Introduction

DeepSeek Harness provides the foundational framework for building agents, but lacks concrete hands-on simulation scenarios. RiverMind is a DSH-based plugin that provides a two-player Texas Hold’em training ground. It makes decisions through an independent Iris Agent, and uses backend validation and persistent memory to simulate a real game environment.

What Is This?

This is a DeepSeek Harness-based Texas Hold’em AI sparring partner and Agent project. The maintainer is jiangzeyuan, the category is Fun Dress-Up, and the open-source license is MIT. The plugin provides a two-player table, an independent Iris Agent, backend validation mechanisms, and traceable decision logs. The rules-based sparring partner does not call a model; when using a model, it reads the model configured in DSH and does not store an API Key separately.

Installation and Enablement

Node.js 22+ and an installed DeepSeek Harness are required. The project is currently developed against DSH 0.2.0-rc.2.

  1. Open DeepSeek Harness to initialize the desktop profile, then fully quit the application.
  2. Run the following command in a terminal:
dsh plugin --profile desktop add dsh-rivermind
  1. Reopen DSH, select RiverMind Texas Hold’em Training Ground in the left panel, and click Start the First Hand.

Core Features

  • Two-player table: supports alternating dealer button, small blind/big blind, fold, check, call, raise, and showdown settlement.
  • Independent Iris Agent: creates a new session, does not inherit other sessions, and restricts tool permissions.
  • Four poker tools: get_observation, recall_oppo.
  • Backend validation: validates the actor, hand number, state version, and legal amounts.

Usage Examples

  • Select RiverMind Texas Hold’em Training Ground in the left panel and click Start the First Hand.
  • Click Restart Training.
  • Click the small document icon on the right side of Trace and choose Copy JSON.

Notes

  • Uninstalling does not clean up training memory.
  • Decision processing is terminated if the model output quota is exhausted, the model request times out, or thinking times out.
  • The rules-based sparring partner does not call a model.
  • The plugin uses the model configured in DSH and does not store an API Key separately.

Ecosystem Context

The DSH philosophy is “everything is a plugin.” The community directory is an independent site and has no official affiliation with DeepSeek / High-Flyer.

Links

  • GitHub: https://github.com/jiangzeyuan/dsh-rivermind
  • Directory: https://www.skillhub.cn/plugins/jiangzeyuan/dsh-rivermind