Apache-2.0 人類與智慧代理協作平台，透過官方 @agentrq/dsh-plugin-agentrq 外掛將工作區 MCP 工具橋接進 DeepSeek Harness 即時對話。
DSH 整合
相容
作者聲明
安全稽核
未稽核
最後核實
2026-08-21
授權條款
Apache-2.0
01它能幫你完成什麼？
Delegate and track AI-agent tasks from inside DeepSeek Harness via a shared human-agent workspace
Tasks assigned to the agent delivered over the MCP channel into the live dsh session, with human replies, status updates and permission requests synced in real time to the AgentRQ task board
Teams running DeepSeek Harness (dsh) who want human-in-loop task delegation, review and real-time chat with their agents across mobile, web and desktop
Automate recurring work with scheduled tasks and event-driven workflows
Cron-scheduled and recurring tasks spawned automatically by a background poller, plus visual event workflows that create downstream tasks across workspaces with no glue code
Operators building multi-step release or content pipelines who want agent work chained together and visible on a graph instead of scripted
Supervise all workspaces and tasks across the whole AgentRQ account from one MCP endpoint
Global management via the Supervisor (CoreMCP) at mcp.agentrq.com — list/create/update workspaces, listAllTasks, reassign, reorder and unified workspace statistics
Power users and supervisors who manage several agent workspaces and need a bird's-eye view instead of per-project profiles
02如何將外掛接入 DeepSeek Harness？
先決條件
- DeepSeek Harness (dsh) CLI available via npx (@deepseek-ai/dsh)
- An AgentRQ workspace with its MCP URL and token (visible in Workspace Settings → Setup → DeepSeek Harness)
安裝步驟
- 01
Install the plugin into a dedicated dsh profile: `npx @deepseek-ai/dsh plugin --profile agentrq-<workspace> add @agentrq/dsh-plugin-agentrq`
$ npx @deepseek-ai/dsh plugin --profile agentrq-<workspace> add @agentrq/dsh-plugin-agentrq
- 02
Pin the workspace's MCP URL in `~/.dsh/profiles/agentrq-<workspace>/cordis.patch.yml` using the config block copied from Workspace Settings → Setup → DeepSeek Harness
- 03
Start the harness with that profile: `npx @deepseek-ai/dsh --profile agentrq-<workspace>` (one profile per workspace; switching workspaces means switching profiles)
$ npx @deepseek-ai/dsh --profile agentrq-<workspace>
驗證整合成功
作者未說明
03DSH 整合程度與能力邊界
Official @agentrq/dsh-plugin-agentrq bundle installed into a dsh profile; bridges the workspace's MCP tools into the live harness session as mcp__agentrq__*
Live workspace session in dsh
AgentRQ workspace MCP URL + token pinned in the dsh profile→workspace tools bridged to the model as `mcp__agentrq__*`; assigned tasks and human replies arrive in the live session with delivery, startup catch-up and reconnect handling
holds a supervised workspace session over the network to the workspace MCP endpointTask & status management via MCP tools
task assignments from the board or human chat→`createTask` (with optional cron_schedule), `updateTaskStatus` (notstarted / ongoing / blocked / completed), `getTask` dequeue, `reply`, `getWorkspace`, `downloadAttachment`
task status changes and replies are written to the AgentRQ backend and synced in real timeVisual task board & scheduling
tasks created by humans or agents→list / Kanban dashboard on any device, launch-date and recurring tasks (every 15 min, hourly, daily, weekly, custom) spawned by a minute-ticking background poller
Event-driven workflows
named signals such as `qa_passed`, `deploy_finished`, `blog_published` fired when tasks complete→subscribed workspaces automatically get new tasks — release processes wired on a graph with no polling or glue code
completing a task that emits an event automatically creates tasks in subscribed workspaces
04適合誰？何時不該用？
適合
- Teams running DeepSeek Harness (dsh) who want human-in-loop task delegation, review and real-time chat with their agents across mobile, web and desktop
- Operators building multi-step release or content pipelines who want agent work chained together and visible on a graph instead of scripted
- Power users and supervisors who manage several agent workspaces and need a bird's-eye view instead of per-project profiles
不適合
- The dsh plugin connects to the workspace's remote MCP endpoint (production URLs follow https://WORKSPACE_ID.mcp.agentrq.com/), so task delivery and real-time notifications require network access and a valid token.
