dsh-terse-mode: Terse Mode for Controlling Thinking, Search, and Subagents

Preface¶

DeepSeek models, when handling simple questions, sometimes engage in unnecessary in-depth reasoning, frequently invoke web search, or delegate tasks to subagents, resulting in slower responses and additional token consumption. The dsh-terse-mode plugin uses controls at the top of the chat page and in the settings page to convert reasoning intensity, web search, and subagent usage into four adjustable levels and three independent switches, enabling immediate control over model behavior.

Plugin Introduction¶

This is a DeepSeek Harness plugin maintained by cspyz. Its core value is that it does not set an output limit, allowing the model to output fully, while using instructions to reduce redundancy. The plugin is implemented in pure JavaScript, has zero runtime dependencies, and stores data locally.

Core Features¶

• Four-Level Intensity Adjustment: Provides four levels — “off / light / standard / strong” — for quickly switching overall behavior modes.
• Independent Control Switches:
• Reasoning Effort: Supports off / low / high / max / auto.
• Web Search: Supports “allow / deny / follow” levels.
• Subagents: Supports “allow / deny”.
• UI Entry: Controls are available at the top of the chat page and on the settings page; changes take effect immediately.
• Technical Features: Zero runtime dependencies and pure JavaScript; incorrect configuration will not cause DSH startup to fail; applies only to DeepSeek models.

Installation¶

Use the official CLI to install the plugin. After installation, a restart of DSH is required on first startup.

dsh plugin --profile web add github:cspyz/dsh-terse-mode
dsh plugin --profile desktop add github:cspyz/dsh-terse-mode

Usage¶

• Top Controls: The top of the chat page displays the current level, subagent, and web search switches. Click a level to switch it, and click a switch to enable or disable it individually. Hover over items to view descriptions, and click ? to expand the full configuration panel.
• Settings Page: Find the “Terse Mode” entry on the settings page to perform more detailed configuration and view descriptions.
• Immediate Effect: After changing a level or switch, DSH does not need to be restarted; the next request immediately applies the new settings.

Configuration Notes¶

The plugin’s default configuration meets most needs. If customization is required, override id: terse-mode in the profile’s cordis.patch.yml.

- id: terse-mode
  config:
    level: standard          # off | light | standard | strong
    reasoningEffort: auto    # auto | none | off | low | high | max
    search: auto             # auto | allow | deny
    subagents: allow         # allow | deny
    applyToAllModels: false  # true = 对非 DeepSeek 模型也压强度

Notes¶

• Local Storage: Settings preferences are stored in local files and persist after restart.
• Profile Independence: Installation only affects the current profile and does not affect other plugins.
• Model Limitation: Currently, the plugin only takes effect for DeepSeek models and will not force-send this parameter to models that do not support it.
• Safety: Configuration errors will automatically fall back to safe defaults and will not cause the program to crash.

Conclusion¶

dsh-terse-mode solves the difficulty of controlling model behavior with simple UI controls, making it suitable for users who want to finely manage reasoning costs and response style.