DeepSeek Harness Plugin Hub

发布与管理完整 Harness Profiles,发现适合你的插件。

探索

插件目录环境预设文档中心动态

社区

发布插件联系我们报告问题

相关链接

Plugin Hub GitHubDeepSeek Harness 官方项目系统状态隐私说明
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

独立、非官方社区项目,与 DeepSeek 官方无隶属、授权或背书关系。

Engram Session — DeepSeek Harness 插件(DSH Plugin)
DeepSeek Harness Plugin Hub
ProfilesPlugins分类动态文档登录管理 Profiles
ProfilesPlugins分类动态文档登录
← Plugins
E

dsh-engram-session

Engram Session

DSH 插件 dsh-engram-session

插件会安装到这里;不确定时保持 web。

npx -y @deepseek-ai/dsh plugin --profile web add github:xiuyuan18/dsh-engram-session#739249257d58aff1c0705f640ed577c75544803c
README兼容性版本

说明

DSH 插件 dsh-engram-session

兼容性与来源证明

Engram Session 以 dsh-engram-session 发布,当前版本为 0.4.2。Plugin Hub 会校验它的 manifest,并保存精确安装来源,便于复现安装结果。

DSH 兼容范围
*
运行环境
any
发布来源
github
Registry 更新时间
2026/8/23

版本

0.4.2stable
2026/8/23
0.4.1stable
2026/8/23
0.3.0stable
2026/8/23

相关插件

正在加载相关插件…

最新版
0.4.2
DSH
*
HMR
重启进程
Tree shaking
未声明可安全裁剪
解包体积
未提供
文件数
未提供
Surface
any
许可证
未声明
发布源
github
GitHub
★ 1
周下载
0
最近提交
2026/8/25
查看源码 ↗
README Badge

点击下方 Badge 复制 Markdown,粘贴到 README 即可。

这是你的 Plugin?认领权益 · 优先安全扫描

验证 package.json 声明的 GitHub 仓库,即可管理这个公开页面。认领后,Hub 会优先安排当前版本的安全扫描,并在通过后公开展示结果。

认领这个 Plugin →
报告问题

相关插件

继续浏览 memory-context 分类下经过校验的插件。

Contextdsh-context用于上下文洞察和管理的 DeepSeek Harness 插件,提供上下文仪表板和上下文命令,帮助了解上下文的构成及其演变过程。Weknora@wxg-prc-cpg/dsh-weknora适用于 DeepSeek Harness (dsh) 的 WeKnora 知识检索工具:通过自有知识库进行语义搜索、文档阅读以及 RAG/代理回答。Memsearch Dsh@zilliz/memsearch-dsh适用于 DeepSeek Harness 的 MemSearch 插件:在多个代理之间共享 Markdown 记忆,支持捕获、步骤前上下文注入、记忆召回技能和技能候选审核面板。Memory@furongjun1999/dsh-memory灵枢(Lingshu·líng shū)DeepSeek Harness 插件:完整大脑——长期记忆/知识飞轮/自我认知/递归反思接入 DSH,对话自动沉淀进 md_cg 认知图(md 文档)

README

dsh-engram-session

Per-session Engram memory for DeepSeek Harness — a standalone bundle project reusable across profiles and presets.

This package is a bundle: its dsh.bundle.patch (cordis.patch.yml) is the composition layer that inserts the plugin row; the plugin then registers the Engram Memory Protocol as its own system-prompt section (engram:memory-protocol, order 10) at load — the deployment persona is never touched. Installing it into any profile is one command:

# from anywhere; <path> may be absolute or relative to your invoking directory
dsh plugin --profile web add /home/xiuyuaned/dsh-plugins/dsh-engram-session

dsh plugin runs pnpm add and reconciles dsh.profile.bundles — a dsh.bundle-declaring package joins the layer stack automatically. The bundle owns everything (plugin row + protocol section), so removing or disabling it leaves zero side effects (no tools, no orphaned memory instructions).

Setup for a new profile or preset

  1. Install dependencies once (self-contained — the project carries its own node_modules):

    cd /home/xiuyuaned/dsh-plugins/dsh-engram-session && pnpm install
    
  2. Add the bundle to the profile (repeat per profile):

    dsh plugin --profile <name> add /home/xiuyuaned/dsh-plugins/dsh-engram-session
    
  3. Restart the profile's app. A preset can also list the bundle directly in its dsh.profile.bundles plus a link: dependency in its package.json.

Removing is symmetric: dsh plugin --profile <name> remove dsh-engram-session. Reinstall at any time with the add command — the source lives outside the profile, so removing the bundle never deletes it.

What it does

One engram mcp stdio child per agent session, spawned with cwd = session.header.cwd (the session's workspace), with the mem_* tools registered in that agent's own scope via agent.ctx.tools.register(). Scoped registration shadows the global tool layer, so concurrent sessions each see their own memory surface under the same mcp__engram__mem_* names without collisions.

Rooting the child at the session workspace makes Engram's project auto-detection (git root / .engram/config.json) deterministic per session — the documented fix for hosts whose MCP child cannot inherit a reliable cwd (DSH web GUI, VS Code, WSL, CI). No --project pinning needed.

The model-facing tool specs are locally corrected: engram's shipped scope parameter descriptions omit global, so the plugin rewrites them (description text only — names, types, and required flags are untouched) in the registered definitions, keeping the agent's tool contract accurate without waiting for an upstream release.

Config

FieldDefaultDescription
serverNameengramNamespace for model-facing tool names
binary(auto-detected)Engram executable; see Binary resolution
args['mcp']Arguments for the MCP stdio server
toolCallTimeoutMs60000Per tools/call timeout

Binary resolution

No machine-specific path is baked in. With binary unset, the plugin picks the first executable it finds: PATH lookup, then ~/.local/bin/engram, /usr/local/bin/engram, /opt/homebrew/bin/engram. If nothing is found it logs a warning and mounts without memory tools — the protocol section is skipped too, so agents are never told about tools that cannot exist.

Set binary in the profile config (an absolute path, or a bare command name) to override detection. A set-but-missing value logs a warning and falls back to detection, so profiles created by pre-0.4 bundles that pinned /home/xiuyuaned/.local/bin/engram self-heal on other machines.

Lifecycle

  • agent/created (and a scan of already-live agents at mount) → spawn + register
  • agent/disposed → unregister tools, close client (kills the child)
  • plugin dispose → same teardown for every live entry

Project layout

  • index.js — the plugin: per-agent Engram MCP child + scoped mem_* tool registration
  • cordis.patch.yml — the bundle composition layer (plugin row)
  • AGENTS.md — guidance for AI agents working on this project
  • package.json / pnpm-lock.yaml — ESM manifest and lockfile

Development

See AGENTS.md for architecture details, conventions, verification steps, and pitfalls. Quick checks:

node --check index.js

Known limitations

  • No reconnect supervision: if the child dies mid-session, its tools fail until the session ends; restart the GUI to recover.
  • Text-only result projection: image/audio/embedded MCP blocks become diagnostics (Engram returns text results).