to convey tone and sentiment.

It does not come with a gallery; the initial gallery is empty. User uploads, image recognition, tag adjustment, and like/dislike feedback will all influence subsequent image selection. Below is an introduction to its functions, installation, and precautions based on verified facts.

Introduction

This is a sticker plugin for DeepSeek Harness. The repository is moononnn/DeepSeek-Harness-biaoqingbao, licensed under MIT © moononnn & 小花.

It solves the problem: assistants can send a sticker from the user’s local gallery based on current emotion, not just text. The plugin requires users to maintain the gallery themselves, adapting to user preferences step-by-step via recognition, tagging, semantic retrieval, and feedback.

Running Requirements:

DeepSeek Harness v0.1 rc series
Node.js 20+

Support Status:

Web mode fully supported
Tools are available in headless mode, but no gallery management panel

Core Capabilities

The plugin supports custom galleries, uploads, folder import, and ZIP import. The initial gallery is empty; images must be added manually after installation.

Currently supported formats:

PNG
JPG
JPEG
GIF
WebP
BMP

Import limits:

Maximum 200 images per import
ZIP files max 50MB
Individual images are also subject to DSH image size limits

AI Image Recognition & Retrieval

The plugin supports AI image recognition, single and batch recognition, and also supports tag matching plus optional semantic vector retrieval.

After configuring the AI recognition model, content analysis model, and vector model, automatic recognition, semantic indexing, and automatic sticker matching can be performed. Changing the vector model will render old vectors incompatible and require regeneration of the index.

Feedback & Tags

The plugin supports “Like / Dislike” feedback and “Chat with Assistant” to modify tags. Recent sticker matching records are provided in a simplified form.

Auto-Selection & Assistant Settings

The plugin provides an auto-selection observer, which can be controlled via Daily / Serious sticker frequency. Setting the frequency to “No Stickers” turns off automatic analysis.

The plugin supports individual settings for each assistant.

Dialect & Style

The plugin supports 9 dialect accents and “Learn my speech”. When using “Learn my speech,” it first analyzes and edits a draft before confirmation, with the option to revert to the previous version.

Assistant Tools

Assistants can call the following tools:

express
search_stickers
list_stickers
update_sticker_tags
report_bad_match

Installation & Enabling

Install from GitHub to a DSH Web profile:

dsh plugin --profile web add github:moononnn/DeepSeek-Harness-biaoqingbao

After installation, enable the plugin and use the gallery management panel in Web mode; in headless mode, assistant tools are still available, but the gallery management panel is not present.

It is recommended to check the source code and license before installation. The plugin runs with the current dsh process permissions, reads local plugin data, and calls the model services you have configured.

Typical Usage

  1. Upload PNG, JPG, JPEG, GIF, WebP, BMP; supports multi-select, folders, and ZIP import.

  2. After configuring AI recognition, content analysis, and vector models, automatic recognition, semantic indexing, and auto-selection can be performed.

  3. Assistants can call express based on emotion, for example:

express({ emotion: '委屈' })

  1. Users can also directly say “发个委屈的表情包” (send a distressed sticker) to let the assistant find an image.

  2. Control automatic observation via Daily / Serious sticker frequency; setting it to “No Stickers” turns off automatic analysis.

  3. When using “Learn my speech,” analyze and edit the draft first, then confirm, with the option to revert to the previous version.

Data, Costs & Privacy

Local data is saved at:

${DSH_HOME}/plugin-data/biaoqingbao/

Which includes the gallery, preferences, usage logs, and vector indexes. Uninstalling the plugin will not actively delete gallery data.

Regarding model calls:

  • Recognition, content analysis, and “Learn my speech” will call configured model services, potentially incurring costs.
  • Recognition sends images to your configured vision model service.
  • Automatic observation sends relevant conversation snippets to your configured content analysis model service.
  • When using “Learn my speech,” collected user speech is sent to your configured content analysis model service.

Other notes:

  • The plugin does not connect to the author’s own server.
  • The plugin does not include the developer’s chat logs, gallery, or API keys.
  • Custom API keys are saved in the plugin’s local configuration; the settings page only displays masked placeholders.

Scenarios & Notes

Suitable for users who want to add sticker expression to DSH assistants in Web mode. The plugin does not decide the gallery for the user; it is suitable for scenarios where users are willing to upload images, maintain tags, and provide feedback.

Things to note:

  • The plugin does not include stickers by default.
  • The full experience depends on the model services configured by the user and may incur costs.
  • Regenerating the index is required after changing the vector model.
  • DeepSeek Harness is still in developer preview; plugin interfaces may evolve with versions.
  • Some features are planned for future updates, including “contextual” feedback, teaching samples, new sticker exposure / unused image exploration mechanisms, Ctrl+V paste import, drag-and-drop import, desktop paper planes, background recognition tasks, automatic duplicate removal, and specialized skipping for abnormal / encrypted files.
  • Verified information does not provide a directory page address; currently verifiable is the GitHub repository.

Links

GitHub:

https://github.com/moononnn/DeepSeek-Harness-biaoqingbao