AI Agent Hub
Back to plugins
🤖

mv-superds

Model Inference Updated 2026.08.22

Run the following command in DeepSeek Harness:

dsh plugin install aitingtingya/mv-superds

Paste the following prompt into your AI chat to install this plugin:

To install this plugin in DeepSeek Harness, please run the command: dsh plugin install aitingtingya/mv-superds. Source: https://github.com/aitingtingya/mv-superds

About this plugin

mv-superds is a minimalist DSH prompt-activation plugin designed to address the core contradiction between system prompt bloat and tool availability in model inference. In complex agent workflows, exposing too many tools too early can often interfere with the model's reasoning path, causing distraction, while hiding tools limits its ability to call upon necessary capabilities at critical moments. This plugin achieves precise control over tool visibility in the first round of reasoning, allowing developers to maximize the use of external capabilities while keeping the system prompt clean.

Its core highlight lies in its unique "Weak" mode and lifecycle management. Unlike conventional approaches, the Weak mode allows external tools to be genuinely available in the first round, but their metadata is stored as a detached copy in the transient context, never polluting the system prompt. The plugin also supports Normal and Force modes, tailoring tool lifecycles to different iteration cycles (epoch or DSH turn). This design is particularly well-suited for developers and researchers who need to customize behavior for specific reasoning stages, such as Standard, PTC, or Creative modes.

Use Cases

  • Need to precisely control the scope of tool exposure in the early stages of model inference.
  • Want to retain tool capabilities while keeping the system prompt clean.
  • Need to flexibly switch tool lifecycle strategies based on task cycles (epoch or turn).

Best For

  • Agent architects focused on the DSH ecosystem.
  • Researchers who need fine-grained management of tool invocation strategies.
  • Agent developers seeking best practices in prompt engineering.