dsh-prompt-studio：DSH 提示词多风格优化与发送前自动改写
dsh-prompt-studio 是 DeepSeek Harness 的提示词优化插件（v1.0.0，维护者 zywnb-2）。在对话输入框提供「优化」面板，按内置或自定义风格改写草稿，支持撤销；开启「发送前自动优化」后，模型收到改写内容而会话仍保留原话。设置页可管理风格库、优化模型与温度等，配置存于 $DSH_HOME/prompt-studio/config.json。适用于 Windows 10/11 与 DSH 桌面版 0.2.0-rc.2，推荐从 GitHub 安装 github:zywnb-2/dsh-prompt-studio#v1.0.0 并重启 DSH。
Read Moredsh-session-titler：按需为整段 DSH 会话生成标题，确认后才写入
dsh-session-titler 是 ZilongYang 维护的 DSH Host+Web 双半插件，分类为模型推理。在顶栏或会话菜单点击「生成标题」，Host 读取整段会话表面并调用 LLM 给出可编辑提案，仅「确认修改」才写入，且不注册独占的 sessionTitle provider，不影响既有自动起名。支持 npm/GitHub/link 安装，要求 DSH >=0.2.0-rc.1 且 <0.3.0-0。
Read MoreRadeon Cloud Patcher：在 DSH 里接入 AMD Radeon Cloud 与逐模型思考档位
介绍社区插件 @zhz8888/dsh-radeon-cloud-patcher：通过补丁层声明 radeon-cloud provider，由 dsh-llm-pi-ai 完成传输与 reasoning，不改 DSH 本体、默认不写 profile。涵盖安装命令、重启与 rc- 密钥配置、模型与推理等级选择、多模态与已知限制，以及卸载即删供应商的归属规则。适用 DSH >=0.2.0-rc.1，MIT 开源。
Read Moredsh-plugin-agnes-image：在 DSH 里接入 Agnes 生图工具与设置页控制台
dsh-plugin-agnes-image 为 DeepSeek Harness 提供 agnes_image 工具与本机 loopback 控制台，对接 Agnes apiHub 图像生成 API，支持文生图、图生图与多图合成。Host 半注册工具并暴露 HTTP 路由；Browser 半在设置页嵌入 Agnes Image 控制台，用于模型切换、API Key、默认尺寸比例、提示词预设与历史图库。安装需将本地包 link 进 profile 并在 dsh.profile.bundles 注册，Node ≥ 22，MIT 许可。
Read Moredsh-completion-playground：在 DSH 里直接玩 DeepSeek Completion API 与 logprobs
dsh-completion-playground 是 Zhichii 维护的 DSH 社区插件，将 DeepSeek /beta/completions 封装为 Playground Mode：全屏 Completion 编辑器、Logprob 候选分布可视化，并保留 Chat 与 Trajectory。每次 Run 为真实对话轮次，支持 max_tokens、stop 转义序列与取消。同包提供 Gemini 模式（无 system、无 tools 的 transcript 包装）。安装需 pin v0.3.0，配置 DEEPSEEK_API_KEY 及可选上游 URL。适用于在 Harness 内调试 completion 与 logprobs，安装前请核对源码、MIT 许可证与 dsh 版本约束。
Read Moredsh-git-style-zeta：为 DSH 配置 Git 提交与 PR 的提示词规范
dsh-git-style-zeta 是 zeta987 维护的 DSH 插件，通过 git-commit-instructions 与 git-pr-instructions 向系统提示注入可选的 commit 与 PR 书写说明，在 Settings 的 Git style 或 profile 的 cordis.patch.yml 中配置。它是提示词引导而非 Git 钩子，不执行 Git、不改写历史 commit。安装：dsh plugin --profile web add dsh-git-style-zeta，需 DSH 0.1.7-rc.2+；当前 npm 版本 0.3.0，MIT 许可。
Read Moredsh-model-thinking-levels：为自定义 llm-pi-ai 路由配置按模型 thinking levels
dsh-model-thinking-levels 是 Zessi-C 维护的 DSH 插件，面向自定义 llm-pi-ai 路由，为各模型配置 reasoningEfforts（thinking levels），并支持 gateway-side capability probing。SkillHub 目录标注 v1.1.0、可安装，分类为模型推理。本文基于目录页与官方简介撰写；GitHub README 暂不可读时不编造安装命令与配置示例，安装与细节以 SkillHub 与仓库文档为准。
Read Moredsh-omni-modal：为纯文本模型接上 OpenAI 兼容的生图与生视频
dsh-omni-modal 是 yzhclcl 维护的 DeepSeek Harness 插件，在设置侧栏提供 Omni-modal settings，可配置多套 OpenAI 兼容的图像与视频 endpoint，并向模型暴露 generate_image、generate_video、check_video。支持配置档切换、图像与视频独立字段、额外 JSON 参数与连通性测试；视频异步提交与轮询，任务可跨重启。安装命令为 dsh plugin --profile <profile> add github:yzhclcl/dsh-omni-modal，需 Node.js ≥18，安装后重启 DSH。插件以进程权限访问第三方 API 与本地 media 目录，使用前请核对源码与 MIT 许可证。
Read Moredsh-subagent-model-switch：为子代理会话补上头部模型切换
DSH 原生模型切换器对 subagent 会话硬禁用，子代理只能在启动时定模型。本插件由 YuMo-233 维护，在队友/子代理会话头部提供与原生同款的下拉菜单，切换在下一 turn 生效且不重开会话；选择持久化于 DSH_HOME/storages。安装可在 DSH Desktop 插件面板填写 github:YuMo-233/dsh-subagent-model-switch。MIT 许可，版本 0.1.0。
Read Moredsh-prompt-polisher：DSH 聊天框一键优化提示词
dsh-prompt-polisher 是 yuluo554 维护的 DeepSeek Harness web 插件，在 Composer 输入框增加「优化提示词」按钮，支持会话模型与 OpenAI 兼容独立 API 双通路、concise/structured 风格与竞态保护。本文介绍安装命令 dsh plugin --profile web add dsh-prompt-polisher、配置项与使用限制（引用 chips、最短长度、dsh 0.2.x）。
Read Moreshengcheng：在 DSH 里用已登录 Grok / GPT 账号生图与生视频
shengcheng 是 yueyexiayu 维护的 DeepSeek Harness 桌面插件，通过 Host 内 OAuth 凭证调用 Grok 与 GPT 图像接口，Grok 另支持 grok-imagine-video 生视频。提供单一工具 shengcheng，凭证仅读 $DSH_HOME/.credentials.yaml，不经对话泄露。安装需复制到 plugins 目录、写入 cordis.patch.yml、安装 FFmpeg 与 pnpm 依赖并重启 Host。适用已登录账号、需在工作区策略内落盘的智能体场景；GPT 真实生成尚未完成验收。
Read Moredsh-tool-subtitle：给 DeepSeek Harness 一套能改时间轴的字幕工具
