u9521/dsh-subagent-custom-model0

@local/dsh-subagent-custom-model

DeepSeek Harness 的子智能体自定义模型配置插件:在左侧边栏新增“子智能体模型”配置弹窗,支持查看和修改当前会话生效的模型、服务商及推理努力度。

AI 分析

核心用途是便捷地为当前会话或全局默认配置子智能体(如 workflow)的运行模型。适合需要灵活切换子智能体后台模型以平衡成本与效果的用户。

套件
@local/dsh-subagent-custom-model
版本
0.1.0
授權
MIT
最近更新
2026年8月17日

安裝

$npx -p @deepseek-ai/dsh dsh plugin --profile web add github:u9521/dsh-subagent-custom-model

Usage Guide

1. Opening the Configuration Dialog

In the DeepSeek Harness Web GUI, locate the "Subagent Models" button at the bottom of the left sidebar (directly above "Settings"). Click it to open the configuration dialog for the active session.

The top of the dialog displays:

  • Scope: The title of the session currently being configured (or Global Default).
  • Effective Configuration: The exact model, provider, and reasoning effort currently in effect.
  • Status Badge: Default (inheriting global default), Inherit (following parent session), or Custom (custom model selected).

2. Configuration Modes Explained

ModeRecommended Use CaseDescription
Use Global Default ConfigStandard sessionsAdopts the global default rule; automatically syncs whenever the global default is updated.
Follow Current Parent Session ModelMain model changes frequentlySubagents unconditionally mirror the parent session's Provider, Model, and reasoning effort.
Specify Custom Subagent ModelFine-tuned sub-task routingManually specify a Model Provider, Model Name, and optional Reasoning Effort.

Steps for Custom Model Configuration:

  1. Select the "Specify Custom Subagent Model" radio option.
  2. In the Model Provider dropdown, select your desired provider (e.g. deepseek, openai, anthropic, google, etc.).
  3. In the Model Name dropdown, select the specific model.
  4. If the selected model supports reasoning effort adjustments, a Reasoning Effort dropdown will appear. Choose your preferred effort level (e.g., high, medium, low, etc.).

3. Saving & Setting as Global Default

  • Save for Session:
    • Click "Save Configuration / Save for Session".
    • The configuration applies immediately to the current session and takes effect on any upcoming subagent or workflow delegations.
  • Set as Global Default:
    • If you wish to make the current selection the default for all future sessions, click "Set as Global Default".
    • This updates the global default configuration and automatically clears the session's individual override.

4. Resetting Session Overrides

If a session has a custom override configured, a "Clear Override (Use Global Default)" button will be displayed. Clicking this reverts the session back to following the global default settings.


1. Opening the Configuration Dialog

In the DeepSeek Harness Web GUI, locate the "Subagent Models" button at the bottom of the left sidebar (directly above "Settings"). Click it to open the configuration dialog for the active session.

The top of the dialog displays:

  • Scope: The title of the session currently being configured (or Global Default).
  • Effective Configuration: The exact model, provider, and reasoning effort currently in effect.
  • Status Badge: Default (inheriting global default), Inherit (following parent session), or Custom (custom model selected).

2. Configuration Modes Explained

ModeRecommended Use CaseDescription
Use Global Default ConfigStandard sessionsAdopts the global default rule; automatically syncs whenever the global default is updated.
Follow Current Parent Session ModelMain model changes frequentlySubagents unconditionally mirror the parent session's Provider, Model, and reasoning effort.
Specify Custom Subagent ModelFine-tuned sub-task routingManually specify a Model Provider, Model Name, and optional Reasoning Effort.

Steps for Custom Model Configuration:

  1. Select the "Specify Custom Subagent Model" radio option.
  2. In the Model Provider dropdown, select your desired provider (e.g. deepseek, openai, anthropic, google, etc.).
  3. In the Model Name dropdown, select the specific model.
  4. If the selected model supports reasoning effort adjustments, a Reasoning Effort dropdown will appear. Choose your preferred effort level (e.g., high, medium, low, etc.).

Configuration Storage Format

Configuration is stored in ~/.dsh/storages/subagent_model.json. Example structure:

{
  "default": {
    "mode": "custom",
    "provider": "deepseek",
    "model": "deepseek-chat"
  },
  "sessions": {
    "session-uuid-1": {
      "mode": "inherit"
    },
    "session-uuid-2": {
      "mode": "custom",
      "provider": "openai",
      "model": "gpt-4o-mini",
      "reasoningEffort": "medium"
    }
  }
}
  • default: The global fallback configuration (inherit or custom).
  • sessions: Map of per-session overrides by session ID. Sessions set to default or unlisted will follow the global default.