MIT-licensed DeepSeek Harness plugin that turns bots with their own role, model and memory into persistent teammates and group chats, with delegated background tasks and a local follow-up queue.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-11-14
License
MIT
01What can it help you accomplish?
Give each recurring job — research, review, writing — its own bot with a standing brief, a chosen model and its own memory inside DeepSeek Harness
A separate main session per bot, an editable role brief and a private memory list you add entries to and approve
DeepSeek Harness users who keep re-pasting the same context and preferences into every conversation
Split one task across several bots in a group chat — for example a research bot collecting comparison dimensions and a review bot checking assumptions and evidence
Concurrent member answers with an expandable per-member process, per-question and per-group usage stats, and the option to top up or stop an individual member
People comparing options or writing reviews who want an auditable division of labour instead of one long prompt
Hand work to the background and come back to a review queue instead of watching a chat
A per-bot task page with delegated subtasks, their evidence and usage, plus follow-up cards for items awaiting acceptance, failed, explicitly waiting or manually registered
Users who delegate multi-step work and need to accept, top up or rework results — without wiring GitHub CI
02How to install into DeepSeek Harness
Prerequisites
- A DeepSeek Harness host with a usable model configured — chat and model tasks need one, while enabling and viewing local follow-up does not call a model
- The `dsh` CLI with the profile you actually run (the README uses `web` for web instances)
- Anyone on the old `dsh-bot` package must run the package-name migration first: stop the Host and back up the profile and its full data
Installation steps
- 01
Desktop app: open 插件 → 添加插件 → 安装第三方插件 (Plugins → Add plugin → Install third-party plugin) and enter `dsh-botgroup@0.6.7`
- 02
Web instance: `dsh plugin --profile web add dsh-botgroup@0.6.7`, replacing `web` with your own profile name
$ dsh plugin --profile web add dsh-botgroup@0.6.7
- 03
Alternative — GitHub package: download `dsh-bot.tgz` and `dsh-bot.tgz.sha256` from the same release (not the auto-generated source archive), verify with `shasum -a 256 -c dsh-bot.tgz.sha256` (or `sha256sum -c` on Linux), then install the `.tgz` through the same plugin entry
- 04
Restart the host as it prompts, fully quit and reopen DSH, then open Bots in the sidebar
Verify the integration
- After the full restart the sidebar has a Bots entry — create a bot, describe its role, and it keeps a main session
- Memory check: add one preference on the bot's 记忆 (Memory) page and confirm it shows up in the list
- After the package-name migration, start the Host and check that config, bots, memory, groups and the old sessions are all still there
Rollback
- Rolling back needs the backup taken before upgrading: keep the original profile and its storage location, add the new package through the same entry, then fully quit and reopen
- Package-name migration: before starting the Host, inspect the plugin override line with `id: dsh-bot` in `cordis.patch.yml` — if it has `name: dsh-bot`, change it to `name: dsh-botgroup` while keeping the original `id`, `config` and `dataDir`; lines without `name` stay untouched
03DSH integration and capability boundaries
A DSH plugin package (`dsh-botgroup`) added through the DSH plugin manager or `dsh plugin add`, which mounts Bots, multi-bot group chats, proactive follow-up and native task delegation inside the host.
Personal bots with role, model, tools and skills
a bot profile: name, long-lived role brief, model, tools and skills→a dedicated main session per bot that keeps that role and its own memory
long-lived role edits are confirmed through the host's native approval flowbot profiles and private memory are stored in the DSH profileMulti-bot group chat
a group of members plus a question, with @-mentions to target one member→concurrent member answers, per-member process, `/` command menu and usage stats per question or per group
each member run consumes model quota, reported in the usage statsa failing member shows the concrete reason, for example exhausted quotaBackground subtask delegation and acceptance
a delegated piece of work inside a bot's session→native subtask records with evidence and usage, plus accept / top up / rework actions where the task supports them
delegated subtasks run through the host's native subagent capability and spend model quotaa hard kill leaves uncertain background results that are not replayed automaticallyProactive follow-up on local tasks and explicit items
a follow-up scope (one bot or one group) plus optional manually registered items with a completion definition→follow-up cards for items awaiting acceptance, failed, explicitly waiting or manually registered, each linking back to the original task
registering a follow-up item starts no task and sends no messagepersonal registrations go through native tool approval; group ones become proposals confirmed in the activity pageDocument analysis and method evaluation
the document body, question and acceptance criteria you explicitly submit→an analysis result in the task list, and method candidates you review before adopting
does not fetch external links and does not treat instructions inside the document as authorisationdoes not modify working methods automatically
04Who is it for? When not to use it?
