Introduction

The core philosophy of DeepSeek Harness (DSH) is “everything is a plugin,” and community developers can combine different skills to create complex workflows. dsh-project-based-learning is one such skill combination specifically designed for teaching scenarios. Its design aims to solve two common problems in AI tutoring: first, knowledge transfer is often mixed with subject-specific content, making it difficult to generalize; second, instruction may focus only on theory without practice, or only show examples without requiring hands-on work. This plugin separates the teaching protocol from subject knowledge and aims to provide a general “project-based teaching coach” framework.

Core Features

This plugin mainly provides the following capabilities:

  1. Project-Based Teaching Coach: It organizes instruction around the artifact the learner wants to create, rather than first throwing a large set of knowledge points at the learner.
  2. Pedagogy and Subject Content Separation: The teaching logic is located in SKILL.md, while domain knowledge is placed in replaceable domain navigation. This means you can swap in a different domain pack (for example, from Unity Shader to Python data processing) without changing the teaching protocol.
  3. Teach First, Practice Later: For unfamiliar knowledge, it first explains the minimum necessary mechanisms, and then returns to the real project for hands-on operation after the learner understands.
  4. In-Situ Explanations: New concepts and built-in functions are explained at their first point of use, avoiding long line-by-line explanations or attempting to deliver an entire subject all at once.
  5. State Management and Migration Support: It supports recording learning progress (state.json) using five descriptive states, ranging from “not yet encountered” to “migrated to a new task.” The plugin includes a state validation script and a lossy migration script for moving from older versions to newer versions.
  6. Zero Dependencies, No Build Steps: The plugin does not rely on a complex build environment and can be run directly.
  7. Replaceable Domain Navigation: The package includes navigation examples for Unity Shader and Unity C#, which are removable or replaceable supplementary materials.

Installation and Enablement

Run the following command in the terminal to install it. If the current profile is not web, replace it with the profile name actually in use:

dsh plugin --profile web add dsh-project-based-learning

After installation is complete, DeepSeek Harness must be restarted to load the new skill. The indicator of a successful installation is that, after invoking the skill, the first line of the body displays “# Project-Based Teaching Coach 2.0”. The skill is registered at runtime and does not copy files into the local skill directory.

Typical Usage

After installing and enabling the plugin, you can start teaching directly by entering a command in a DSH session.

Teaching Mode Example:

Teaching mode: I have never learned . Please assess my level.

The plugin will automatically select an appropriate teaching lane (for example, a quick lesson, a single guided practice, or a continuous course). It first judges your proficiency gap, then teaches the minimal necessary mechanisms required right now, and then quickly helps you produce an observable result, rather than first throwing a set of diagnostic questions at you.

Applicable Scenarios and Notes

Applicable Scenarios:
- Learners who need to learn a new skill from scratch but want to start doing hands-on practice immediately.
- Teams that want their AI coach to focus on “how to teach” and “how to guide,” rather than directly writing code for them.

Notes:
- It is not a question bank or drilling tool: The domain navigation only provides rough dependencies and threshold concepts; it does not include diagnostic question banks or fixed class hours.
- It is not a code-writing tool: The plugin does not automatically authorize code changes or write the project for you.
- Third-party plugin: This project is maintained by community developer Kirisame1969 and is not an official DeepSeek plugin.
- Lossy migration: The state migration script compresses old data into now/next/later, but it does not preserve numeric proficiency levels; manual review is required.
- Environment dependency: The plugin depends on a Node.js environment.

Conclusion

dsh-project-based-learning provides an executable teaching protocol that distinguishes between “having explained” and “having mastered,” emphasizing evidence from practice. For developers who want to build automated teaching workflows, it is a practical tool for decoupling pedagogy from specific business logic. For more details and source code, see the GitHub repository.