MIT 授權的 DeepSeek Harness 原生外掛，把單一對話變成隊長領銜的多代理團隊——`dsh plugin add` 一鍵安裝，不需額外工作流程引擎。
DSH 整合
原生執行環境
作者聲明
安全稽核
未稽核
最後核實
2026-08-21
授權條款
MIT
01它能幫你完成什麼？
Turn one natural-language goal into a coordinated multi-agent team that runs and consolidates the result inside DeepSeek Harness
A captain-led team of durable sub-agents with dependency-aware tasks, direct messaging, and one consolidated final report
Developers and tech leads running DeepSeek Harness (dsh) who want ai agent automation with automatic task splitting and delegation
Watch and audit multi-agent runs in real time without a separate workflow engine
A live Web activity panel with segmented progress, a collapsible roster, and an interactive task DAG; completed archives retain full member and task history
Engineering teams that need visibility into parallel agent work inside the DSH Web UI
02如何將外掛接入 DeepSeek Harness？
先決條件
- An existing DeepSeek Harness installation (the README requires it before install)
- git and pnpm only when building from source
安裝步驟
- 01
Install from npm: `dsh plugin --profile web add @nanmicoder/dsh-agent-teams`
$ dsh plugin --profile web add @nanmicoder/dsh-agent-teams
- 02
Or build from source: `git clone https://github.com/NanmiCoder/dsh-agent-teams.git && cd dsh-agent-teams && pnpm install && pnpm build`, then `dsh plugin --profile web add .`
$ git clone https://github.com/NanmiCoder/dsh-agent-teams.git && cd dsh-agent-teams && pnpm install && pnpm build
- 03
Validate the composed profile and restart DSH: `dsh --profile web --dump-config` then `dsh web`
$ dsh --profile web --dump-config
- 04
Ask for a team directly in the session, e.g. ask AgentTeams to review commits and return one consolidated report
驗證整合成功
- Run `dsh --profile web --dump-config` to validate the composed profile, restart DSH with `dsh web`, and refresh the Web UI
- Ask in the session: "Use AgentTeams to review the commits after v0.5.3 from performance, security, and product perspectives. Return one consolidated report."
03DSH 整合程度與能力邊界
Native DeepSeek Harness plugin installed with `dsh plugin --profile web add @nanmicoder/dsh-agent-teams`; the current session becomes the team captain inside DSH
Captain-led delegation
a natural-language goal in the current DSH session (or `/agent-teams <goal>`)→a team where the current session creates the team, assigns roles, and consolidates the final result
spawns DSH sub-agents that consume model calls on the snapshotted provider/model routeDependency-aware tasks & automatic scheduler
a goal split into tasks with owners and explicit dependencies→tasks claimed atomically per idle member; idle members reused for ready work; revoked stale attempts and cold recovery of stranded open attempts
team state is written to disk under `<workspace>/.agent-teams/`Live activity panel (Web UI)
persisted team state plus live sub-agent activity→segmented progress, collapsible roster, interactive task DAG; completed archives keep full member and task history
Deterministic /agent-teams activation
`/agent-teams <goal>` from the Web GUI slash menu or the headless CLI gesture boundary→the handler queues one deterministic activation message and the captain protocol starts immediately; invocation logged as `command/run` / `command/done`
invocations are durably logged
04適合誰？何時不該用？
適合
- Developers and tech leads running DeepSeek Harness (dsh) who want ai agent automation with automatic task splitting and delegation
- Engineering teams that need visibility into parallel agent work inside the DSH Web UI
不適合
- State is file-backed and serialized within one DSH process; concurrent processes editing the same team are not coordinated.
05相容性、維護與安全提醒
- State is file-backed and serialized within one DSH process; concurrent processes editing the same team are not coordinated.
- One captain leads one active team at a time.
- The activity panel reports persisted state as-is; models may occasionally finish work without performing the expected task-state update.
MIT · actively maintained (latest release v0.1.8, 2026-08-19)
06常見問題
dsh-agent-teams 如何接入 DeepSeek Harness？
它是 DSH 原生外掛：執行 `dsh plugin --profile web add @nanmicoder/dsh-agent-teams` 安裝，再用 `dsh --profile web --dump-config` 驗證設定、`dsh web` 重啟並重新整理 Web UI。它不是 MCP 伺服器，而是讓目前對話直接成為團隊隊長。
需要什麼前置條件？
只需要既有的 DeepSeek Harness 安裝。預設設定開箱即用；受信賴的 profile 可選擇性覆寫成員行為（stateDir、memberProvider、memberModel、maxMembers 等）。
怎麼發起一個團隊任務？
在 Web GUI 的斜線選單選擇 `/agent-teams`，在 headless CLI 等介面以 `/agent-teams` 開頭輸入，或直接用自然語言提問，例如「Use AgentTeams to review the commits after v0.5.3 … Return one consolidated report.」。
團隊狀態存在哪裡？
存在工作區的 `<workspace>/.agent-teams/` 目錄下。Web 面板讀取這份磁碟上的真實狀態，並結合子代理的即時活動展示；狀態在單一 DSH 程序內序列化。
有哪些主要限制？
同一時間只能有一位隊長領導一個運作中的團隊；多個 DSH 程序並行編輯同一團隊不會被協調；活動面板如實反映持久化狀態，模型偶爾完成工作但不執行預期的任務狀態更新。
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
`dsh-agent-teams` turns the current DeepSeek Harness session into a captain that can assemble durable sub-agents, split…
dsh plugin --profile web add @nanmicoder/dsh-agent-teams
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。