dsh-lark

面向飞书 / Lark 的 DeepSeek Harness（dsh）原生渠道插件。在 IM 中运行 Agent、切换工作区和模型，并通过人机协作卡片在聊天里掌控关键决策（提问、计划审阅、工具审批）。

DSH 适配

原生运行时

作者声明

安全审计

未审计

最后核验

2026-09-01

许可证

BSD-3-Clause

01它能帮你完成什么？

Run DeepSeek Harness agents from Feishu/Lark chat
Task progress, tool calls and final answers stream into Feishu/Lark as messages and interactive cards; the active workspace and model can be switched from chat
Teams and developers who run DeepSeek Harness (dsh) and want to dispatch and supervise agents from Feishu/Lark IM
632807d9abaf…632807d9abaf…
Approve or answer agent prompts without leaving Feishu/Lark
Human-in-the-loop cards — single/multi-choice or free text answers, plan approval/edits, and allow/deny tool calls
Operators and reviewers who need to control key agent decisions (questions, plan review, tool approval) from chat
632807d9abaf…
Collaborate multiple agents in one group chat
Multiple bots added with one command, handing off turns via @ mentions, with a turn cap to prevent infinite loops
Teams that want several DeepSeek Harness agents to coordinate inside a single Feishu/Lark group
632807d9abaf…

02如何接入 DeepSeek Harness？

前置条件

DeepSeek Harness (dsh) installed — if not yet, run `npm i -g @deepseek-ai/dsh`
A Feishu (Lark) account to scan the QR code and create the bot app
Node.js runtime (the channel is distributed as an npm package)

安装步骤

01
Install the channel globally: `npm i -g dsh-lark-channel`
$ npm i -g dsh-lark-channel
02
Start it: `dsh-lark-channel start` — a QR code appears in the terminal
03
Scan the QR code with Feishu to create the app, then DM the bot or @ it in a group

验证接入成功

Check status with `/status` to see workspace, model, session and current permission preset
Run a task in Feishu chat and confirm the agent's execution shows up as messages/cards

回滚

To stop, terminate the `dsh-lark-channel start` process (Ctrl-C); if run with `npx dsh-lark-channel@latest start`, nothing was installed globally
To remove the global package: `npm uninstall -g dsh-lark-channel`

03DSH 适配与能力边界

DSH 适配原生运行时

Native DeepSeek Harness (dsh) channel for Feishu/Lark — chat-driven agent execution; replies and approvals returned as messages and interactive cards.

Chat-driven agent execution
Natural-language tasks and DSH commands typed in Feishu/Lark chat (e.g. /status, /ws, /cd, /model)→Agent reasoning, tool calls and final answers rendered in Feishu/Lark; the active workspace and model can be switched from chat
632807d9abaf…632807d9abaf…
Human-in-the-loop approval cards
Agent questions, plan proposals and tool-call requests→Interactive cards letting you answer via single/multi-choice or text, approve/modify plans, or allow/deny tool calls
632807d9abaf…632807d9abaf…
Safe dependency build strategy
First-run dependency install of the channel→pnpm build strategy written to the profile; unapproved build scripts warned and skipped so install does not fail
writes a pnpm build strategy into the profile before installing dependencies on first runskips unapproved dependency build scripts with a warning instead of failingrecords protobufjs's print-only postinstall as skipped by name
632807d9abaf…

04适合谁？何时不该用？

适合

Teams and developers who run DeepSeek Harness (dsh) and want to dispatch and supervise agents from Feishu/Lark IM
Operators and reviewers who need to control key agent decisions (questions, plan review, tool approval) from chat
Teams that want several DeepSeek Harness agents to coordinate inside a single Feishu/Lark group

不适合

A Feishu/Lark account is required to scan the QR code and create the bot app; the channel cannot start without completing app creation via the QR scan.

05兼容性、维护与安全提示

On first run the channel writes a pnpm build strategy to the profile and skips unapproved dependency build scripts (warned, not failed); protobufjs's print-only postinstall is recorded as skipped by name, so `pnpm approve-builds` is not needed manually.
A Feishu/Lark account is required to scan the QR code and create the bot app; the channel cannot start without completing app creation via the QR scan.

2026-08-142026-08-19v0.0.7

BSD-3-Clause · actively maintained (latest release v0.0.7, 2026-08-19)

06常见问题

如何安装 dsh-lark？

全局安装渠道包 `npm i -g dsh-lark-channel`，然后运行 `dsh-lark-channel start`。终端会显示二维码，用飞书扫码完成应用创建，之后私聊机器人或在群里 @ 它即可。如果还没有 DeepSeek Harness，先运行 `npm i -g @deepseek-ai/dsh`。

需要公网服务器或回调地址吗？

不需要。README 说明无需公网服务器，也无需配置回调地址——只要用飞书扫码完成应用创建即可。

如何在聊天里管控 Agent 决策？

模型提问、计划审阅和工具审批都会回到当前聊天，可用按钮或文字作答：单选 / 多选 / 文字回答问题，批准或修改计划，允许或拒绝工具调用。

awesome-gpt-image-2

作者 freestylefly

Prompt as Code | GPT Image 2 / 2.5 提示词与案例库，530+ 个案例、20+ 套工业级模板与可复用 Skills，新增 2.5 同提示词对比专区，附完整提示词与生成记录，持续更新。

聊天与 IMJavaScript

33.7K3,238

voyager

作者 voyager-crew

Enhancement suite for Gemini, AI Studio, Claude, ChatGPT & DeepSeek — plus a prompt manager for any website, DeepSeek Harness included. / 面向 Gemini、AI Studio、Claude、ChatGPT 与 DeepSeek 的增强套件；其中的提示词管理器可用于任意网站，如 DeepSeek Harness。

聊天与 IMTypeScript

20.3K675

dsh-im

作者 xmanrui

通过扫码或机器人凭据把IM机器人接入DeepSeek Harness（支持飞书、微信、钉钉、企业微信、QQ、Slack、Telegram、Discord和WhatsApp）。 Connect IM bots to DeepSeek Harness via QR code or credentials (9 channels).

精选聊天与 IMJavaScript

1,544186

wegent

作者 wecode-ai

开源、可自托管的 AI 工作空间，用于规划、构建和交付代码、协作与自动化任务。

聊天与 IMTypeScript

864142