Introduction

DeepSeek Harness (DSH) uses a plugin-based architecture, allowing developers to adapt specific workflows through functional extensions. In conversational scenarios, when it is necessary to quote, explain, or record thinking processes related to specific text, traditional methods such as taking screenshots or manually copying are often inefficient and can easily break the flow of context transmission. The dsh-plugin-text-quote plugin addresses this issue. It mimics Codex’s interaction model, allowing users to annotate text directly within a conversation and inject this information into the message stream as structured data, enabling the model to recognize and process the supplementary content.

Plugin Positioning

This is a conversational text annotation plugin maintained by sunnystarye-ui. It does not change the basic interaction logic of conversations. Instead, it provides an annotation dock on the input side and a pill display after messages are sent. Its core value lies in converting unstructured quotes and comments into machine-readable <!-- dsh-annotations --> markers, ensuring that the model can generate responses based on this annotated context.

Core Features

  1. Selection Annotation: Drag to select a piece of text in the conversation window, and the system displays a context menu. Click “Add to Conversation” to open the annotation box, saving the selected original text and the user’s comments.
  2. Input-side Annotation Dock: The annotation list appears in the area to the left of the input box. Users can expand, view, edit, or delete created annotations here.
  3. Quote Injection: When a message is sent, the plugin automatically injects annotation blocks (formatted as <!-- dsh-annotations -->) into the message body. This enables the large model to read the annotation content when processing the message.
  4. Post-send Pill: In sent messages, annotation blocks are collapsed into “N annotations” pills. Clicking a pill expands it to view the original text and corresponding comments.
  5. Keyboard Shortcut Support: While typing in the annotation box, press Enter to save the annotation, press Shift+Enter to insert a line break, or use Ctrl/Cmd+Enter to save.
  6. Local Operation: The plugin runs entirely locally, contains no telemetry data, and does not send requests to external services, maintaining a clean runtime environment.

Installation and Enablement

This plugin is based on the DeepSeek Harness web client architecture. Before installation, ensure that your environment meets Node.js version >= 18.

dsh plugin --profile web add dsh-plugin-text-quote

After installation is complete, the plugin runs as a client plugin (dsh.client.platform = web). It relies on the host environment to provide the @deepseek-ai/dsh-client-ui-slots interface.

Typical Usage

  1. Drag to select the text in the conversation history that you want to quote or annotate.
  2. Click “Add to Conversation” in the pop-up menu.
  3. Enter comments in the opened annotation box and press Enter to save.
  4. Click send. After the message is sent, the annotation block is automatically injected and displayed in the message content as an “N annotations” pill. Click the pill to expand and view the details.

Applicable Scenarios and Notes

  • Supported Platform: DeepSeek Harness web client (dsh.client.platform = web).
  • Dependencies: The host must provide @deepseek-ai/dsh-client-ui-slots slot support.
  • License: Governed by the MIT License; the code is open source.

This plugin helps developers manage contextual references in conversations more clearly through standardized data injection. If you want to view the source code or submit an Issue, please visit the GitHub repository.