Preface

The DeepSeek Harness (DSH) plugin ecosystem aims to solve efficiency challenges in specific scenarios through modular extensions. A single thinking model often has blind spots when facing complex decisions. The following section introduces a plugin that can turn classical thinking models into analytical tools. It simulates multiple advisors conducting cross-dimensional analysis to provide systematic decision support.

Plugin Overview

Plugin Name: @weibaohui/dsh-thinktank

Maintainer: weibaohui

Positioning: This plugin builds a library of 161 classical thinking models. After the user enters a question, the plugin invokes AI models from nine perspectives—decision-making, strategy, cognition, psychology, communication, learning, systems, product, and execution—to perform analysis. Finally, it compiles a comprehensive report containing consensus, disagreements, blind spots, and an action checklist.

Core Features

This plugin mainly provides the following capabilities:

  1. Library of 161 Classical Thinking Models
    The model library is organized into nine categories and includes first principles, inverse thinking, second-order thinking, SWOT, PESTEL, business model canvas, moat, North Star metrics, confirmation bias, anchoring effect, Dunning-Kruger effect, Feynman learning technique, deliberate practice, flow, compound interest effect, feedback loops, black swans, antifragility, JTBD, Fogg behavior model, PDCA, 5Whys, premortem, and more. Each model includes a definition, key points, applicable scenarios, and guided analytical questions.

  2. Four-Step Wizard Workflow
    The workflow consists of four steps: input question (optionally with background) → clarification prompting (optional) → model selection → confirmation and submission. After submission, the interface automatically redirects to the history page to view progress.

  3. Model Selection Mechanism
    Manual selection is supported, with two recommendation modes provided:

    • Keyword Matching: Based on a local n-gram algorithm, it performs weighted matching using scenario keywords, names, and definitions, with fast response speed.
    • AI Intelligent Recommendation: After the AI reviews the model library, it generates a recommendation list, with each recommendation accompanied by a specific reason.

  4. Dual-Mode Analysis

    • Automatic Analysis: The host side distributes selected models to multiple agent sessions in batches for parallel execution (default 3-way concurrency), then performs cross-model synthesis after completion.
    • Prompt Round-Trip: Generates a prompt containing all selected models and the output contract, which users can copy into any AI session to execute and then import the result. This serves as a fallback path when agent services are unavailable.

  5. Comprehensive Report Generation
    The report includes the overall conclusion, model consensus, differing viewpoints, blind-spot reminders, an action checklist (with the supporting models noted), and a signal overview (favorable/unfavorable/risk/neutral). Each model has a card, supports filtering by category, and can be exported as Markdown.

  6. Conclusion-Based Follow-Up Questions
    At the bottom of the report page, users can directly continue asking questions to the “Chief Advisor.” The system retains the full report context, and answers are persisted so they are not lost after restart.

Installation and Enablement

Run the following command in the terminal to install the plugin:

dsh plugin add @weibaohui/dsh-thinktank

After installation, the entry point is located by default in the left sidebar under Global panels (🦉 Think Tank). Refresh the page to use it.

Typical Usage

  1. Input Question: Enter the question to be analyzed in the wizard and add background information (such as constraints, resources, timeline, etc.).
  2. Clarification Prompting (Optional): Let the AI ask clarifying questions about the issue. The answers are automatically merged into the background.
  3. Select Models: Manually check models, or use AI intelligent recommendation. It is recommended to select 6–24 models for cross-validation.
  4. Confirm Submission: After submission, you are automatically redirected to the history page to view real-time progress.
  5. View and Export: On the report page, view the overall conclusion and per-model analysis, and export as Markdown if needed.
  6. Continue Follow-Up Questions: Based on the report conclusions, probe details further at the bottom.

Technical Notes

  • Architecture: Uses a separated architecture with a Host half (pure CJS, zero runtime dependencies) and a Client half (React function components).
  • AI Output Protocol: The end of an AI response must include the ===MIND-JSON=== marker and a JSON object; the plugin is responsible for parsing and validation.
  • Environment Requirements: The engine requires Node >= 22.15, and DSH version >= 0.1.7-rc.2.
  • Data Persistence: History records retain the most recent 50 entries and are based on the dsh_thinktank storage domain.
  • Test Coverage: Includes 25 offline tests covering model library integrity, protocol fault tolerance, route interception, and the end-to-end pipeline.

Applicable Scenarios and Cautions

This plugin is suitable for complex scenarios that require examining problems from multiple dimensions, identifying decision blind spots, or conducting systematic strategic planning. Before using it, ensure that the current DSH process has permission to invoke AI models, and check the source code and license before installation.

Conclusion

With @weibaohui/dsh-thinktank, static thinking models can be transformed into dynamic decision-support tools. It does not provide a single answer; instead, it uses cross-validation across multiple models to help developers organize their thinking, identify blind spots, and create action checklists. For more details, refer to the GitHub repository.