dsh-plugin-followup

Reading a lengthy assistant reply in DeepSeek Harness and wanting to drill into a specific code block or conclusion? The usual workflow is clunky: you copy, paste, format the quote by hand, and the follow-up Q&A floods into the main thread, pushing the original text further out of view. dsh-plugin-followup cuts this to three actions — select text, right-click, ask — and keeps the entire Q&A thread in a right-side panel while the corresponding turn in the main conversation is hidden (not deleted), so the model retains full context.

The plugin is built for one scenario: asking targeted follow-up questions on any text fragment of a completed reply. It works with body text, code blocks, wide tables, and tool output. Each follow-up card is a tab-style thread — you can keep asking in the same tab or start a new one by selecting different text. Full answers are read from the turn-tail node without truncation, and a built-in zero-dependency Markdown renderer handles code fences, tables, lists, and more. In multi-workspace setups, cards are scoped by session ID so nothing bleeds across.

Ideally suited for developers and researchers who regularly read long technical answers and need to probe specific passages repeatedly, as well as power users juggling multiple sessions who want follow-ups out of the main flow. It is a pure client-side plugin — no server-side deployment required.