MIT TypeScript plugin pair that turns DeepSeek Harness into a digital-employee team — channels with a dispatching lead, threads and DMs, employees with their own model / skill / three-layer memory, versioned deliverables, a job canvas and a reviewable skill-writing loop, all on native DSH sessions.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-11-16
License
MIT
01What can it help you accomplish?
Run a multi-step task by describing it to the lead in a channel, instead of splitting the steps, switching sessions and re-pasting context yourself
Steps split by dependency, dispatched to the right employees in order, and the finished work handed back as versioned deliverable files you can finalize or download
People who regularly push AI through multi-step work — research plus writing, material sorting plus drafting, analysis plus reporting
Give every employee a standing brief so a new task does not have to be re-explained
Employees with their own name, responsibility, model and skill, plus global / channel / employee memory that the employees honour in later tasks
Teams that repeat the same kinds of jobs and want a fixed division of labour inside DSH
Separate what you handed in as material from what the employees produced, without losing earlier rounds
Every deliverable version kept — markable as final, openable online or downloadable — in one file view that separates files by source
Anyone who has to hand over or archive results and still wants the earlier drafts around
Turn a process that finally worked into a reusable skill instead of re-describing it next time
A skill draft the lead assembles, which you edit, approve or send back, then install in one click and bind to an employee
Teams whose tasks repeat on a cycle — weekly reports, recurring research — and who want the method to improve over time
02How to install into DeepSeek Harness
Prerequisites
- DeepSeek Harness `0.2.1-rc.5` installed, with at least one usable model configured — yolo-bot reuses the models and credentials registered in DSH instead of a separate account
- Node.js 22.19 or newer for the DSH Web edition; the source workspace asks for Node.js `^22.19.0 || >=24.0.0` and pnpm >= 9
- Both packages — `@cyancity/yolo-bot` (backend) and `@cyancity/yolo-bot-ui` (UI) — installed at the same version
Installation steps
- 01
Add `https://raw.githubusercontent.com/cyancity/yolo-bot/main/market/index.json` as an index source: on DSH's Plugins page open Experimental plugins → More, enable the custom plugin marketplace, switch to Discover and click Add index source
- 02
Install **YOLO Bot** and the UI package **YOLO Bot 界面** from the list; the two packages must stay on the same version
- 03
Fully quit DSH and reopen it once the marketplace says the plugin takes effect on next start
- 04
Off-marketplace alternative: download the two `.tgz` files from Releases and install the backend first, then the UI
- 05
DSH desktop edition: Plugins → Add plugin, paste the absolute path of the `.tgz` and enable it
- 06
DSH Web edition: `npx @deepseek-ai/dsh@0.2.1-rc.5 plugin --profile web add <yolo-bot.tgz absolute path>`, the same command for the UI package, then `npx @deepseek-ai/dsh@0.2.1-rc.5 web` — for another profile replace `web` with the target profile
$ npx @deepseek-ai/dsh@0.2.1-rc.5 plugin --profile web add <yolo-bot.tgz absolute path>
Verify the integration
- The marketplace reports that the plugin takes effect on next start; after a full quit and reopen, open the workbench
- First start already ships one channel and three employees — lead, analysis and execution — so you can describe a task there right away
03DSH integration and capability boundaries
DSH plugin pair — `@cyancity/yolo-bot` (Cordis host plugin) plus `@cyancity/yolo-bot-ui` (`dsh.client` web plugin), installed from the repository's marketplace index source or with `dsh plugin --profile … add …`; every employee runs as a native DSH session on DSH tools and models
Channels and lead dispatch
A task described in a channel→Steps split by dependency and dispatched in order; missing material or a needed human decision stops the run to ask you; the summary comes back as the result
Threads, @-mentions and DMs
Any message in a channel→A thread under that message for follow-up questions, a direct reply from an @-mentioned employee, and an independent DM conversation with one employee
DM history is stored separately from the channel and does not disturb the main lineEmployees with their own model, skill and memory
An employee's name, responsibility, model and skill→A standing employee configuration — model picked from the providers registered in DSH, blank meaning the global default — plus global / channel / employee memory
Memory persists: a standing instruction you write down is followed by the employees in later tasksMaterials and versioned deliverables
Files you upload, and files the employees hand over→Deliverables that keep every version, can be marked final, opened online or downloaded, with files separated by source in one view
Every deliverable version is kept, so earlier drafts stay retrievableDistilling a working process into a skill
A process that ran smoothly, turned into a skill draft by the lead→A skill you edit, approve or send back for rework, then install in one click and bind to an employee
A draft with blocking findings cannot be installed until it passes your reviewJob canvas, layouts and themes
Which employees are working, at which step, and which runs finished or failed — clicking a node shows its session and task; two layouts, light/dark following the system, and a Chinese or English UI
04Who is it for? When not to use it?