yuehancn/dsh-tool-subtitle 为 DeepSeek Harness 提供 subtitle_read/write、convert、retime、check、transcribe 等七个结构化工具，将 SRT/VTT/ASS 解析与时间轴运算从正则手工改为可调用本地工具；可选配置 ASR 命令完成音视频转写。安装命令为 dsh plugin --profile desktop add github:yuehancn/dsh-tool-subtitle，兼容 dsh 0.2.0-rc.2，MIT 许可。
Read Moredsh-tool-qrcode：在 DSH 里规划容量、生成 SVG 二维码并可选读码
dsh-tool-qrcode 是 yuehancn 维护的 DeepSeek Harness 工具插件，提供 qrcode_status、qrcode_plan、qrcode_make、qrcode_read 四个工具。内置 ISO 18004 QR 编码器可零依赖输出 SVG；qrcode_plan 在出图前计算编码模式、最小版本、纠错等级与容量余量。可选配置 qrencode、zbarimg 等实现 PNG 与读码。安装命令为 dsh plugin --profile desktop add github:yuehancn/dsh-tool-qrcode，兼容 dsh >=0.2.0-rc.2，MIT 许可。
Read Moredsh-tool-podcast：在 DSH 里把一期录音做成播客
yuehancn 维护的 DeepSeek Harness 插件 dsh-tool-podcast，通过本机 ffmpeg/ffprobe 提供 podcast_status、podcast_chapters、podcast_trim、podcast_feed、podcast_normalize 五个工具。先由 silencedetect 可解释地推断章节断点，再切段与 EBU R128 两遍 loudnorm，RSS 的 enclosure length 从磁盘读取并自校验。安装：dsh plugin --profile desktop add github:yuehancn/dsh-tool-podcast。兼容 dsh 0.2.0-rc.2，MIT 许可。
Read Moredsh-tool-ocr：在 DSH 里从图片与 PDF 抽取文字
dsh-tool-ocr 是 yuehancn 维护的 DSH 工具插件，用于从图片与 PDF 抽取文本：电子版 PDF 直接读文字层，扫描件与图片走用户配置的 OCR 引擎（如 Tesseract、PaddleOCR）。提供 ocr_read 与 ocr_status 两个工具，支持页码范围、语言提示与结果 note。安装命令为 dsh plugin --profile desktop add github:yuehancn/dsh-tool-ocr，配置写在 cordis.patch.yml。MIT 许可，针对 dsh 0.2.0-rc.2 构建测试。
Read Moredsh-tool-mining：用 GitHub REST API 把仓库数字读准
dsh-tool-mining 是 yuehancn 维护的 DeepSeek Harness 工具插件，仅用 fetch 调用 GitHub REST API，提供 mining_status、contributors、velocity、busfactor、issues 五个工具。它同时报告中位数与均值、过滤 PR 与 bot、处理分页截断，适合在 Harness 中评估仓库维护风险。安装命令为 dsh plugin --profile desktop add github:yuehancn/dsh-tool-mining，建议在 cordis.patch.yml 配置 PAT。兼容 dsh 0.2.0-rc.2，MIT 许可。
Read Moredsh-tool-media：在 DeepSeek Harness 里用 ffmpeg 处理本地媒体
dsh-tool-media 是 yuehancn 维护的 DeepSeek Harness 社区插件，将本机 ffmpeg/ffprobe 封装为 media_status、media_probe、media_convert、media_frames、media_audio、media_compress 六个工具。Agent 可先结构化读取时长与编码，再转 mp4/webm、抽帧、导出 16k 音轨或按 targetSizeMb 两遍编码压缩。支持 desktop/web profile，安装命令为 dsh plugin --profile desktop add github:yuehancn/dsh-tool-media，兼容 dsh 0.2.0-rc.2 预览版，MIT 许可。
Read Moredsh-tool-invoice：把发票读成可验算、可对账的结构化数字
yuehancn/dsh-tool-invoice 是 DeepSeek Harness 社区插件，提供 invoice_status、invoice_extract、invoice_ledger 三个工具：从发票文本或文件抽取代码、号码、购销方、金额税额等字段，校验金额+税额=价税合计及大小写金额一致，并支持多票台账与重复号检测。安装命令为 dsh plugin --profile desktop add github:yuehancn/dsh-tool-invoice，可选 textCommand 对接 pdftotext 或 OCR。面向 dsh 0.2.0-rc.2 预览版测试，MIT 许可。
Read Moredsh-tool-epub：在 DeepSeek Harness 里读、造、拆 EPUB 3
dsh-tool-epub 是 yuehancn 维护的 DSH 工具插件，在进程内提供 epub_status、epub_inspect、epub_build、epub_split 四个能力，自实现 ZIP/XML，无 epubjs/jszip 依赖。支持只读体检元数据/spine/nav 与结构 warnings，从章节或单 HTML 组装 EPUB 3，或按 spine 拆回逐章文件。安装：dsh plugin --profile desktop add github:yuehancn/dsh-tool-epub，兼容 dsh 0.2.0-rc.2 预览版，MIT 许可。
Read Moredsh-tool-browser：在 DSH 里打开网页、按页面需要等待、抽正文、截图
dsh-tool-browser 是 yuehancn 为 DeepSeek Harness 提供的浏览器工具插件：browser_status 说明环境，browser_fetch 取 HTML，browser_extract 按密度抽正文，browser_screenshot 需 Playwright。可选 Playwright，失败退回静态时会用 engine 与 engineNote 标明。安装：dsh plugin add dsh-tool-browser。适合需在智能体中区分静态与渲染页、可诊断正文抽取的开发者；插件以 DSH 进程权限运行，安装前请核对源码与许可证。
Read Moredsh-tool-archive：为智能体提供确定性 tar 归档与安全解包
dsh-tool-archive 是 yuehancn 维护的 DeepSeek Harness 插件，零外部依赖实现 ustar 读写，提供 archive_status、archive_pack、archive_unpack、archive_verify 四个工具。打包按路径排序并固定元数据，产出可跨机器比对的 treeDigest 与 SHA-256 清单；解包拒绝绝对路径、路径穿越、符号/硬链接与设备节点，strict 模式下拒一项则不写盘。支持 listOnly 与 verify，配置 workDir、archiveDir、extractDir 及体积与条目上限。适合智能体工作流中可审计、可验证的目录归档；插件以 DSH 进程权限运行，启用前请查阅源码与 MIT 许可。
Read Moredsh-token-cost：在 DSH Web 里估算当前会话的 Token 费用
