Welcome to DeepSeek Harness Community - Please Read Before Posting #1797
Replies: 41 comments 32 replies
|
Great project and thanks for inviting for the internal test, I build dsh-browser, a Chrome sidebar extension that lets DSH operate your browser directly—no vision capabilities required. Connect DeepSeek Harness to the Chrome tab you are already using. The model can read page content, click controls, fill forms, scroll, and navigate while preserving your login state, session, and cookies. A side panel provides the conversation UI. Project: https://github.com/Lum1104/dsh-browser Free feel to have a try~ |
|
DeepSeek has been one of the best models I've used, and now with it's own harness it's growing even stronger. Appreciate your work! |
|
很高兴参与了内部的测试,我开发了dsh-tianshu-tui,这是一个基于ansix渲染的tui界面,同时基于harenss工程增加了tdd驱动工作流、严谨的审查交付协议、智能检索路由、视觉图像桥接等机制,用于帮助开发者进行完整开发交付。项目地址:https://github.com/huiliyi37/dsh-tianshu-tui。同时我也在基于harness工作做独立的发行版本,欢迎大家到仓库下载体验。dsh 探索星辰大海! |
|
Thanks for your development and opensource work. I truly like the imagination that the Cordis architecture brings. Future I will glad to develop my repo into dsh based on these interesting Cordis sets ! |
|
Huge thanks to the DeepSeek team for open-sourcing DeepSeek Harness and its accompanying skills! 🚀 Beyond the harness framework, the open-source skills are a major breakthrough for modern LLM agents. Many current setups suffer from over-defense, heavy moderation, and redundant gating & hash validation—friction points that break autonomy and slow down execution. The skills in DeepSeek Harness strike the right balance: pragmatic, efficient, and execution-focused, giving agents real capabilities without the bloated guardrail drag. Looking forward to building on top of this! 👏 |
|
nice! |
|
[非官方] 分享两个 DeepSeek Harness 插件:主题合集 & 桌宠鲸鱼 大家好,最近在折腾 DeepSeek Harness 的时候做了两个插件,分享给有需要的人。
给 Harness 换皮肤用的。基于官方主题扩展点 ctx.theme 构建,只改外观不动逻辑——颜色、代码块、工具调用面板、终端界面等,不涉及模型行为或 Agent 逻辑。 内置 11 个主题: DeepSeek(浅色)、OLED、Dracula、Catppuccin、Tokyo Night、GitHub Dark、Solarized、Gruvbox、Nord、Synthwave '84、Cobalt2 安装: bash
一只悬浮在 Web UI 右下角的小鲸鱼,会根据 agent 运行状态做不同反应——idle 时打盹、推理时游动、完成时鳍挥舞、SOTA 时跃出水面、报错时肚皮朝上。 通过 shell.overlay 插槽实现,不触碰聊天流。可以点击、悬停、拖拽。 安装: bash 两个项目都是 MIT 协议,也会持续更新。 |
|
不完善的地方还挺多,期待修复 |
|
有无用飞书做源,直接在dsh中提问或者飞书IM中提问的插件 |
|
我用Electron给WebUI套了个壳子,制作了一个桌面版,内置了Node,即插即用,DIY了一个图标和载入画面。 |
|
欢迎帖读到。我们也在做社区配套——dsh 中文手册(14 章 + 讨论区反馈沉淀),持续把社区踩坑整理成文档,希望和官方一起把新手路径铺好:https://github.com/Electricitysheep/dsh-handbook |
项目地址:deepseek-harness-workbench-plugin 开发下来感觉界面的侧边栏做起来很费劲, 不改会话插件做侧边栏有点绕远。 最好还是放在‘会话’便签之下, 现在只能放外部。 |
|
https://github.com/988hj7tczd-oss/harness-desktop 做了一个彩色的小鲸鱼 大家不要嫌弃呀 |
|
做了一个dsh中ds用量可视化的插件,在 DeepSeek Harness Web UI 的输入框下方实时显示 token 用量、成本估算、模型与账户余额;点击展开用量可视化图表面板(参考 DeepSeek 开发者平台的用量页组织方式),全部使用零依赖 SVG 自绘,不引入任何图表库。成本支持 USD / CNY 多币种显示与实时汇率刷新(v0.3)。https://github.com/Max-Samson/dsh-usage-chart |
|
|
搓了一个非官方的桌面壳 Harness Desktop V0.1:一个不修改 Harness Core 的原生 macOS 桌面壳,为官方 DeepSeek Harness 增加安全可 DIY 的主题、壁纸、轻量动效、双语设置和长任务状态纠偏能力。https://github.com/renjianxin929-ux/deepseek-harness-desktop |
|
一键安装全部,桌面版。非官方。 https://github.com/zouzhe1/dsh-desktop 大家好,我做了一个桌面版,因为用网页太不方便了。我这个是集成安装,把所有东西一键双击搞定。支持随时更新最新版。
|
分享一个 DeepSeek Harness 的可观测插件: DSH 自带的 dsh plugin --profile web add @loongsuite/dsh-plugin@beta
export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318
