DeepSeek Harness (DSH) defaults to using a dropdown menu to select the model’s reasoning level. The dsh-effort-slider plugin converts it into a sci-fi-style energy slider, supporting drag, click, and arrow-key interaction. In addition to the visual upgrade, it adds an Ultra level, Lightning orchestration mode, and a subagent token-rate readout.

Core Features

Ultra Level

The 6th position on the slider. When selected, the plugin locks the reasoning level to the model’s highest actual level (MAX) and injects an English policy named “Maximum Strictness” into the current session. No matter how the UI is interacted with, as long as Ultra is active, the effective reasoning level is always MAX.

Lightning Mode

Disabled by default. When enabled, the plugin injects a policy that the parent agent only orchestrates and all work is delegated to subagents. This is a prompt-level control and will significantly increase token consumption (about 15 times). The policy includes anti-recursion instructions, and when disabled, an OFF notice is appended.

Fleet tok/s

Displays in real time the token output rate (including generation and input, excluding cache reads) of all subagents in the currently tracked session. The interface shows full integers (e.g., 12,840) and hides the entire block when no data is available.

Installation and Configuration

dsh plugin add github:yuhub233/dsh-effort-slider

Manual Installation

Copy the repository directory to the DSH plugin directory. The default Windows path is:
C:\Users\<you>\.dsh\plugins\

Optional: Install Accompanying Skills

The plugin package includes 4 accompanying skills (dispatching-parallel-agents / subagent-driven-development / verification-before-completion / systematic-debugging). They must be manually copied to ~/.dsh/skills/ to be loaded into sessions.

Usage Instructions

UI Interaction

The slider supports dragging, clicking tick marks, and adjusting with keyboard arrow keys. The level names dynamically follow the model’s actual levels, and when a model has no reasoning levels, the control automatically hides.

Using Ultra

Drag to the Ultra position to activate. At that time, the plugin sets the reasoning level to MAX according to the model configuration and injects a strict acceptance policy.

Lightning Mode

Click the lightning toggle in the upper-right corner of the panel to switch it. The policy text changes before and after switching to ensure the orchestration contract takes effect.

Technical Details and Limitations

Environment Dependencies

  • DSH Desktop version must be ≥ 2.0.9.
  • Node.js version must be ≥ 22 (used only for building; zero runtime dependencies).

Implementation Principles

The plugin is divided into client and host halves. The client registers through window.__ModuleLoader__.load and is responsible for UI rendering and interaction; the host injects policy text through the agent/pre-step waterfall, and state is stored in <DSH_HOME>/storages/effort-slider.json.

Known Limitations

  • Nebula skin: Taken offline per the author’s request; the code remains but is unavailable.
  • Token statistics: Only counts descendant subagents and excludes cache reads.
  • Package name consistency: dsh-effort-slider is simultaneously the registration ID, endpoint prefix, and JSON property name, so changes must be synchronized in multiple places.