Good for
- DeepSeek Harness users who keep re-pasting the same context and preferences into every conversation
- People comparing options or writing reviews who want an auditable division of labour instead of one long prompt
- Users who delegate multi-step work and need to accept, top up or rework results — without wiring GitHub CI
Not for
- Development is pinned to DSH `0.2.1-rc.2`: 0.6.7 was verified on macOS against the isolated `0.2.1-rc.2` and `0.2.2-rc.1` hosts, and native rc.2 still has a managed-bot fork limitation.
- Everything runs on the local DSH Host: after the machine shuts down background tasks are not promised to continue, and zero-model scheduling, external IM delivery, system/phone push and cloud hosting are explicitly out of scope for this version.
- A bot's working directory is not a file-isolation boundary — bots inherit the host's permissions, approvals and sandbox, so the guard rails are the ones DSH already enforces.
- The package name moved from `dsh-bot` to `dsh-botgroup`; existing installs must back up the profile and full data, remove the old package, install the new one and fix the `id: dsh-bot` override line in `cordis.patch.yml` before starting the Host.
05Compatibility, maintenance and safety notes
- Development is pinned to DSH `0.2.1-rc.2`: 0.6.7 was verified on macOS against the isolated `0.2.1-rc.2` and `0.2.2-rc.1` hosts, and native rc.2 still has a managed-bot fork limitation.
- Everything runs on the local DSH Host: after the machine shuts down background tasks are not promised to continue, and zero-model scheduling, external IM delivery, system/phone push and cloud hosting are explicitly out of scope for this version.
- A bot's working directory is not a file-isolation boundary — bots inherit the host's permissions, approvals and sandbox, so the guard rails are the ones DSH already enforces.
- The package name moved from `dsh-bot` to `dsh-botgroup`; existing installs must back up the profile and full data, remove the old package, install the new one and fix the `id: dsh-bot` override line in `cordis.patch.yml` before starting the Host.
MIT · README evidence only, latest release v0.6.7 (2026-11-13), no security audit
06Frequently asked questions
How do I install DSH-Bot into DeepSeek Harness?
In the desktop app open 插件 → 添加插件 → 安装第三方插件 and enter `dsh-botgroup@0.6.7`; on a web instance run `dsh plugin --profile web add dsh-botgroup@0.6.7` (swap `web` for your profile). A GitHub `.tgz` from the release also works through the same plugin entry — verify it against the matching `.sha256` first, and never use the auto-generated source archive. Then fully quit and reopen DSH and open Bots in the sidebar.
Do the bots need their own model account or API key?
No. The plugin reuses the host's model configuration and lets you pick a model per bot and per group member. Chat and model tasks need a usable model already configured in DSH; enabling, viewing and refreshing local follow-up does not call a model. Model usage may cost money.
Is it safe to let several bots run in a group chat?
Members stop when there is nothing left to do: the default is at most 4 concurrent members per group with an automatic follow-up allowance of 32 executions, both configurable, and you can stop a member or the whole task. Native model and tool time limits still apply, and a failing member shows the concrete reason, for example exhausted quota.
Can background tasks keep running after I close my laptop?
No — the README explicitly refuses to promise that. Tasks depend on the local DSH Host, and this version ships no cloud residency or continued execution after the system exits. Zero-model scheduling, external IM delivery, system/phone push and cloud hosting are also outside this release.
I installed the old `dsh-bot` package — what changes?
Stop the Host, back up the profile and its full data, remove the old package through DSH and install the new one. Before starting, check the plugin override line with `id: dsh-bot` in `cordis.patch.yml`: if it has `name: dsh-bot`, change it to `name: dsh-botgroup` while keeping the original `id`, `config` and `dataDir`. Once only the new package is left in that profile, start the Host and verify config, bots, memory, groups and old sessions.
07Related DSH workflows
archify
by tt-a1i
Turn any idea, plan, or codebase into a beautiful interactive diagram. An agent skill for Claude Code, Codex, and more.
openviking
by volcengine
Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
aitoearn
by yikart
Let's use AI to Earn!
learn-harness-engineering
by walkinglabs
Harness engineering beginner tutorial, from 0 to 1
08Data and sources
**你的个人智能体工作区。** 在 DeepSeek Harness 中创建各有职责与记忆的 Bot，持续对话、跟进工作，也可以把它们组成群聊，分工处理同一个任务。
桌面版打开 **插件 → 添加插件 → 安装第三方插件**，输入 `dsh-botgroup@0.6.7`，安装后按宿主提示重启。
通过原生子任务委派工作，查看证据和用量，对支持的任务验收、补充或返工；普通日程沿用 DSH 原生能力。
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-14. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.