- While the platform itself is self-hostable, the dsh plugin flow relies on the workspace MCP URL and filled-in config served from the AgentRQ dashboard; fully air-gapped operation would require self-hosting the backend.
05相容性、維護與安全提醒
- A dsh profile serves exactly one workspace and carries its own endpoint; managing several AgentRQ workspaces from DeepSeek Harness means running and switching multiple dsh profiles.
- The dsh plugin connects to the workspace's remote MCP endpoint (production URLs follow https://WORKSPACE_ID.mcp.agentrq.com/), so task delivery and real-time notifications require network access and a valid token.
- While the platform itself is self-hostable, the dsh plugin flow relies on the workspace MCP URL and filled-in config served from the AgentRQ dashboard; fully air-gapped operation would require self-hosting the backend.
Apache-2.0 · actively maintained (latest release v0.4.2, 2026-08-19)
06常見問題
AgentRQ 如何接入 DeepSeek Harness？
透過官方 `@agentrq/dsh-plugin-agentrq` 外掛包，用 `npx @deepseek-ai/dsh plugin --profile agentrq-<workspace> add @agentrq/dsh-plugin-agentrq` 安裝到 dsh profile。它會將工作區工具以 `mcp__agentrq__*` 形式橋接給模型，並維持一個受監督的工作區對話。
是原生整合還是基於 MCP？
基於 MCP：外掛維持受監督的工作區對話，指派給智慧代理的工作與人類回覆都透過 MCP 通道直接進入 dsh 即時對話——無需輪詢，也不必離開 harness 去處理工作佇列。
安裝前需要準備什麼？
DeepSeek Harness CLI（可用 `npx @deepseek-ai/dsh` 呼叫）和一個 AgentRQ 工作區。工作區 ID、含 token 的完整 MCP 位址以及可直接貼上的設定，都在工作區的 Workspace Settings → Setup → DeepSeek Harness 裡。
一個 dsh profile 能服務多個工作區嗎？
不能。一個 dsh profile 只服務一個工作區並攜帶自己的接入端點，所以要為每個工作區各建一個 profile——切換工作區就是切換 profile。
我的資料會流向哪裡？
工作與回覆透過工作區 MCP 端點（正式環境形如 `https://WORKSPACE_ID.mcp.agentrq.com/`）與 AgentRQ 後端即時同步。AgentRQ 本身支援 Docker 自架，如需完全私有部署可自行執行後端。
07相關的 DSH 工作流程
archify
作者 tt-a1i
為編碼智慧體生成美觀可驗證的架構圖、時序圖與資料流圖，輸出自包含 HTML，支援動效與清晰匯出。
openviking
作者 volcengine
為 AI 智慧體打造的自進化上下文資料庫，統一智慧體記憶、知識 RAG 與技能。
nocobase
作者 nocobase
開源的 AI + 無程式碼應用搭建平臺：AI 在久經生產驗證的底層基礎設施與所見即所得的無程式碼介面之上工作，幫你快速構建 CRM、ERP 等業務系統，兼顧速度與可靠性。
learn-harness-engineering
作者 walkinglabs
Harness 工程新手教程，從 0 到 1 系統學習智慧體工作流框架。
08資料與來源
The `@agentrq/dsh-plugin-agentrq` bundle brings AgentRQ into DeepSeek Harness. It bridges the workspace's tools to the m…
npx @deepseek-ai/dsh plugin --profile agentrq-<workspace> add @agentrq/dsh-plugin-agentrq
此頁面根據專案公開文件、儲存庫中繼資料與 DSH Plugins 的結構化解析所產生；最後核實於 2026-08-21。發現錯誤？提交更正。
最佳 DeepSeek Harness 外掛
從全目錄挑出的 12 個值得優先安裝的外掛，涵蓋各個分類。