dsh-mermaid

In the DeepSeek Harness web client, mermaid fences that appear in a conversation are currently shown only as raw code blocks; to see the actual diagram you have to copy the syntax into an external renderer. dsh-mermaid removes that friction: whether the message comes from the user or the assistant, any mermaid fence is automatically rendered as an inline SVG diagram with no extra steps.

The rendering pipeline reuses Mermaid official parser for syntax validation, so errors surface as warnings directly beneath the diagram instead of failing silently. Supported diagram types include flowchart, sequenceDiagram, classDiagram, erDiagram, and stateDiagram. Every rendered diagram ships with a collapsible source block for quick review or editing, and the /mermaid command lets you re-render the last assistant message on demand. Results are cached at (session, turn) granularity so repeats are not recomputed, and oversized fences exceeding 400 lines or 20000 characters are skipped with a notice to keep the page responsive.

If you frequently draw flowcharts, sequence diagrams, or class diagrams alongside AI in Harness conversations and want mermaid syntax to appear as visual diagrams right in the chat instead of copying it to another window, this plugin is built for exactly that workflow — lightweight, backend-free, and ready to use out of the box.