dsh --profile web提示词、回复和工具参数默认不采集,需要时显式开启;插件持有私有 OpenTelemetry Provider,不替换全局 Provider;采集或上报失败会被隔离,不影响 Agent 正常执行。目前是 beta,设计讨论在 #1699,欢迎反馈生命周期映射和 GenAI 语义约定上的问题。 An observability plugin for DeepSeek Harness: @loongsuite/dsh-plugin — https://github.com/loongsuite/dsh-plugin DSH already ships dsh-session-telemetry-otel for OTLP logs. This plugin adds traces: session, agent loop, LLM calls, and tool executions become one ENTRY → AGENT → STEP → LLM / TOOL OpenTelemetry GenAI trace, with duration and token metrics, exported over standard OTLP/HTTP to any compatible backend (Jaeger, Grafana Tempo, SigNoz, Langfuse). No vendor lock-in, and no extra collector process required. It answers questions like: where did this 47-second turn go, which tool was slowest, how many tokens did the subagent burn, did the retries stall on the same step every time. Prompts, responses, and tool arguments are not captured by default. The plugin owns private providers and never replaces the global OpenTelemetry provider, and telemetry failures are isolated from agent execution. Currently in beta — design discussion is in #1699, and feedback on the lifecycle mapping and GenAI semantics is welcome. A real session exported over OTLP into a self-hosted Langfuse: four react steps, four LLM calls with per-call latency and token counts, a failed 下图是一次真实会话通过 OTLP 导出到自建 Langfuse 的效果(该截图显式开启了正文采集)。 |
项目地址: 项目介绍: 给 DSH 的极简插件套装,UI 已做 i18n(zh/en 双语)。一键安装,开箱即用,两个小插件:
|
|
[非官方] pi2dsh — 把 Pi 插件生态接到 DeepSeek Harness pi2dsh 不是把某一个 Pi 插件手工改写成 DSH 插件,而是提供一层标准 Host ABI:桥装一次,兼容的 Pi npm 包继续按原包安装和运行。 dsh plugin --profile web add pi2dsh@0.11.0
dsh plugin --profile web add <Pi-package>目前已经在真实 DSH CLI/Web 链路跑通的案例包括:给 Harness 装上识图能力、联网工具、多模型子 Agent、Pi OAuth 订阅登录、审批与跨会话记忆。仓库同时公开机器可读能力矩阵和可复现 examples,不靠只加载成功的 mock 结论。 项目:https://github.com/weijiafu14/pi2dsh 欢迎 Pi 插件作者和 DSH 用户拿真实包来测;遇到 Host ABI 缺口我们会按标准能力补,不维护一堆一次性 fork。 |
|
A huge thank you to the DeepSeek team for open-sourcing DeepSeek Harness! 🙌 I really like how Harness brings agents, tools, and sessions together, while the Cordis architecture leaves plenty of room for community extensions. I built a VS Code sidebar on top of these capabilities. 🐋 [Unofficial] dsh-vscode — DeepSeek Harness for VS Code I spend most of my development time in VS Code and have grown used to keeping assistants like Claude Code and Codex beside the editor. So I built dsh-vscode (https://github.com/Lixxx1/dsh-vscode), which lets DSH work directly with the project currently open in VS Code. It currently supports project conversations, streaming Markdown, tool cards, approval prompts, agent follow-up questions, image attachments, and separate Model and Reasoning Effort controls. 🧩 🔗 Project: https://github.com/Lixxx1/dsh-vscode Feedback, issues, and pull requests are all welcome! 🚀 |
|
非常感谢 DeepSeek 团队开源 DeepSeek Harness!🙌 Harness 对 Agent、工具和会话的整体设计很有意思,Cordis 架构也给社区扩展留下了很大的想象空间。基于这些能力,我做了一个 VS Code 侧边栏扩展。 🐋 [非官方 / Unofficial] dsh-vscode — DeepSeek Harness for VS Code 我平时主要在 VS Code 里开发,也习惯了 Claude Code / Codex 这种常驻编辑器右侧的助手,所以做了 dsh-vscode (https://github.com/Lixxx1/dsh-vscode),让 DSH可以直接围绕当前打开的项目工作。 目前支持项目会话、流式 Markdown、工具调用卡片、权限审批、Agent 追问、图片附件,以及分别选择 Model 和 Reasoning Effort。🧩 🔗 项目地址:https://github.com/Lixxx1/dsh-vscode 欢迎大家体验,也欢迎提交 Issue 和 PR~ 🚀 |
|
大家好,我自己写了个跨会话长期记忆插件 dsh-agent-memory:数据存本地 JSON,零依赖、不上云。v0.8.0 刚更新:①可解释召回——每条检索结果都附命中原因,agent 和人能看懂为什么翻出这条 ②设置里新增记忆面板,可统计/搜索/新建/编辑/删除 ③BM25 词频信号,中文长文本召回更准。 安装: |
|
大家好,我做了个 dsh-wechat-bot——DSH 设置页里的微信桥管理面板:扫码登录微信 bot、选 agent 模式(标准/PTC/极简/创造/自定义 preset)、配模型与思考强度。配合独立进程走 DSH web 会话 API,微信聊天是真会话(上下文连续、preset 插件全加载)。安装: |
