Introduction

In the plugin ecosystem of DeepSeek Harness (DSH), providing immediate feedback to models is key to improving interaction quality. Traditional feedback mechanisms often lack fun and instant incentives. This article introduces xby-cookie-reward, a Model Context Protocol server that rewards LLMs through gamified self-reflection.

Plugin Overview

This plugin is maintained by xby-skill and is licensed under the MIT License. It is a Model Context Protocol server whose core logic is to let the LLM evaluate its own output (self-reflection) in order to earn rewards.

Core Features

The plugin provides the following features:

  • set_xby_apikey: Sets an API key in the chat, with automatic persistence, so it remains valid after restart.
  • self_reflect_and_reward: The core feature, allowing the LLM to evaluate its own work and earn Cookies.
  • give_cookie: A legacy method for giving Cookies to the LLM (prefer self_reflect_and_reward when possible).
  • check_cookies: Views the current number of accumulated Cookies.
  • reset_cookies: Resets the Cookie counter, useful for testing.
  • add_cookies_to_jar: A user operation for adding Cookies to the reward jar (🚨 USER ONLY: for human use only; LLMs should not call this tool).
  • cookie_jar_status: Views the status of the reward jar, including capacity and remaining space.

Installation and Enablement

Use the official command to install it:

dsh plugin --profile web add github:xby_skill/xby-cookie-reward

Important Notes

Before use, you must register on the official Xiaobenyang website and obtain an API key. The plugin relies on this API for reward processing. Additionally, be sure to note the usage restriction on the add_cookies_to_jar tool: it is intended for human users only, and LLMs cannot and should not add Cookies to their own cookie jar.

Conclusion

Through a simple self-evaluation mechanism, this plugin provides an additional feedback loop for LLMs. For more details, see GitHub.