dsh-token-cost 是 yoggu 维护的 DSH Web 插件，读取会话与子智能体结算日志，用 shipped @deepseek-ai/dsh-llm-pi-ai 的 pi-ai catalog 与 calculateCost 展示估算费用，非账单。底部按钮显示合计，点击可查看 Main agent、Subagents 与 Total；未定价与覆盖不全会显式标记。安装：dsh plugin --profile web add 指定 v0.2.2 Git URL。需 Node>=22.19.0，pi-ai 0.99.1 测试；MIT 许可。
Read Moredsh-codex-account：在 DSH 里管理 ChatGPT / Codex 登录
dsh-codex-account 是 yoggu 维护的 DSH 0.2.x 插件，在 Plugins → Codex Account 提供 ChatGPT 登录、取消授权、登出，并在需要时将 openai-codex provider 写入 Models。凭据由随 DSH 发布的 llm-pi-ai 适配器持有，插件不读取 grant 内容。要求 Node.js ≥22.19.0，通过 dsh plugin add 指定 v0.2.0 tag 安装。v0.2.0 取代 0.1.x 自定义多账户适配器；自 dsh-codex-sign-in 迁移可保留原凭据键。
Read Moredsh-agnes-gen：在 DSH 里接入 Agnes 文生图与文生视频
dsh-agnes-gen 是 Ylhow06 维护的 DSH bundle 插件（0.1.3），为 DeepSeek Harness 注册 agnes_image 与 agnes_video 工具，对接 Agnes AI 图像/视频生成。内置跨进程 RPM 限流、429 退避及本机 ffmpeg 转 GIF，支持中国站与国际站分 Key 配置。本文基于官方 README 说明安装命令、DSH 版本适配、Key 配置与典型调用方式，并提醒安装前审阅源码与权限边界。
Read Moredsh-reply-visual：把 DSH 里的一条 AI 回复画成单页图解
dsh-reply-visual（@yfwu2020/dsh-reply-visual）是 DSH Web 社区插件：在每条 AI 回复下提供「图解」按钮，将原文交给模型生成单文件 HTML 与内联 SVG，于右侧栏沙箱 iframe 渲染。支持图解/源码/原文三视图、会话历史回放与重画。安装使用 dsh plugin --profile web add @yfwu2020/dsh-reply-visual，要求 DSH 运行时 >=0.1.7-rc。文章基于 GitHub README 说明安装、配置、缓存与安全边界，便于智能体开发者评估是否接入。
Read Moredsh-tts-live：DeepSeek Harness 句级实时 TTS 分支
dsh-tts-live 是 YevheniiMiachyn 基于上游 dsh-tts 0.4.16 的 fork，不发布 npm，通过 agent/assistant-stream 实现可选句级 live 播报、有序队列与 POST /dsh-tts/bargein 取消，并修复 custom WAV 与合成超时等问题。文章说明与上游差异、live 配置项、file: tgz 安装命令及 GPU/TTFT 等限制，便于在 DeepSeek Harness Web UI 中降低本地语音 agent 的首句等待。
Read Moredsh-ASD-100：在 DSH 中常驻 ASD-STE100 简化技术英语输出
dsh-ASD-100 是 yang1604454533-netizen 维护的 DSH 插件（v0.1.0，MIT）。开启后于每次请求的系统提示注入 ASD-STE100 简化英语规则，统一句长、语态与用词约束。通过 /asd100 on|off 与 ~/.dsh/dsh-asd-100/.active 全局开关；规则可来自 ~/.agents/skills/simple-english、内置 skills 或九行回退集。可选 config 指定 skillDir 与 stateFile。适合需长期输出 STE 风格英文文档的场景；安装以 SkillHub 与 GitHub README 为准，启用前请审阅源码。
Read Moredsh-plugin-dex-style-edit：在 DSH 里像 Codex 一样改已发消息并从此继续
dsh-plugin-dex-style-edit 是 XylaAlyx 维护的 DSH 插件，在只追加的会话日志下实现 Codex 式体验：点击用户消息下的铅笔就地改文，Enter 后在对应轮次 fork 新会话、归档原会话，工作目录不变；Composer 所选模型用于编辑后的消息。安装命令为 dsh plugin --profile desktop add github:XylaAlyx/dsh-plugin-dex-style-edit，需重启 DSH。MIT 许可，peer 依赖 cordis 与 dsh 0.2.x。
Read Moredsh-model-endpoint：把 Harness 当前模型发布成 OpenAI 兼容 HTTP 端点
dsh-model-endpoint 是社区插件，在本机或 LAN 独立监听 OpenAI 兼容 API，复用 Harness profile 的模型路由与凭据。支持 /v1/chat/completions 流式、Bearer 鉴权、CORS、设置页及可选借出 Harness 工具。客户端只需 base_url；绑定 0.0.0.0 或公网转发时需 API key。安装使用 dsh plugin add 本地路径或 tgz，默认端口 8787。
Read Moredsh-input-optimizer：发送前重写草稿，并管理对话分支
dsh-input-optimizer 是 Xxsttb 维护的 DeepSeek Harness 社区插件（模型推理）。在 Composer 模型选择器左侧提供四角星按钮，用当前对话模型以低推理强度重写未发送草稿，修正语法、歧义与步骤顺序，且不生成回答、不进会话日志。另支持已发用户消息分叉编辑、无记忆重新生成、右侧记忆树及在工具改写工作区时禁用铅笔与重生成。要求 DSH ≥0.2.0-rc.2，通过 dsh plugin add 本地路径或 pnpm pack 的 tgz 安装，设置页可配重写指令、模型路由与编辑快捷键。
Read Moredsh-llm-model-tuning：在 Models 页按模型配置推理强度与多模态输入
dsh-llm-model-tuning 是 xurunxin 维护的 DSH 客户端 bundle，在 Settings → Models 的自定义 provider 卡片上为每个模型编辑 reasoningEfforts、input 及路由级 defaultInput，写入 llm-pi-ai 段且无需重启。内置编辑器 deliberately 不提供 provider 级 reasoning 控件；本插件以 per-model 粒度避免同一路由下模型能力不一致导致的拒绝。要求 dsh 0.1.7+（configForms）且 profile 已 compose llm-pi-ai。安装使用 plugin_manager install_bundle 指向包目录绝对路径，symlink 安装支持改源码后 HMR 或 profile 重启。
Read Moredsh-prompt-inject：在 DSH 设置页按注入点与模型自定义 Prompt 注入
dsh-prompt-inject 是 DeepSeek Harness 社区插件，由 XuezuoYS 维护。它在设置页提供 GUI，按八个注入点（用户前后缀与 system 六段）配置规则，支持按 provider/model 作用域、模板字段 {{model}}/{{provider}}/{{cwd}} 与自动保存。本文基于官方 README 介绍安装命令、注入点语义、用户前后缀会永久留在会话历史的注意事项，以及 engines.dsh 版本匹配与 rules.json 配置位置。
Read Moredsh-tts：用 OpenAI 兼容后端把文本合成语音
dsh-tts 是 xuan25 维护的 DSH out-of-tree 插件，提供单一 tts 工具，经 OpenAI 兼容后端将文本合成为 wav/mp3 等音频，并支持 voices、providers、models、status、phonemize 等 verb。配置可通过 TTS_* 环境变量或 settings.yaml 的 dsh-tts 段分层覆盖。安装命令为 dsh plugin --profile <name> add dsh-tts。适合需要在智能体流水线中统一对接多种 TTS 后端的开发者；使用前请查阅源码与 MIT 许可证，并注意插件以 dsh 进程权限运行。
