Introduction¶

When using DeepSeek Harness (DSH) for agent development or technical learning, it is a common pain point to establish a structured diagnostic path, record learning evidence, and follow up continuously. Plain conversational AI often lacks this closed loop. The dsh-diagnostic-tutor plugin separates the “teaching brain” from the “learning runtime” and provides a diagnostic-first interactive environment, enabling learners to clarify requirements, obtain evidence, and proceed step by step under tutor guidance.

Plugin Introduction¶

dsh-diagnostic-tutor is a workflow plugin for DeepSeek Harness maintained by SenmuuuuW. It is a diagnostic-first AI learning runtime whose core value lies in:
* Separating teaching logic from runtime logic.
* Visualizing the learning process through an interactive map and side-docked UI.
* Requiring learners to operate step by step; the runtime will not skip automatically.

Core Features¶

Based on verified facts, the plugin has the following capabilities:
* Diagnostic-first AI learning runtime: Uses diagnosis as the core driver of the learning process.
* Interactive learning map: Builds knowledge paths step by step with clear node states.
* Tutor-authored structured courses: Provides structured teaching content with checkpoints.
* Persistent learner state: Saves learning progress, diagnostic maps, and evidence records.
* Evidence recording: All state changes are supported by evidence.
* Tutor handoff to the next step: The tutor decides the next step and provides reasons.
* Side-docked UI surface: Map, course, and chat are in the same interface, eliminating frequent switching.
* Runtime does not auto-advance: Progress must be continued by the learner actively pressing Continue.

Installation and Enablement¶

The plugin uses prebuilt output and requires no cloning or building.

1. Open a terminal or command line.
2. Run the following command to install the plugin (requires the Web profile):

    dsh plugin --profile web add dsh-diagnostic-tutor

3. After installation, restart or refresh DeepSeek Harness to load the plugin.

Typical Usage¶

1. Define the goal: Use your own words in the chat to define the learning objective (for example, “I want to learn machine learning”).
2. Provide evidence: Answer the tutor’s check questions directly in the chat to provide evidence for the diagnostic map.
3. Advance the step: Follow the next step indicated by the tutor and press the Continue button to move to the next stage.
4. Resume learning: Use the resume feature to continue a previous scenario, avoiding long waits caused by cold starts.

Applicable Scenarios and Cautions¶

The plugin is suitable for scenarios that require a systematic and structured learning path. When using it, note the following:

• Math rendering: Mathematical formulas currently receive styling only, and LaTeX is not actually rendered, which may affect the reading experience in some STEM courses.
• Agent dependency: The plugin depends on an active Agent and a valid Agent scope. If the Agent is not activated or the scope is invalid, the panel may display cannot tell and remain silent.
• Storage location: All learning state is stored in a single document <dsh-home>/storages/udt. This file is readable and exportable, but a corrupted format will cause the plugin to fail to load.
• Progress control: The runtime will never advance progress automatically; the learner must manually confirm the next step. The tutor may not converge and may repeatedly evaluate the same node as more-practice.

Summary¶

dsh-diagnostic-tutor provides DeepSeek Harness with a structured, evidence-driven learning environment that does not automatically skip steps. It combines the diagnostic map, course, and chat through a sidebar UI, making it suitable for scenarios that require a rigorous learning process.

• Directory page: dsh-diagnostic-tutor - SkillHub
• Source repository: SenmuuuuW/dsh-diagnostic-tutor