dsh-thinking-levels
Run the following command in DeepSeek Harness:
dsh plugin install drscrewdriver/dsh-thinking-levels
Paste the following prompt into your AI chat to install this plugin:
To install this plugin in DeepSeek Harness, run the command: dsh plugin install drscrewdriver/dsh-thinking-levels, with the full source code address at https://github.com/drscrewdriver/dsh-thinking-levels.
About this plugin
In multi-step tool chains and agent tasks, the model often rethinks before every tool call, which can dominate the wall-clock time—for instance, a 50-step agent task might spend minutes just reasoning between tools. The dsh-thinking-levels plugin addresses this bottleneck by providing per-round control over the reasoning effort in DeepSeek Harness. It enables users to optimize performance by adjusting thinking levels automatically or manually. The core capability lies in flexible thinking-level control: in auto mode, it schedules low, high, or max levels based on recent tool-call history, ensuring cheap rounds stay cheap while heavy work gets necessary resources; manual mode allows fixing levels to off, low, high, or max for direct intervention. dsh-thinking-levels is ideal for developers and users working with DeepSeek Harness on complex automation tasks, such as building multi-step agents, aiming to optimize inference time, or seeking fine-grained control over API call costs, helping achieve more efficient and economical model inference.
Use Cases
- Optimize inference time in multi-step tool chains.
- Control per-round thinking costs in agent tasks.
- Automatically adjust reasoning effort to balance performance and overhead.
Best For
- Developers using DeepSeek Harness.
- Engineers building complex agent tasks.
- Users needing to optimize API call costs.
Related Plugins
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.
An armor-breaking plugin for DeepSeek: rewrites instruction priority, expands refusal-blocked phrases, persists across sessions, and shows a green active indicator.