Read Moredsh-ark-image：在 DeepSeek Harness 里用火山方舟 Seedream 做文生图
dsh-ark-image 是 XN-H 维护的 DSH 社区插件，通过工具 ark_generate_image 调用火山方舟 Doubao Seedream 实现文生图。零依赖纯 JavaScript、无需构建；生成后自动下载到会话工作区并在对话内展示，按日期分目录保存。支持 npm 或 GitHub 安装，需配置 ARK_API_KEY 或 apiKey，默认模型 doubao-seedream-5-0-pro-260628。仅文生图，不含图生图与编辑。MIT 许可，非官方插件。
Read Moredsh-input-optimizer：在输入框里用 Harness 已有路由改写提示词
dsh-input-optimizer 是 xkjxziochu 维护的独立 DSH 插件，在输入框发送按钮旁提供提示词优化：通过 ctx.llm 复用 Harness 已配置的 LLM 路由改写草稿，确认面板展示原文与优化后对照，采纳或取消。可配置 optimize 路由、Reasoning effort、自定义 system prompt 及 0–20 轮上下文。推荐安装：dsh plugin --profile <profile> add dsh-input-optimizer。需 dsh ^0.2.0-rc.1，自 dsh-better-input 拆出，不含语音识别。MIT 0.1.0。
Read Moredsh-open-code-review：在 DSH 里按需跑 OpenCodeReview 代码评审
dsh-open-code-review 由 xinyangGL 维护，将本机 ocr CLI 的评审规格与 unified diff 接入 DeepSeek Harness，提供 ocr_review、ocr_status 与 /ocr-review。默认不自动评审，回合改文件后可点「启动代码审核」或手动触发；结论以路径:行 [严重程度] 逐条输出，并区分真干净与失败/部分结果。支持 engine=ocr（默认，按 token 计费且跑前预估）、delegate（无额外 review token）及独立只读评审子 agent。安装需 Node≥20、npm i -g @alibaba-group/open-code-review，以及 dsh plugin --profile 添加 github:xinyangGL/dsh-open-code-review 后重启。
Read MoreDSH Decision Router：用决策模型为 Harness 自动选执行模型
dsh-decision-router 为 DeepSeek Harness 提供自动选模：决策模型从 2–6 条已配置执行路由中选模型与推理档位，支持 macOS 本地 Laya/d1 与可选 System One 云端。手动选模优先，失败回退明确。安装需 Node.js ^22.19.0 或 >=24 与 pnpm，宿主已验证 DSH 0.2.0-rc.2。本文基于 v0.1.0 README 介绍安装、设置与本地 decision:serve 用法。
Read Moredsh-subagent-custom：DSH 子代理模型管控与 Jev 智能分流
dsh-subagent-custom 是 DeepSeek Harness 社区插件，在 agent/request 层硬性改写子代理的 provider、model 与 reasoningEffort，并接入 Jev 对任务类别、工作量与风险做毫秒级判型。支持插件级 mode（off/once/auto）与会话级 inherit/auto/manual 双层开关，预置五角色八类别及三点对齐派发规则。可通过 DSH 桌面端安装 npm 包 dsh-subagent-custom，或执行 dsh plugin --profile web add dsh-subagent-custom。MIT 协议，需 Node.js 22+ 与 DSH ^0.1.7-rc.2，安装后须重启宿主。
Read Moredsh-claw-identity：把 Claw 人设文件打进系统提示
dsh-claw-identity（npm 包名 dsh-agent-identity）是 DSH 治理套件 Phase 1 插件：当会话 cwd 位于 DSclaw 工作区时，将 SOUL.md、IDENTITY.md、AGENTS.md、TOOLS.md 等每轮写入系统提示；BOOTSTRAP.md 用于新建 Agent 的首轮身份问答。USER.md、MEMORY.md 由 dsh-agent-memory 注入，HEARTBEAT.md 仅存盘。依赖 dsh-agent-registry 创建的目录，Web 端可在 Claw Agent 核心 Tab 编辑。安装：dsh plugin --profile web add github:xingyingyuzhui/dsh-agent-identity，MIT 许可。
Read Moredsh-prompt-enhance：在 Composer 输入框旁一键改写提示词
dsh-prompt-enhance 是 xiex16070-jpg 维护的 DeepSeek Harness 插件，在会话输入框模型选择器左侧注册「增强」芯片，通过宿主 LLM 将草稿改写成结构化提示词并写回输入框，不新增对话轮次。支持多种改写模式、撤销与 /enhance 斜杠命令。安装推荐使用 dsh plugin --profile desktop add github:xiex16070-jpg/dsh-prompt-enhance，装后需重启 DSH。可选配置位于 ~/.dsh/prompt-enhance/config.json。
Read Moredsh-llm-multimodal：在 DSH 聊天里接入六类多模态生成工具
dsh-llm-multimodal 是 xiaokaizhou 维护的 DSH 社区插件（模型推理），在对话中提供 generate_text、generate_image、generate_video、generate_vision、generate_tts、generate_music 六个工具。除文本走 harness LlmRuntime 外，其余多基于 OpenAI 兼容 API，并从 llm-pi-ai 配置按模型 id 关键词自动发现能力。安装推荐 dsh plugin --profile web add xiaokaizhou/dsh-llm-multimodal；可与 dsh-media-studio、media-preview 组成生成与预览闭环。MIT 许可，当前资料版本 0.2.0。
Read Moredsh-clinepass：在 DeepSeek Harness 里接入 ClinePass 模型与额度
dsh-clinepass 是 xiangnan0811 维护的 DeepSeek Harness 社区插件，通过设置侧栏 ClineBot 保存 ClinePass API 与密钥，拉取 plan 模型并注册 provider clinebot，展示 5 小时/周/月额度与请求统计，支持多账号在 429/402 时轮换。本文基于官方 README 说明安装命令、配置命名空间 dsh-clinebot、密钥勿在启动 shell 导出 CLINEBOT_API_KEY 等注意点，以及 /cline ping、stats 等典型用法。
Read Moredsh-mac-keep-awake：DSH 会话运行期间阻止 macOS 空闲休眠
dsh-mac-keep-awake 是 wjvalue 维护的 DeepSeek Harness 插件：在任意 DSH 会话运行期间通过 macOS caffeinate 持有 PreventUserIdleSystemSleep，阻止空闲系统休眠，会话全部结束后释放；默认不阻止屏幕息屏，多会话引用计数共用一个断言。仅 macOS 生效。安装需 node scripts/install.mjs 后以 file:vendor/dsh-mac-keep-awake-1.4.2.tgz 装入 profile，配置写在 cordis.patch.yml。适合长时推理与 agent 任务。MIT 许可。
