dsh-md-memory

Across multiple sessions, the model forgets the preferences you keep repeating, the corrections you have made, and the reusable lessons that matter. The community already offers heavier memory plugins with auto-extraction, vector search, and settings dashboards, but if you want the lightest, most transparent way to carry key context forward, dsh-md-memory exists for exactly that.

The idea is straightforward: the model writes what should persist as Markdown files under ~/.dsh/memory using a single tool called memory. Three layers keep things organized - user.md holds cross-project preferences, agent/MEMORY.md stores hot rules, and agent/topics/*.md keeps on-demand topic knowledge. At the start of each new session the host injects the stored content into the context. Subagents have no write access, so parallel tasks cannot overwrite each other; and if you say "remember this" or "from now on do it this way" but nothing was saved in that turn, the host nudges the model to write it down.

Who is it for? If your ideal of memory is a handful of plain text files you can open in any editor, read, edit, and delete at will - with no vector engine, no automatic extraction, no browser panel - then this plugin fits. It stays small and bounded (8 KB for user, 12 KB for hot rules, 20 KB per topic), coexists with AGENTS.md and heavier memory suites, as long as no other plugin registers a tool with the same name memory.