DSH HUB
HomePlugin StorePlugin PacksCommunityRankingsResourcesPublish Guide
Plugin source
Back to catalog

QuinnKevin /

QuinnKevin/dsh-concept-explainer

Verified

Explain an unfamiliar concept selected in a DeepSeek Harness conversation in an isolated right-sidebar tab; the conversation is read-only and the explanation prompt is customizable.

★ 0 Stars0 Forks0 IssuesN/A Community rating0 Confirmed installs
View on GitHub
READMESource: main@139da75f

Concept explainer

English | 简体中文

Select an unfamiliar term in a DeepSeek Harness conversation and get an explanation grounded in the discussion around it, in the right sidebar. The explanation and your follow-ups stay there; the conversation gains, loses, and rewrites nothing.

Usage

  1. Select the term you do not understand — in an assistant answer or in your own prompt.
  2. Two equivalent entry points:
    • Click Explain in the composer's left tool row. It uses the current selection; with nothing selected it just opens the panel.
    • Or right-click the selection and choose Explain this.
  3. The right sidebar opens the Concept explainer tab, streams the answer, and lists which conversation fragments it used.
  4. Keep asking follow-ups at the bottom: Enter sends, Shift+Enter adds a line.
  5. Customize prompt at the top of the tab edits the system prompt sent to the model. Saved changes apply to later answers; Restore default brings back the built-in rules.
  6. The tab is also listed in the sidebar's + menu; its empty state explains how to start.

Why the conversation stays untouched

  • Requests go to the plugin's own POST /api/concept-explainer/stream, which only reads the session's messages as context.
  • The answer is written into the tab's React state only; no message is appended to the host session log.
  • The transcript lives in the panel's memory. Switching tabs keeps it (the tab stays mounted); restarting Harness clears it.
  • The panel header always states: "Reads conversation fragments as context only; never writes to the conversation".

How it is built

Part Where Host API used
Entry on selected text client.js ctx.uiConversation.interactions.register + the conversation.chat.text-interaction seat
Composer entry client.js the conversation.input.left seat
Right-sidebar tab client.js ctx.sidebarRightTabs.register + ctx.sidebarRight.openTab + the sidebar.right.pane.tab[.title] seats
Explanation index.js ctx.connection.fetch.register + ctx.llm.stream (the conversation's model, falling back to the default model)
Context prompt.js ctx.sessions.get(id).deriveMessages(), read-only
Custom prompt client.js + index.js the browser's localStorage holds the user's replacement system prompt and sends it with each request; an empty value falls back to the built-in rules

index.js and prompt.js import no Harness package: plain ESM plus the platform's own Request / Response / ReadableStream.

Install

The package is a plain dsh.bundle bundle with no build step. Install it through the Harness Plugin Manager's add-plugin flow, pointing at a local checkout of this directory; it then appears in the right sidebar's tab menu. The client half needs a page reload, a change to the host half needs a full Harness restart.

Local checks

node --check index.js; node --check prompt.js; node --check client.js
node test/host.test.mjs
node test/client.test.mjs

test/host.test.mjs drives the real request handler with a fake Cordis context: payload validation, the context window, prompt assembly, the NDJSON stream, model failures and follow-up history — and asserts that the session write path is never called. test/client.test.mjs captures the browser module with a fake loader and drives apply against a recording context: every registration, the interaction admission rules, the prompt editor, and both panel states.

Known limits

  • Only two entry points exist on purpose: the host allows one text interaction at a time, and the built-in "Quote text" bubble already owns the left-selection gesture. Taking it over would replace a shipped feature.
  • The custom prompt lives in this browser's localStorage. It does not sync across devices and is never written into DSH settings or the session log.
  • A custom prompt replaces the built-in explanation rules. Conversation fragments are still delivered as read-only context inside the user message, so conversation text is never treated as instructions.
  • Explanations render as plain text with line breaks preserved; Markdown code blocks are not rendered.
  • No web search or dictionary lookup: the answer comes from the model and the conversation. Unknown concepts are reported as such rather than invented.
—/ 5

No ratings yet

Verified DSH bundle

Commit 139da75ff2fc

Community comments

No comments yet. Be the first to write one.

DSH HUB

A community index for DSH plugins. Not an official GitHub or DeepSeek AI product.

CommunityResourcesAPIAbout