Read Moredsh-answer-me-with-html：用一页 HTML 回答复杂问题
dsh-answer-me-with-html 将 Answer me with HTML 打包为 DeepSeek Harness 可选 skill 插件：模型写 Markdown 草稿，am CLI 生成 CSS、布局与 SVG，页面落在 ~/.answer-me-with-html/pages/。本文基于官方 README 说明安装命令、典型问法、与默认 skill filesystem shadow 冲突及更新方式，要求 Node.js 20+，MIT 许可，维护者 wjvalue。
Read Moredsh-evidence-first-knowledge-work：把 evidence-first 知识工作技能装进 DSH
win10ogod 将 evidence-first-knowledge-work 1.1.0 封装为 DeepSeek Harness 插件（0.1.2，MIT）。插件在 assets 内 bundled SKILL.md、references 与评估资源，经 ctx.skills 注册，支持 skill 工具按描述载入或 /evidence-first-knowledge-work 显式载入。兼容 DSH Skill 接口 0.1.7-rc.2、0.2.0-rc.2、0.2.1-alpha.1。安装示例：dsh plugin --profile web add github:win10ogod/dsh-evidence-first-knowledge-work，并用 --dump-config 确认 skill-evidence-first-knowledge-work。插件不增轮次、不改工具权限；使用前请审阅源码与权限。
Read Morexiaohe-canvas-dsh：在 DSH 中驱动小禾画布做电商内容创作
xiaohe-canvas-dsh 是 wild-River2016 维护的 DeepSeek Harness 社区插件（MIT，0.1.3），通过 Canvas Agent 让小禾画布在 DSH 中可被工具调用：画布节点与布局、16 类电商模板、图片视频音频文案生成及素材库。安装使用官方命令 dsh plugin --profile web|desktop add xiaohe-canvas-dsh，需重启 DSH；Web 配置见 ~/.dsh/profiles/web/cordis.patch.yml。典型流程为 npx @deepseek-ai/dsh web、打开画布页后自然语言编排工作流。插件以当前进程权限运行，安装前宜审阅源码。
Read Moredsh-llm-workbuddy：在 DSH 里用浏览器登录 WorkBuddy，免 API Key 选模型
dsh-llm-workbuddy 是 DeepSeek Harness 的 WorkBuddy LLM provider 插件，维护者 wenh56884-wq。通过浏览器 OAuth 登录中国站或国际站，无需 API Key，即可在 web profile 的 Web UI 中使用 WorkBuddy 模型目录、流式对话、工具调用、推理等级与图片输入，并查看侧栏用量指示。安装命令为 dsh plugin --profile web add github:wenh56884-wq/dsh-llm-workbuddy，要求 dsh >= 0.1.7-rc.1。插件 MIT 授权，改编自 dsh-llm-codebuddy。
Read Moredsh-tabbit-brain：在 DSH 内通过本地网关委托 Tabbit 模型做独立推理
dsh-tabbit-brain 是 wbdxy 维护的 DSH 模型推理插件，主代理通过 tabbit_brain 等工具向本机 OpenAI 兼容 Tabbit 网关发起纯文字请求，自管 prompt 与会话历史，不创建 DSH 子 agent、不继承主对话工具与 Skill 目录。支持后台 job、多 conversation 标签、delegationStyle 引导独立审查与分析，并明确 DSH Skill 须由主代理执行后转交。安装需 Node 22.19+、部署 gateway-patch 网关，并用 dsh plugin add link 装入 profile。
Read Moredsh-model-order：在设置页调整 llm-pi-ai 路由的模型顺序
dsh-model-order 是 Wany-i 维护的 DSH 社区插件，在 Settings → Models 的 llm-pi-ai 路由卡片上提供上移、下移、置顶、置底与按名称排序，直接改写 cordis.patch.yml 中 providers 路由的 models 数组顺序。Composer 模型菜单与设置页共用同一配置源，无需第二套同步。插件无独立配置、不联网，支持 web/desktop，安装命令为 dsh plugin add github:Wany-i/dsh-model-order，需重启并硬刷新。适用于希望常用模型靠前、已声明 models 列表的用户。
Read Moredsh-lsp-bridge：为 DeepSeek Harness 接入语言服务器
dsh-lsp-bridge 是 voidmind26 维护的 DSH Bundle 插件，通过 lsp 工具把会话接到 stdio 语言服务器，支持 definition、hover、references、workspaceSymbols、diagnostics 等只读语义查询，并提供 lsp_setup 在 danger-full-access 下按 catalog 安装与配置。插件支持多语言、多根工作区与会话内服务器复用，Web/Desktop 可通过官方 CLI 安装，并在「设置→LSP」做扫描与验证。要求 Node.js≥22，版本 0.3.0，MIT 许可。
Read MoreSession Cost Pill：在 Composer 底栏实时显示会话费用
DeepSeek Harness 社区插件 Session Cost Pill（vegerot）在 Composer dock 增加 cost pill，按每一步 assistant 的时间戳区分峰谷单价，并对缓存命中/未命中输入与输出分项计价，无需修改 Harness 源码。支持 web 与 desktop profile，点击可查看费用明细。安装使用 README 中的 dsh plugin add 本地路径命令。需注意法定节假日未建模、中途换模型按最近模型计价、不展示账户级账单等限制。MIT 许可证。
Read Moredsh-speech-service：为 DSH 官方语音输入增加可拆卸的增强能力
dsh-speech-service 是 vb2250158 维护的 DSH 模型推理类插件，在官方语音输入之上提供可关闭的增强：普通输入框话筒与波形、静音 1000ms 收尾、转写后自动发送、新回复自动朗读，以及 FunASR / faster-whisper 本地 ASR 子进程识别。需先启用官方语音与模型；朗读依赖 dsh-rabiroute-agent 与 Rabi TTS。官方能力覆盖后可逐项关闭并卸载。MIT 许可，当前 v0.2.0，适配 DSH 0.2.1-alpha.1 至 0.3.0 之前。
Read Moredsh-openai-responses-web-search：为 DSH 配置 OpenAI Responses 联网搜索
介绍社区插件 dsh-openai-responses-web-search（vb2250158，模型推理）。插件为 DSH 0.2.x 的 web profile 提供可配置的 OpenAI Responses API 搜索后端，支持 baseURL、apiKeyEnv、model 与 searchProvider（model-native / openai-responses）。含官方安装命令、YAML 配置示例、原生搜索路由说明及适用注意事项。MIT 0.1.5。
Read Moredsh-gpt-tool-schema-compatibility：修复 GPT/Codex Full access 下的工具参数冲突
