DeepSeek Harness (DSH) provides plugin-based extensibility. When running model inference tests, you may sometimes need to verify how a model behaves under extremely high load, or simply want to consume extra tokens. This requires a tool that can force the model to perform heavy computation without outputting actual content.

What This Is

This is a DSH plugin maintained by HuaJi2077. It provides an “Empty Fort Strategy” mode: the model will engage in multiple rounds of thinking and tool calls, but ultimately will not answer the question and will only output a newline. This is a purely entertainment-oriented tool.

Core Features

  • Empty Fort Strategy mode: Once activated, the model enters a loop of thinking and tool calls without generating a valid answer.
  • Custom configuration: Supports adjusting thinkingRounds (number of thinking rounds), toolCalls (number of tool calls), and outputFinal (whether to output the result).

Installation and Enablement

  1. Install using the CLI:
    pnpm dsh plugin --profile web add github:HuaJi2077/empty-fort-strategy
  1. Manual installation (clone the repository):
    git clone https://github.com/HuaJi2077/empty-fort-strategy.git
    cp -r empty-fort-strategy/preset ~/.dsh/.agent-presets/empty-fort-strategy
  1. Restart dsh (pnpm dsh web), start a new session, and select “Empty Fort Strategy mode” in the mode selector.

Configuration

Behavior parameters are in the preset/scenario.yml file:

  • thinkingRounds: Number of thinking rounds, minimum 2. If it is less than toolCalls + 1, the value is automatically increased.
  • toolCalls: Number of tool calls, minimum 1.
  • outputFinal: Whether to output the result, default is false (only outputs a newline).

After modifying the configuration, you need to copy the directory again and restart dsh. The configuration is read when the plugin is mounted.

Uninstallation

  1. Delete the plugin directory ~/.dsh/.agent-presets/empty-fort-strategy.
  2. Remove it from the plugin list:
    pnpm dsh plugin --profile web remove empty-fort-strategy

This plugin is mainly used to consume tokens and test the limits of model thinking; it does not change the actual output. Check the directory or GitHub for more information.