Skip to content
  • Home
  • Plugin manual
  • dsh-tool-user-memory

IAMLieutenant/dsh-tool-user-memory

30Last commit Aug 15, 2026

dsh-tool-user-memory DSH plugin

This plugin provides persistent user memory for DSH agents. It automatically injects a user profile into system prompts every session, so the agent knows your preferences from the start. You can tell the agent to remember something, and it will persist across sessions.

skills-workflows

How to install the dsh-tool-user-memory DSH plugin

dsh plugin --profile web add dsh-tool-user-memory

Copying does not run this command. Review the repository and version before installing the dsh-tool-user-memory DSH plugin.

dsh-tool-user-memory DSH plugin data source

dsh-tool-user-memory DSH plugin snapshot date: Aug 16, 2026

discovered

What the dsh-tool-user-memory DSH plugin can do

  • memory_update tool: record, append, or remove user preferences with key-value pairs
  • memory_get tool: query the stored user profile with optional keyword filtering
  • Automatic injection of the user profile into every new session's system prompt, making the agent aware of preferences without additional tool calls

Where the dsh-tool-user-memory DSH plugin fits

  • Remember language preference: 'I prefer concise Chinese answers' — agent will use that style in all future sessions
  • Remember project context: 'I work on Python backend development' — agent can tailor responses accordingly
  • Remember communication style: 'Be direct and avoid pleasantries' — agent adapts its tone without repeated instructions

Who the dsh-tool-user-memory DSH plugin is for

  • DSH users who want a personalized, consistent agent experience across sessions
  • Developers who frequently switch contexts and want to avoid re-explaining their background every time

dsh-tool-user-memory DSH plugin limitations

  • Memory is global across all profiles and workspaces; no per-workspace isolation yet (planned for v2)
  • Currently only supports key-value text memory; semantic search (vector recall) is not available (planned for v2)
  • Memory file is human-readable and editable, but security relies on file system permissions

dsh-tool-user-memory DSH plugin: from the repository README

Quoted from the IAMLieutenant/dsh-tool-user-memory README, the upstream source of the dsh-tool-user-memory DSH plugin. Copyright remains with the original authors.

**DeepSeek Harness 用户偏好记忆插件**：让 agent 跨会话记住你的偏好——语言习惯、沟通风格、项目背景、目标……每个新会话都不必重新自我介绍。 > **独立开源插件** —— 作为 DeepSeek Harness 社区生态的一部分独立开发与维护 > （GitHub 话题：[`dsh-plugin`](https://github.com/topics/dsh-plugin)）。 > 与官方仓库无关；直接通过 npm 安装，30 秒启用。 [English](README.en.md) ｜ [更新日志](CHANGELOG.md) --- ## 1. 项目介绍 ### 它解决什么问题 默认情况下，DeepSeek Harness 的 agent **每次新会话都是"陌生人"**：不知道你偏好简洁还是详细、不知道你做什么项目、不知道你用什么语言交流——每个新会话都得重新交代一遍。 这个插件给 agent 加了一块**持久化的用户画像**： - 你说一句"我喜欢简洁的中文回答"，agent 把它写进记忆文件； - **之后每一个新会话**，这段记忆会自动注入系统提示词，agent 天生就知道——不用你提醒，也不用调工具。 ### 核心能力 | 能力 | 说明 | |---|---| | `memory_update(key, value, mode?)` | agent 学到你的稳定偏好时，自动记录 / 追加 / 删除 | | `memory_get(query?, limit?)` | 需要个性化回答时，主动读取你的画像 | | `{{user_profile}}` 系统提示词注入 | **每个会话每轮自动携带**你的画像（空画像零 token 成本） | | 持久化存储 | `$DSH_HOME/user-memory/user.md`，人类可读、可手改、可删除 | ### 工作原理（30 秒版） ``` 你说"记住：我喜欢简洁的中文回答" → agent 决定调用 memory_update → 写入 $DSH_HOME/user-memory/user.md（原子写、仅属主可读） → 之后每个新会话：系统提示词自动注入画像 → agent 天生认识你 ``` --- ## 2. 下载与安装 ### 前置条件 - 已安装并跑通 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness)（`dsh` CLI；已在 0.1.0-rc.x 验证） - 无需单独安装 npm 包！`dsh plugin` 会替你装好 ### 安装（推荐：一条命令） 在你想启用的 profile 上安装，例如 web： ```sh

Read the full READMERepository license: MIT

dsh-tool-user-memory DSH plugin questions

How do I install dsh-tool-user-memory?

Run `dsh plugin --profile web add dsh-tool-user-memory` for the web profile, or replace `web` with your profile name. Then restart your DSH session (e.g., `dsh web`). The plugin will be activated automatically. No additional npm install is needed.

How can I make the agent remember something about me?

Just tell the agent naturally, like "Remember: I prefer concise Chinese answers." The agent will detect this as a stable preference and call the `memory_update` tool to store it. You don't need to manually invoke any tool.

How do I check what the agent currently remembers about me?

Ask the agent directly: "What do you know about me?" or "What is my language preference?" If the memory is already injected into the system prompt, the agent will answer without calling any tool. Otherwise, it can use `memory_get` to retrieve the stored profile.

Can I delete a specific memory?

Yes. You can say "Forget my preference about X" and the agent will call `memory_update` with mode `remove` to delete that key. Alternatively, you can manually edit the memory file at `$DSH_HOME/user-memory/user.md` and remove the relevant lines.

Does the memory persist across different DSH profiles?

Yes, the memory is stored globally under `$DSH_HOME/user-memory/user.md`, so it is shared by all profiles (web, headless, etc.) and all workspaces. It is automatically injected into every new session regardless of the profile.