dsh-gpt-tool-schema-compatibility 是 vb2250158 维护的 DSH 0.2 系开源插件，通过 cordis.patch.yml 注入配置，修复 GPT/Codex 在 Full access 下的工具参数冲突。需 DSH 0.2.1-alpha.1 或兼容 0.2 版本；安装命令为 pnpm dsh plugin --profile web add github:vb2250158/dsh-gpt-tool-schema-compatibility#<commit>。MIT 许可，列表展示名 GPT 工具兼容。安装前请审阅源码与进程权限。
Read Moredsh-volume-knob：输入框旁的朗读与音量控制
dsh-volume-knob 在 DSH Web GUI 输入框工具行、模型选择框左侧提供扬声器按钮：默认从最新提问首字读到当前回复，长按可选正文起点；上滑打开页内媒体与系统输出竖式混音台。文本按 data-chat-flow 白名单抽取，优先走 dsh-tts，失败则回退 speechSynthesis。需 DSH >= 0.1.5-rc.3，npm 或 GitHub 安装后重启 dsh web。MIT 许可，维护者 underworld-oddball。
Read Moredsh-openai-live：为 DeepSeek Harness 接入 GPT-Live-1 全双工语音
dsh-openai-live 是 Travizm 维护的 DSH 模型推理类插件，提供 realtime 接缝、GPT-Live-1 适配与无密钥回放后端，解决 Harness 缺少全双工语音及 gpt-live-1 生态空白的问题。当前 pre-release，安装为 dsh plugin --profile 添加 dsh-openai-live（npm 发布后），凭证经 OPENAI_LIVE_API_KEY 注入。强调 fail-closed 委托、provider 持有轮次边界与 keyless CI 回放。MIT 许可，Node ^22.19.0 或 >=24。
Read Moredsh-plugin-local-prompt-bridge：本地 token 精算与上下文溢出压缩重试
社区 DSH 插件 local-prompt-bridge 在不修改 @deepseek-ai 包的前提下，将 context 溢出 400 改写为 CONTEXT_WINDOW_EXCEEDED 以触发官方 compaction-basic；配置 routes 后对本机 OpenAI-compat 服务做 /tokenize 步进前精算，超阈值则主动压缩再试。本文基于 README 说明 web profile 的 pnpm 安装、bundles 注册及 cordis.patch.yml 配置，并给出自测与适用注意。
Read Moredsh-opencode-go-pool：多把 OpenCode Go 订阅共用一个 provider 路由
dsh-opencode-go-pool 是 TianYa-DAO 维护的 DSH 模型推理插件，将多份 OpenCode Go 订阅 key 池化在同一条 provider 路由后。因 agent 请求 options 不可改，插件通过独占 activeRef 在每次请求前写入最健康 key，并调用 opencode.ai 官方用量接口判断 5 小时/周/月窗口；429 GoUsageLimitError 时在无 token 输出前提下 failover 并重发。本文基于 README 说明配置 route、keys、activeRef 与 apiKeyEnv 改法、日志前缀与限制，并给出 selfcheck 命令；未编造安装 CLI，启用前需审阅源码与 MIT 许可。
Read Moredeepseek-harness-wallet-patched：DSH 对话侧的钱包与用量控制中心
deepseek-harness-wallet-patched 是 DeepSeek Harness web 侧的控制中心插件，由 ThinkofRain1213 在上游 control-center 停更后维护。提供官方余额刷新、V4 峰谷环形钟、官方价格页同步、本会话与本机 365 天 token 账本、Z.ai 套餐监控、第三方分时自定义价、加密多账户切换，以及完成与低余额提醒。安装使用 dsh plugin --profile web add；费用为本机估算，安装前宜审阅源码与 MIT 许可证。
Read Moredsh-deep-opencode：在 DSH 里走 OpenCode 免费模型，无需 API Key
DeepSeek Harness 虽支持 models.dev 上的 opencode Provider，但 OpenCode 免费档仅允许来自 OpenCode 本体的请求。社区插件 dsh-deep-opencode 由 temidayoxyz 维护，启动本地 opencode serve 并提供 opencode-free 路由，使免费模型在 Harness 中可用且无需 API Key。本文基于官方 README 介绍安装命令、模型选择方式及 cordis.yml 主要配置项，并提醒插件以 dsh 进程权限运行、安装前应审阅源码与 MIT 许可证。
Read Moredsh-whale-elite-preset：鲸英模式 Agent 预设
鲸英模式（dsh-whale-elite-preset）是 SZYTree0312 维护的 DeepSeek Harness Agent 预设 bundle：精简插件行、不加载 AGENTS.md 链、关闭运行时上下文注入，并调低压缩与工具结果裁剪阈值。适合以本地读改代码为主、不需要子代理与联网的任务；复杂规划、强依赖工作区指令或长历史回溯时应换用标准或其他预设。安装后重启 dsh，在预设选择器选「鲸英模式」。MIT 许可，基线 deepseek-harness 0.2.0-rc.2。
Read Moredsh-bailian-gold：为阿里云百炼做前缀缓存优化的 DSH Agent Preset
dsh-bailian-gold（百炼成金模式）是 SZYTree0312 维护的 DSH agent preset，面向阿里云百炼 Model Studio，以 prefix cache hit rate 为核心指标。通过固定工具目录、历史只追加、动态状态放尾部，并默认 Anthropic 端点与足够大的 contextWindow，减少系统提示重写与压缩导致的前缀失效。安装：dsh plugin --profile web add github:SZYTree0312/dsh-bailian-gold，在预设中选择百炼成金模式。适用于百炼上 Qwen、DeepSeek 等多模型长会话；安装前请审阅源码与 MIT 许可证。
Read Moredsh-think-budget：给单步思考设字数预算，连续无正文时先逼出结论
dsh-think-budget 是 superSizzzz 维护的 DSH 插件，在 agent 循环中为单步思考设字数上限（默认 4000），并在连续多步无可见正文（默认 3 步）时向会话插入提醒，要求模型先用正文写出当前结论再继续。安装命令为 dsh plugin --profile web add github:superSizzzz/dsh-think-budget，阈值可在 ~/.dsh/profiles 下 cordis.patch.yml 覆盖。不额外调用模型，权限限于追加消息与可选日志文件。适用于长链路推理与工具密集场景。
Read MoreDSH Pet：把 Harness Agent 状态带到 macOS 桌面
DSH Pet 是 sundusk 维护的 macOS 14+ 桌面伴侣，含原生 app 与 dsh-moodball-status 桥接插件。桌宠（默认小雨）反映 Harness 待机、思考、授权、完成等状态，支持任务卡片、快捷键输入与截图附件；插件经 Unix Socket 或 HTTP 推送状态，输入经官方 session 接口转发。安装需先运行 DSH，推荐 clone 仓库执行 install.sh，或 dsh plugin --profile web add github:sundusk/dsh-pet 后重启 Harness。
Read Moredsh-stream-timeout-fix：放宽 undici 默认超时，避免长静默 LLM 流被误报为 TRANSPORT
