In the plugin-based architecture of DeepSeek Harness (DSH), model inference typically requires developers to manually configure external interfaces. For the specific scenario of dish recognition, directly calling a vision model to obtain names and probabilities is repetitive work. The xby-dish plugin encapsulates this workflow and provides a standardized input and output mechanism.

This is a model inference plugin under xby-skill. It provides dish recognition capabilities, taking an image as input and outputting possible dish names and probabilities.

Core Features

The plugin provides three recognition methods, each corresponding to a different image input format:

  • dish_recognition: Accepts an image file link.
  • dish_recognition_for_data_base64: Accepts a BASE64-encoded image file.
  • dish_recognition_for_data_file: Accepts a file path to the image file.

In addition, the plugin supports setting the API key in chat. The key is automatically persisted and remains valid after a restart.

Installation

Direct installation from GitHub is recommended.

# 格式: dsh plugin --profile <profile> add github:<owner>/<repo>
dsh plugin --profile web add github:xby_skill/xby-dish

Configuration and Usage

Before use, register on the Xiaobenyang official website to obtain an API key.

In the DSH interactive environment, simply call the functions mentioned above. For example, provide an image link and call dish_recognition; the plugin returns the recognition results and probabilities.

Notes

The plugin runs with the permissions of the current DSH process. Before installing, it is recommended to review its source code and license (MIT).