CuteSamurai24/dsh-md-memory

CuteSamurai24★ 0JavaScript最后同步: 2026-08-18

在 GitHub 打开

Small file memory for DeepSeek Harness. The model writes markdown, the host injects it.

README 摘要

dsh-md-memory English 简体中文 A small file memory plugin for DeepSeek Harness. The model writes what should persist as markdown under /.dsh/memory . Next session, the host injects it. This is not a second brain. The community already has fuller memory plugins (auto-extract, vector search, settings dashboards). This one stays small: the disk holds files you can open in a text editor, and nothing is saved unless the model calls memory . MIT What it does - Remembers preferences, corrections, and reusable lessons across sessions - Three files: user.md (cross-project) / agent/MEMORY.md (hot rules) / agent/topics/ .md (on-demand topics) - Project-only facts belong in the repo AGENTS.md , not here - Subagents cannot write , so parallel tasks cannot scramble memory - If you say things like “remember / from now on / don’t ever”, and nothing was stored this turn, the host nudges the model What it does not do - No automatic extraction from the chat - No secret scanning (do not let the model write tokens here) - No vector / full-text engine ( search is a plain substring) - No browser settings page Do not install this next to another plugin that also registers a tool named memory . Install Then re…

在 GitHub 查看完整 README →
工具/开发deepseek-harnessdeepseek-harness-plugindeepseek-harness-pluginsdsh-plugindsh-pluginsagentsearch

分类