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。

Session Tools — 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
S

dsh-session-tools

Session Tools

DeepSeek Harness 的会话归档、跨会话读取、消息、等待和复制 ID 工具

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:Unintendedz/dsh-session-tools#037e88cefb984a31aef70bc0420cbaddf9bf4931
README兼容性版本

兼容性与来源证明

Session Tools 以 dsh-session-tools 发布,当前版本为 0.2.2。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
web
发布来源
github
Registry 更新时间
2026/9/8

版本

0.2.2stable
2026/9/8
0.2.1stable
2026/9/6
0.2.0stable
2026/8/24
查看其余 2 个版本收起版本
0.1.1stable
2026/8/23
0.1.0stable
2026/8/23

相关插件

正在加载相关插件…

最新版
0.2.2
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
web
许可证
MIT
发布源
github
GitHub
★ 1
周下载
0
最近提交
2026/9/10
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-session-tools

English | 中文

A DSH Web plugin for session management across conversations:

  • session_archive: archive the calling session, or another session by exact ID.
  • session_read: read DSH's bounded, untrusted text snapshot of another session by exact ID.
  • session_send: send a message to another ordinary session and receive a durable request ID.
  • session_wait: wait for, or poll, the exact reply associated with a session_send request.
  • Copy session ID in every populated session row's three-dot menu.
  • ID header button: copy the exact ID of the currently open session.

session_send places one durable user message in the target's next ordinary turn. DSH's native agent resolver can resume an inactive ordinary session with its recorded context before delivery. session_wait correlates the request ID with that exact target turn and returns only its visible assistant text; it does not expose reasoning or tool traces.

Archiving is persistent but non-destructive: the session disappears from normal lists while its log remains stored. Cross-session reads reuse DSH's session-reference projection, so each reference keeps the host's configured byte limit.

Requirements

  • DSH 0.1.1-rc.2 / 0.1.2-rc.1
  • A Web profile (web in the examples below)

Version 0.2.2 reads live event snapshots on DSH 0.1.2-rc.1 and retains the legacy event-array path for DSH 0.1.1-rc.2. Polling sees newly committed replies and terminal status without reusing an older snapshot.

Install

dsh plugin --profile web add github:Unintendedz/dsh-session-tools#v0.2.2

Restart the running DSH Web service. Plugins are loaded when the service starts.

Usage

Use Copy session ID in a row's three-dot menu, or open a session and click ID in its header.

To ask another ordinary session to do work, call session_send:

{
  "session_id": "session-target-id",
  "message": "Check the test results and summarize any failures."
}

The tool returns immediately with a request_id. Pass both IDs to session_wait:

{
  "session_id": "session-target-id",
  "request_id": "request-id-from-session-send",
  "timeout_seconds": 60
}

session_wait returns completed, running, or failed. Its wait defaults to 60 seconds and accepts 0–600 seconds; use 0 to poll. A running result is safe to wait on again with the same IDs because waiting never resends the message or cancels the target.

You can also paste a copied ID into another conversation and ask it to call session_read, or ask it to call session_archive with that ID. Calling session_archive without session_id archives the current conversation.

The sidebar menu is repositioned after the copy action is injected, so all four rows remain inside the viewport even when the source conversation is near the bottom of the screen.

Update

Install the desired tag, then restart DSH Web:

dsh plugin --profile web add github:Unintendedz/dsh-session-tools#v0.2.2

Uninstall

dsh plugin --profile web remove dsh-session-tools

Safety

  • session_read uses DSH's native session-reference resolver. Returned text is bounded by the host and explicitly treated as untrusted context.
  • session_send and session_wait require two different ordinary sessions. Self-targeting is rejected; subagent sessions must use DSH's built-in send_message tool.
  • Only the session that created a request may wait for its result.
  • session_wait returns only visible assistant text from the request's exact turn. Reasoning and tool traces are excluded, and returned UTF-8 text is bounded to 65,536 bytes.
  • A wait timeout returns running; it does not resend the request or cancel the target. Cross-session wait cycles are rejected.
  • session_archive preserves the session log; it does not delete conversation data.
  • Tool arguments require exact, non-empty IDs and reject unknown fields.

Development

npm install
npm test

Set DSH_NATIVE_ROOT to the installed DSH package root containing its node_modules, and use a fresh temporary DSH_HOME, to include the native Session polling regression. It uses synthetic data and does not boot a profile.

The host entry point is lib/index.js; browser source lives in src/client.jsx. npm test rebuilds lib/client.js and runs the host and browser-bundle tests.

Request lifecycle and native output

session_wait associates an inbox claim with its active turn. If that turn is aborted, errors, or is blocked before the user message is recorded, waiting returns failed with that reason. A claim whose turn is still active remains running; ownership and queued-request cancellation checks still apply. Native tool output always includes the request status and any terminal reason, followed by the available reply text bounded to 64 KiB of UTF-8. A partial reply therefore remains visibly failed or running.