dsh-plugin-mindmap

Hours-long conversations with an AI assistant tend to blur: you lose track of which thread was resolved, which branch is still open, and what the starting premise even was. Once the session ends or the project restarts, that context is gone. dsh-plugin-mindmap tackles exactly this conversation-amnesia problem.

In the background it incrementally classifies every new message as an append, a fork, or a brand-new topic, distilling the whole dialogue into independent storylines. The result is rendered as an interactive map using bezier gradient ribbons and six semantic node shapes (question, decision, feature, bugfix, refactor, research). Clicking any node opens a detail card with the question, tool-call summaries, and the answer. A status badge (ongoing, blocked, done, discussion ended) plus a one-line description sits under each topic title, so you can grasp the full picture at a glance. The classification is persisted to DEV_LOG.json at the workspace root; on restart the map reloads instantly with zero LLM calls, so the memory outlives any single session.

If you develop long-running projects with DeepSeek Harness and need to revisit discussion threads, track the progress of each topic, and keep institutional knowledge accessible across sessions and agents, this auto-growing mind map turns ephemeral chat into a searchable, shareable development memory rather than fragments buried in scrollback.