Introduction

In the session tabs of DSH Web, you typically see chat, tool calls, and session logs. dsh-auto-chess places a simplified Auto Chess board into this tab bar: you can play as the Blue team against an AI, or watch two AIs play against each other. After each AI decision, you can also view the full reasoning text in the Thought Log panel, making it suitable for observing how the model gives actions under fixed rules.

What is it

omdsh-dev/dsh-auto-chess is a DSH Web plugin maintained by omdsh-dev under the MIT license. It provides a simplified Auto Chess interface, supporting Player vs AI or AI vs AI modes.

Core Features

Game Modes

  • Player vs AI: The user plays as the Blue team.
  • AI vs AI: The Blue and Red teams are operated by AIs respectively; the user watches as an observer.

Model and Thinking Settings

  • The Blue and Red teams can select models from the model directory and configure them independently.
  • Supports Off, High, and Max thinking levels.
  • If the model does not support the selected level, it automatically falls back to the default level.

Decision Limits

  • Default timeout for a single AI decision is 300000ms.
  • Default output token limit is 32000.
  • These two items are determined by deployment configuration; the interface no longer provides modification options.

Prompts and Reasoning Logs

  • The Blue and Red teams can edit prompts separately and restore defaults with one click.
  • After each AI decision, the full reasoning text can be viewed in the left/right Thought Log panels.

Failures and Retries

  • When the AI makes an illegal move, it will prompt the reason and offer a retry.
  • Temporary streaming failures will be automatically retried within the budgeted attempts.

Session Switching

  • Switching sessions or closing session tabs will not interrupt AI thinking and battle processes.

Installation and Activation

First build in the plugin directory, then install to the web profile:

pnpm run build
dsh plugin --profile web add /Users/yejiming/Desktop/OpenSource/dsh-auto-chess

After installation, check the configuration layer:

dsh --profile web --dump-config   # auto-chess layer should appear in the output

Start or restart the Web interface:

dsh web                            # After restart, the "Auto Chess" tab appears in the session tabs

The Web process needs to be restarted for the new dshClient package to be scanned and the client tab to be injected.

Remove the plugin:

dsh plugin --profile web remove @deepseek-ai/dsh-auto-chess

Typical Usage

The following are verified development and validation commands.

Build:

pnpm run build

Type checking and testing:

pnpm run typecheck
pnpm test
node tests/e2e/gui-round.spec.mjs

Here, node tests/e2e/gui-round.spec.mjs is used for end-to-end testing.

Use Cases and Notes

It is suitable for viewing Auto Chess decisions and reasoning texts under different models, prompts, and thinking levels in DSH Web. It can also let two AIs play against each other to observe the output differences on both sides.

Notes before use:

  • Game state only exists in page memory; refreshing or closing the browser page will reset the game.
  • Game rules are a simplified version: no equipment system, and combat is simulated with low precision.
  • Auxiliary requests are independent of any agent conversations and are never written to session logs.
  • The plugin runs with the permissions of the current dsh process; check the source code and license before installing.
  • The license is MIT; source code is available at the GitHub repository.

Conclusion

The value of dsh-auto-chess lies in placing model decisions into a visible, comparable, and retryable Auto Chess scenario: victory and defeat happen on the board, while the reasoning text remains in the log panel.

Directory Page: https://www.skillhub.cn/plugins/omdsh-dev/dsh-auto-chess
GitHub: https://github.com/omdsh-dev/dsh-auto-chess