dsh-memoryhub

Running long-term projects in DeepSeek Harness means every new session starts with the model forgetting earlier architecture decisions, coding conventions, and debugging context. MemoryHub stores project memory as git-versioned checkpoints, and this plugin wires that memory pipeline directly into the dsh session lifecycle so the model begins each conversation with full project recall.

Core capabilities include: auto-loading checkpoint memory on session start and injecting it as durable context with no extra prompt or tool call; an mh_save bridge that renders dsh durable event logs into a transcript format MemoryHub recognizes, where re-saving replaces by stable per-session identity rather than duplicating; six tools (mh_load, mh_save, mh_status, mh_list, mh_search, mh_checkpoint) that let the model perform common operations without shelling out; and a runtime-registered workflow skill that teaches the model when to load, when to save, and how to write compact summaries. The web UI gains a Memory tab that embeds MemoryHub's own checkpoint timeline map in an iframe, providing per-turn editing, token budget views, and every upstream feature unchanged.

This plugin is for users who run multiple related dsh sessions and want project-level context such as design decisions, API contracts, and hard-won debugging notes to persist across conversations without manual copy-paste or scattered notes. It turns memory into a versioned, searchable, revertible git record that both the model and you can read at any time.