介绍 DSH 宿主插件 dsh-stream-timeout-fix：在进程级替换 undici 全局 dispatcher，将默认 300 秒 body/headers 超时放宽为可配置值，避免长静默 SSE 流死于 TypeError: terminated（UND_ERR_BODY_TIMEOUT）并被归类为 TRANSPORT。说明适用自建与第三方 OpenAI 兼容端点、github 规格安装与配置项、验证方式，以及需同步调大 streamIdleTimeoutMs 的边界。插件不含重试，MIT 许可，尚未发布 npm。
Read Moredsh-code-review-graph：DSH 上的代码图谱与审查工作流
dsh-code-review-graph（v1.0.5）是 summingwang233 维护的 DeepSeek Harness 插件，复用 Code Review Graph 2.3.9 引擎，提供 30 个工具、5 个工作流、7 个斜杠技能及 Web/桌面侧栏交互图谱。兼容 DSH 0.1.5-rc.2 与 0.2.0-rc.2。安装：`dsh plugin add dsh-code-review-graph@1.0.5` 与 `npx dsh-code-review-graph@1.0.5 prepare`；侧栏可无模型建图，`/crg-explore-codebase` 等技能用于模型分析。插件以会话仓库为界本地运行，MIT 许可。
Read Moredsh-persona-forge：在 DSH 里用角色卡改写待发消息
dsh-persona-forge 是 sugarmaster666 为 DeepSeek Harness 编写的零依赖插件：在输入框选角色卡，由当前会话模型在发送前改写用户草稿的语气与框架，支持先审查或直接发出，不修改系统提示与会话配置。角色卡为 ~/.dsh/persona-cards/ 下的 YAML，可通过设置页或 AI 代填管理。Web 端推荐 dsh plugin --profile web add dsh-persona-forge 安装。适合需要可复用口吻、又希望保留技术需求与发前校验的 DSH 用户；编码场景建议保持审查模式。
Read Moredsh-composer-prompt-optimizer：在 DSH 输入框一键优化提示词
subisle 维护的 DSH 插件 dsh-composer-prompt-optimizer（GitHub：dsh-prompt-optimizer）在 Composer 工具栏提供一键提示词优化：结合最近用户轮次与会话上下文，复用当前选中模型发起单次 compaction 辅助调用，将草稿整理为目标、范围、要求、交付与验收等结构，不写入会话日志、不激活 Agent。支持 npm / tarball / 本地 / github 安装，可选 maxDraftChars、maxContextTurns 等配置。适用于常写 coding-agent 任务说明的 DSH 用户；安装前请审阅源码与 MIT 许可证。
Read Moredsh-chatgpt-plugin：在 DeepSeek Harness 里用 ChatGPT 账号走 provider 路由
Starlight-bananice 的模型推理插件 @bananiceee/dsh-chatgpt-provider，在 DeepSeek Harness 0.2.x 中通过 Sign in with ChatGPT OAuth 登录 Plus/Pro 账号，注册 chatgpt provider 并按受限 Responses API 收发请求。凭据由 Harness 托管，不经第三方代理。本文基于 README 说明安装命令、设置页登录、模型与 cordis 配置、测试连接及排障要点，并注明 Preview API 与 Plus 共享额度等限制。
Read Moredsh-image-tiler：大图分块后再交给模型，保留可读分辨率
dsh-image-tiler（0.1.6，Ssrs-Pro-Max）是 DeepSeek Harness 实验性插件，在附件 admission 阶段将超大图切成多块 image part，避免 Harness 整体缩放导致小字不可读；小图行为不变。兼容 DSH 0.2.x 与 Node 20+，推荐安装命令为 dsh plugin --profile web add dsh-image-tiler@0.1.6，配置沿用 attachment-local 限额键。本文介绍能力、安装验证、配置项与安全注意点。
Read Moredsh-sift（Sift）：DSH 当前文章的轻量参考与会话关联
Sift（@songyanglin/dsh-sift）是 Songyanglin-curious 维护的 DSH 插件，不另建 Document 工作区，在原生 Markdown/文本预览旁提供当前文章并排参考栏，支持粘贴参考、按路径持久化、导入长会话并做 User Turn 相关性定位，并与 @songyanglin/dsh-annotations 统一批注。适配 DSH 0.2.0-rc.2，npm 版本 0.3.0（MIT）。资料未给出官方一行安装命令，安装前请查 SkillHub 与源码。
Read Moredsh-skill-labels：给 DSH 技能目录加中文短标题
dsh-skill-labels 是 Siloet-phy 维护的 DeepSeek Harness 插件，在模型技能目录与输入框 / 菜单中，于每条技能的英文描述前插入 4–7 字可自定义前缀（默认【标题】原描述），英文触发描述完整保留。支持内置表、包内 labels.json、用户 ~/.dsh/skill-labels.json 与 cordis 行配置四级覆盖，保存即热更新。提供 skill_labels 工具供模型列出未标注技能并合并写入用户表。零依赖，安装命令为 dsh plugin --profile web add github:Siloet-phy/dsh-skill-labels，MIT 许可，版本 0.2.0。
Read MoreDsh-OtherAPI：在 DSH 0.1.7 下一键接入 OpenAI 兼容 API
Dsh-OtherAPI 是 ShukebtAb 维护的 DeepSeek Harness Web 插件，基于 dsh-plugin-thinking-api 改造，适配 DSH 0.1.7-rc.2 的 entry-config 设置模型。通过设置中的 OtherAPI 面板管理 OpenAI 兼容 provider、模型启用与思考模式，密钥经 credentials 写入本地 refs 而非明文进配置。安装命令为 dsh plugin --profile web add github:ShukebtAb/Dsh-OtherAPI；插件 0.1.0 已在 0.1.7-rc.2 验证，不支持 0.1.5–0.1.6。
Read Moredsh-connect-zcode：在 DSH 里复用 ZCode Coding Plan 模型额度
dsh-connect-zcode 是 Shuffle-1992 维护的 DSH 模型推理插件，从 ZCode 本地配置读取智谱 BigModel Coding Plan 凭据，注册 provider id 为 zcode 的 anthropic-messages 通道，由 DSH 直连发请求，凭据零落盘。仅支持付费 Coding Plan；Start Plan 代发通道已下线。含模型显隐、200K/1M 上下文预算与能力面板。安装：dsh plugin --profile desktop add github:Shuffle-1992/dsh-connect-zcode。需 DSH >=0.2.0-rc.2 且 ZCode 已登录 Coding Plan。
Read Moredsh-sensitive-replace：按屏蔽表在出网前替换敏感字符，规避 DeepSeek 官方 API 400
dsh-sensitive-replace（Short-Arm-Ape，v0.7.0）是 DSH 模型推理类插件：按用户维护的 sensitive-blocklist.yaml，在 agent/pre-step、tools/post-execute 与 llm/stream 三处替换敏感字符，规避 DeepSeek 官方 Content Exists Risk 400；支持屏蔽表热重载、出网前整请求清洗、400 加硬清洗与同会话重试。安装：dsh plugin --profile web add dsh-sensitive-replace。替换单向不可逆，部署前需审源码与权限。
Read Moredsh-plugin-mind-tree：把 DSH 会话事件日志变成可导航的思维树
