Introduction

In DeepSeek Harness (DSH) usage scenarios, mind maps usually require manual drawing or are separated from source files. The dsh-mindmap plugin unifies the two: a Markdown file is the mind map, the chat window serves as the editor, and the AI progressively modifies the .md file through tools while the right panel renders the map in real time. The plugin is in the pre-1.0 stage, with core functionality fully covered and tested.

Core Features

The plugin drives workflows through four core tools (mindmap_create / mindmap_open / mindmap_get / mindmap_update).

  1. File and root node synchronization
    The root heading of a Markdown file stays synchronized with the file name. When the AI changes the root node name, the file name is updated accordingly (conflicts must be handled).

  2. Real-time rendering and state restoration
    The plugin uses a real-time rendering mechanism without extra channels, updating the view based on session snapshots (mindmap_* tool results). It provides reliable opening mechanisms and recoverable loading states, even when structure fingerprints change or paths mismatch.

  3. Canvas and interaction

    • MarkGrove-style mapping: Supports heading levels, nested lists, code blocks, and paragraphs as independent nodes.
    • Operations: Supports canvas centering, zooming (25%-300%), panning, node search and quick navigation, and subtree collapsing.
    • Layout: Supports a floating panel (280px~80% viewport) or a native sidebar (requires installing dsh-better-sidebar).

  4. Draft and refresh protection

    • Draft protection: Uses host input snapshots to protect drafts, references, and attachments.
    • Refresh from disk: Refreshes the file without sending a model request, preserving drafts and attachments and preventing stale requests from overwriting the new view.

  5. Export and security

    • Export: Supports one-click PNG image export and copying the full text of the current mind map as Markdown source.
    • Security: The default write approval is scoped to a single session/document, preventing continuous AI writes from interrupting workflows. The client has no direct file-system write permissions; all edits are performed through AI tools.

Installation and Enablement

This plugin depends on the DSH main environment and an optional sidebar plugin.

  • Dependencies:

    • @deepseek-ai/dsh-tools (must meet specific version requirements)
    • dsh-better-sidebar (optional, for native sidebar integration, version >= 0.10.0)
    • @deepseek-ai/schemastery (>= 3.18.0)

  • Where to get it:
    The plugin source code is available in the GitHub repository: https://github.com/guhanfei-ai/dsh-mindmap

Typical Usage

  1. Open and render: Open a Markdown file in the session working directory, and the plugin parses it into a mind map. The root node heading automatically corresponds to the file name.
  2. Iterative editing: Enter instructions in the chat, and the AI calls the mindmap_update tool to modify the file; the right panel updates instantly.
  3. Export: Click “Export Image” in the toolbar to generate a PNG, or click “Copy Full Text” to copy the Markdown source to the clipboard.
  4. Navigation: Use the zoom bar or keyboard shortcut (⌘/Ctrl+F) to search nodes, and press Enter to jump between matches.

Applicable Scenarios and Notes

  • Applicable scenarios: Suitable for scenarios that require converting structured documents (such as outlines and notes) into visual mind maps, as well as for gradually refining mind-map structures through AI collaboration.
  • Notes:
    • The project is in the pre-1.0 stage, and cross-version compatibility has not yet been certified.
    • Write approval defaults to a single session/document; pay attention to security settings.
    • The hierarchical structure, nested lists, and code blocks in Markdown files all support MarkGrove-style mapping.

Conclusion

dsh-mindmap provides a workflow that tightly integrates Markdown files with mind maps. Through its toolchain and real-time rendering mechanism, it reduces the cost of converting text into structure. Project address: https://github.com/guhanfei-ai/dsh-mindmap