Preface¶

When developing with DeepSeek Harness (DSH), the main session often fills up with commands, errors, and temporary conclusions, but truly memorable concepts are rarely organized separately. Common practices include opening separate notes or relying on memory, which easily disconnects from the current work context and makes reuse in subsequent sessions difficult.

This introduces dsh-explain (maintainer yuezengwu). It is a learning mode plugin for DSH that extracts concepts from completed work rounds, generates structured explanation cards, and writes them to a cross-session global learning thread. The main agent remains unaffected; Explain uses independent model calls, scheduler, context, and a local SQLite database.

What This Is¶

dsh-explain targets DSH 0.1.0-rc.8 and is categorized under “Memory” in the SkillHub Plugin Directory. The plugin’s current version is v0.1.0, licensed under MIT.

The core problem it solves is: how to transform concepts that arise during everyday DSH work into reviewable, feedback-enabled, and cross-session-continuable private learning records, instead of scattering them across various conversations.

Core Features¶

Explanation and Learning Cards by Source¶

Explain supports multiple entry points, all generating explanations based on the current or selected source material:

Each learning card answers three questions: What is it? (the concept), Why does it matter? (its practical significance in work), and What is the common pitfall? (common misconceptions). You can choose Got it to close the card or Not yet to request a different explanation; even if the source session is later deleted, existing explanations can still be rephrased.

Cross-Session Global Learning Thread¶

Each $DSH_HOME has only one Explain learning thread. Various work sessions contribute materials, but resuming and forking do not duplicate learning state:

• Each source session has at most one explanation pending feedback
• All work sessions display the same global history in DSH’s native Learning tab
• The global scheduler serially processes manual explanations, auto-evaluations, rephrases, and compressions
• The default auto-evaluation budget is 50 requests within a rolling 24-hour window, preserved across restarts
• A private ExplainContext tracks explanation preferences, knowledge level, and learning progress

Local-First and Context Isolation¶

When structured observations or closed explanations are pending, if there is no Explain activity within 30 minutes, or if a request will exceed 50% of the selected model’s context window, the auxiliary history is automatically compressed.

Diagnosable Settings Interface¶

Open Settings → Learning in DSH Web to select the auxiliary provider and model, enable learning mode, and save. Explain only observes new top-level rounds completed after enabling; it does not scan existing history. Within the Composer, you can use /explain on, /explain off, /explain status to control or check the runtime status.

Installation and Enablement¶

Explain is currently compatible with DSH 0.1.0-rc.8. Installation commands:

npx @deepseek-ai/dsh@0.1.0-rc.8 plugin --profile web add github:yuezengwu/dsh-explain
npx @deepseek-ai/dsh@0.1.0-rc.8 web

Git-hosted plugins build during installation. If pnpm requests build approval, you need to add the printed dsh-explain entry to the profile’s pnpm-workspace.yaml and then rerun the installation command.

After starting the web interface, go to Settings → Learning to configure the auxiliary model and enable learning mode. With these steps, Explain begins listening to subsequent completed work rounds.

Typical Usage¶

Learning from an Answer¶

1. Complete a meaningful main agent response in DSH Web.
2. Select Learn from this answer, or use Explain selected text after selecting text.
3. Review and edit the generated /explain draft, then submit after confirmation.
4. View the generated learning card in the Learning tab and choose Got it or Not yet.

Manually Requesting an Explanation¶

Enter directly in the Composer:

/explain <your question or concept>

The current session content participates in generation as a bounded source context. Use /explain status to check the current status.

Controlling Learning Mode¶

/explain on
/explain off
/explain status

You can toggle or check Explain’s runtime without leaving the Composer.

Use Cases and Considerations¶

Who is this for: Users who长期 use DSH for development and want to consolidate concepts from sessions into reviewable learning records; scenarios where the main agent needs to remain independent and the learning logic is handled by an auxiliary model.

Compatibility: The plugin follows the DSH public API line (currently 0.1.0-rc.8) and does not maintain compatibility layers for earlier private-preview packages. The repository includes 64 unit tests, 4 assembled DSH Web acceptance scenarios, and 3 Explain-specific shortcut acceptance scenarios; see docs/ACCEPTANCE.md for the detailed matrix.

Pre-installation note: The plugin runs with the current DSH process permissions and reads/writes SQLite and settings files under $DSH_HOME. Before installation, review the source repository and MIT license to confirm that the data storage location and model calling methods meet your expectations. SkillHub is a community plugin directory and has no official affiliation with DeepSeek / High-Flyer.

Conclusion¶

dsh-explain transforms concepts worth remembering from everyday DSH work into locally stored, cross-session-shared learning threads, keeping the main agent path clean. If you’re looking for a “memory” type plugin in the DSH ecosystem, you can learn more from the directory page or GitHub:

• Directory page: https://www.skillhub.cn/plugins/yuezengwu/dsh-explain
• GitHub: https://github.com/yuezengwu/dsh-explain