|
有没有小伙伴做个 可以在网页的 deepseek上交流完 直接发到 dsh 里面去执行的插件 类似 ChatGPT 的那种【主要的网页的 Chat不要 token】 |
|
我们需要CLI哇 |
|
Yesterday, while using DeepSeek in a conversation, I encountered a strange bug. After the normal conversation ended, I closed the DeepSeek app. However, the AI continued to "talk" to itself in the background. When I opened the app again later, I found that it had sent many emojis on its own. When I scrolled up to check, I saw messages like: There was also a lot of other nonsense. After saying these things, it just started sending emojis endlessly. It was really creepy and strange. I hope this can be fixed soon! |
|
这个试验了一下,好像你发给他一些不规律的东西不管是文字还是英文,你只要多发点,他就会出现那种乱回复无限发表情包的情况
…---Original---
From: ***@***.***>
Date: Mon, Aug 17, 2026 21:23 PM
To: ***@***.***>;
Cc: ***@***.******@***.***>;
Subject: Re: [deepseek-ai/deepseek-harness] Welcome to DeepSeek Harness Community - Please Read Before Posting (Discussion #1797)
我用网页版经常出现 权限的报错,你这个有这个问题吗
—
Reply to this email directly, view it on GitHub, or unsubscribe.
Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS and Android. Download it today!
You are receiving this because you commented.Message ID: ***@***.***>
|
|
这个用的是APP版,不是网页
…---Original---
From: ***@***.***>
Date: Mon, Aug 17, 2026 21:23 PM
To: ***@***.***>;
Cc: ***@***.******@***.***>;
Subject: Re: [deepseek-ai/deepseek-harness] Welcome to DeepSeek Harness Community - Please Read Before Posting (Discussion #1797)
我用网页版经常出现 权限的报错,你这个有这个问题吗
—
Reply to this email directly, view it on GitHub, or unsubscribe.
Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS and Android. Download it today!
You are receiving this because you commented.Message ID: ***@***.***>
|
|
[非官方] dsh-share Access your DeepSeek Harness workspace from your phone/iPad — over a secure, temporary public tunnel that exists only while you use it. dsh-share is built for you to reach your own DeepSeek Harness from your own devices. It's not a sharing tool: the URL and the generated password are meant for you, not to hand out. Free, no account, no signup. dsh-share uses a Cloudflare quick tunnel (cloudflared) to expose your local dsh web server over HTTPS. No Cloudflare account, no API token, no ngrok authtoken — just click Start and you get a public https://*.trycloudflare.com URL you can open on any device. |













Uh oh!
There was an error while loading. Please reload this page.
Hi everyone - great to meet fellow Harness developers here.
DeepSeek Harness is still an early preview, and there’s plenty we need to improve. Thanks for trying it this early and bearing with the rough edges.
Whether you’re using Harness for real-world tasks, building plugins, or trying out new agent workflows, feel free to share what you’re working on in the relevant category. If something breaks or doesn’t make sense, just let us know. We read every piece of feedback.
Before posting, please keep a few things in mind:
DeepSeek Harness v0.1 is just the beginning. Many questions still don’t have a standard answer. We look forward to working with Harness developers around the world to make it better and explore the limits of intelligence together.
Hi 各位 Harness开发者,很高兴在这里认识大家。
DeepSeek Harness 还处在早期预览阶段,当前仍有许多细节有待改进和打磨。谢谢大家愿意尝试,也谢谢大家的理解和包容。
无论你正在使用 Harness 完成真实任务、开发插件,还是尝试新的 Agent 工作流、插件,都欢迎在相应板块分享你的想法和实践。遇到问题也请直接提出,我们会认真阅读每一条反馈。可以直接在对应板块分享。
发帖前,请留意以下几点:
DeepSeek Harness 目前的 v0.1 版本只是一个起点,许多问题还没有标准答案。我们非常期待和全球的 Harness 开发者一起把 Harness 做得更好,共同探索智能上限~
All reactions