@poiuyjie

DSH plugin: Auto-convert images to text for pure-text LLMs (DeepSeek etc.) via any vision model. No need to switch your main model.

DeepSeek Harness plugin: open .md references from chat in a dockable side-by-side preview panel instead of the system editor; per-session recent-file tracking; pairs with dsh-vision-opencode for visual-verification workflows.