Good for
- People who regularly push AI through multi-step work — research plus writing, material sorting plus drafting, analysis plus reporting
- Teams that repeat the same kinds of jobs and want a fixed division of labour inside DSH
- Anyone who has to hand over or archive results and still wants the earlier drafts around
- Teams whose tasks repeat on a cycle — weekly reports, recurring research — and who want the method to improve over time
Not for
- The backend and UI packages must run at the same version, and neither upgrading DSH nor pulling new source updates them automatically — you update both packages and then fully quit and reopen DSH. Keeping the existing `DSH_HOME`, profile and data directory is what preserves employees, channels, sessions and materials.
- Employees spend your own model quota and read/write files and execute commands as the task requires, so important operations have to be verified by hand.
- A one-click backup writes workbench data and native sessions to the local machine — materials, every deliverable version, conversations and execution directories — but waits for running tasks and queues changes meanwhile. The DSH credential store is not included, so keys must be re-entered on a new machine, and a backup may contain sensitive content you typed, so it must not be shared.
05Compatibility, maintenance and safety notes
- The backend and UI packages must run at the same version, and neither upgrading DSH nor pulling new source updates them automatically — you update both packages and then fully quit and reopen DSH. Keeping the existing `DSH_HOME`, profile and data directory is what preserves employees, channels, sessions and materials.
- Employees spend your own model quota and read/write files and execute commands as the task requires, so important operations have to be verified by hand.
- Workbench data stays on your machine in `workbench/` under the DSH data directory and employee sessions are managed by DSH; there is no usage telemetry and no model credentials are stored. The one network exception is the site icon of a web-search result, requested from Google's favicon service, which can see those domains and your IP.
- A one-click backup writes workbench data and native sessions to the local machine — materials, every deliverable version, conversations and execution directories — but waits for running tasks and queues changes meanwhile. The DSH credential store is not included, so keys must be re-entered on a new machine, and a backup may contain sensitive content you typed, so it must not be shared.
MIT · no release yet · actively developed (last push 2026-10-30)
06Frequently asked questions
How do I install yolo-bot into DeepSeek Harness?
Add `https://raw.githubusercontent.com/cyancity/yolo-bot/main/market/index.json` as an index source — on DSH's Plugins page open Experimental plugins → More, enable the custom plugin marketplace, switch to Discover and click Add index source — then install **YOLO Bot** and the UI package **YOLO Bot 界面** from the list, both packages on the same version, and fully quit and reopen DSH. Without the marketplace, download the two `.tgz` files from Releases and install the backend first, the UI second; the Web edition uses `npx @deepseek-ai/dsh@0.2.1-rc.5 plugin --profile web add <tgz absolute path>`.
Which DSH and Node.js versions does it need?
DSH `0.2.1-rc.5` with at least one usable model configured, and Node.js 22.19 or newer for the DSH Web edition. The source workspace asks for Node.js `^22.19.0 || >=24.0.0` and pnpm >= 9, matching the DSH host.
How does it behave inside a DSH session?
Every employee is a native DSH session that runs on DSH's tools and models; the lead only dispatches and never does the employees' work, and you can bypass it at any time to @ an employee or DM it. It reuses the models and credentials already registered in DSH, so there is no second account to create.
What permissions and risks should I expect?
Employees spend your own model quota and will read/write files and execute commands as a task requires, so check important operations yourself. Workbench data stays on your machine in `workbench/` under the DSH data directory, no usage telemetry is reported and no model credentials are stored; the single network exception is the site icon of a web-search result, fetched from Google's favicon service, which sees those domains and your IP.
What are the operating limits?
The backend and UI packages must be updated together to the same version, and neither upgrading DSH nor pulling new source updates the plugins automatically. Backups exclude the DSH credential store — keys must be re-entered on a new machine — and a backup may contain sensitive content, so do not publish it. The core ships no industry catalogue or production workflows; those belong in separate extensions.
07Related DSH workflows
dsh-agent-teams
by nanmicoder
AgentTeams 多智能体团队协作插件:自然语言驱动建团队、拉成员、拆任务声明依赖、成员间消息直达,Web 右上角活动面板
dsh-blender-plugin
by sixtysevenlf
Adds tools for Blender scene editing, viewport capture, render profiling, headless work, and verification.
dsh-free-search
by dddmuc
DSH 免费搜索插件,执行联网检索。
dsh-bili-scroll
by c1overzz
Schedules Bilibili discovery, reads subtitles, writes daily reports, and optionally posts a comment tagging you.
08Data and sources
**2. 安装。** 在列表里安装 **YOLO Bot** 和 **YOLO Bot 界面**,两个包必须保持同一版本。看到下次启动生效的提示后,完全退出 DSH 再重新打开。
每名员工背后都是一个 DSH 原生会话,用 DSH 的工具和模型执行。负责人只调度,不替员工干活;你随时可以绕过负责人,直接 @ 某名员工或私信它。
| `@cyancity/yolo-bot` | [`workbench-host`](workbench-host) | Cordis host 插件 |
| `@cyancity/yolo-bot-ui` | [`workbench-ui`](workbench-ui) | `dsh.client` web 插件 |
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-16. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.