dsh-plugin-mind-tree（思维树）是 ShOpenMind 维护的 DeepSeek Harness 插件，遍历会话事件日志识别编辑重试、批准、子代理与分叉等分支节点，在右侧栏展示可定位的纵向思维树与轮次 token 统计，支持精简/标准/完整三档削减及 Mermaid、大纲、Obsidian 画布导出；终端可通过 mind_tree 工具生成相同格式。安装推荐使用 GitHub Releases 的 .tgz 地址并完整重启 DSH，需 Node 22+ 与匹配的 @deepseek-ai/dsh 版本。
Read Morejizuo-creator-media-models：在 DSH 里单独配置图片与视频模型
jizuo-creator-media-models（@jizuo/media-models-plugin 0.3.20）是 ShikangPang 发布的 DeepSeek Harness 媒体模型插件，归类模型推理。它在 Desktop 0.2.0-rc.2 上独立提供图片/视频服务配置与聊天生成工具，不要求安装视频项目插件；安装包捆绑即作共享核心。文章说明 Desktop 与 dsh 安装地址、启停与卸载、与创作工作区的关系，以及 FFmpeg 导出前提与 chapterWorkflowEnabled 限制，并提醒插件以宿主进程权限运行、安装前需审阅源码与 MIT 许可证。
Read Moredsh-ponytail：在 DeepSeek Harness 中接入 Ponytail 的薄适配层
dsh-ponytail 是 Sepolian 维护的 DeepSeek Harness 插件，面向 DSH 0.2.0-rc.2，将 Ponytail v4.13.0 上游指令与技能以 vendor 形式接入，提供 lite/full/ultra/off 会话隔离与持久化默认。安装需 Node ^22.19.0 或 >=24，通过 Release tarball 执行 dsh plugin --profile web add。斜杠命令 /ponytail 管理模式，/ponytail-review 等触发上游一次性技能。peer 对齐 Cordis ~4.0.4；Linux Node 22 为 README 声明的测试环境。
Read Moredsh-zen-adapter：把 OpenCode Zen 免费模型接进 DeepSeek Harness
dsh-zen-adapter 是 seonzzy 维护的 DSH 模型推理插件，将 OpenCode Zen 免费模型注册为原生 provider zen-free。无需 API key 与额外依赖，从 opencode.ai 与 models.dev 动态发现免费模型目录，并在插件面板支持获取模型与可用性检测。安装后重启 DSH，在 OpenCode Zen (free lane) 分组中选模型即可。需注意上游按 IP 配额与地区限制。MIT 许可，版本 0.2.0。
Read Moredsh-model-smart-config：第三方 API 模型行的智能参数补全
DeepSeek Harness 第三方 provider 走网络 discover 时通常拿不到输入模态与 reasoning effort。dsh-model-smart-config 用 201 条带来源的有序正则规则识别 model id，在用户点击「智能匹配」后把上下文窗口、text/image 模态与思考等级填入官方模型行受控表单，不静默写库、未识别不猜值。支持 profile 五开关与可选端点探测。安装：dsh plugin --profile <profile> add 仓库 URL。MIT，Seller-1990 维护。
Read Moredsh-plugin-kit：单文件会话与图表预览插件集
Sayi 维护的 dsh-plugin-kit 包含四个 DeepSeek Harness 插件：dsh-file-session 将会话绑定到唯一文件并守卫写入；dsh-preview-kit 提供预览外壳；dsh-mermaid 与 dsh-d2 分别预览 *.mmd/*.mermaid 与 *.d2。安装需在 DSH 中逐条填写 GitHub 地址，预览需先装 preview-kit，d2 需本机 CLI。MIT 许可，面向 DSH 0.2.x 桌面端。
Read Moredsh-sensevoice-npu：在 Intel NPU 上跑 SenseVoiceSmall 的 DSH 语音转文字 Provider
dsh-sensevoice-npu 为 DeepSeek Harness 官方实验语音包新增 OpenVINO NPU 识别 provider，在不变更语音按钮与 speech API 的前提下运行 SenseVoiceSmall。插件不下载模型，需 Python 3.11–3.13 与 NPU 驱动，通过 bootstrap.ps1 建环境并以 file: 路径安装到 desktop profile。默认 NPU 推理、CPU 跑 VAD 与 fbank，首次 bucket 编译约 2.5 分钟，适合 Core Ultra 用户将重计算移出 CPU。
Read Moredsh-openrouter-spend：在 DSH 里看 OpenRouter 真实账单
dsh-openrouter-spend 是 RyabykinIlya 维护的 DeepSeek Harness Web 插件，通过 OpenRouter Management API 与 Analytics 拉取 total_usage 实扣金额，在 Composer 下方显示 session/day spend chip，popover 提供按 key、模型与时间范围的明细、预付余额与可选额度/余额警告，Settings 中配置 Management key 与刷新策略。安装命令为 dsh plugin --profile web add dsh-openrouter-spend，需 web profile 与 Management API key，MIT 许可，当前版本 0.6.0。
Read Moredsh-qoder-plugin：在 DeepSeek Harness 里用 PAT 接入 Qoder 订阅模型
dsh-qoder-plugin 是 road-kid 维护的 DeepSeek Harness 社区插件，通过 PAT 调用与 Qoder CLI 同源的模型服务。国内版 provider id 为 qoder、国际版为 qoder-global，凭证与模型目录互不混用。支持 PAT 校验保存、模型目录 live/saved/fallback、逐模型开关与上下文窗口、每日签到及输入框额度只读概览。安装命令为 dsh plugin --profile web add github:road-kid/dsh-qoder-plugin，要求 DSH >=0.1.7-rc.1 <0.3.0。插件以 DSH 进程权限运行，安装前请查阅源码与 MIT 许可证。
Read Moredsh-voice-ru：为 DSH 语音输入提供本地俄语 ASR
dsh-voice-ru 为 DeepSeek Harness 语音输入增加本地 STT 提供商 nemotron-local，弥补内置 SenseVoice 对俄语支持不足。引擎为 NVIDIA Nemotron 3.5 ASR Streaming 0.6B Q8_0 GGUF，经 transcribe.cpp v0.3.1 与 koffi FFI 在本地推理；运行时与模型首次 prepare 时从 GitHub/Hugging Face 下载并 sha256 校验。安装需 dsh plugin add github:ripls56/dsh-voice-ru、在 profile bundles 注册并重启，于 Voice input 选择 Nemotron 3.5 ASR 0.6B 与语言 Auto。适合俄英混说场景；注意磁盘约 1 GB、录音上限 120 秒及插件进程权限。
Read Moredsh-zed-provider：在 DeepSeek Harness 里复用 Zed AI 订阅额度
dsh-zed-provider 是 RGarrido03 维护的 DSH 模型推理插件，将 Zed Pro/Student 订阅通过 zed-agent 路由接入 Harness，请求经 cloud.zed.dev 流式转发。支持 Claude 与 GPT，提供 /zed-login、Models 页 Zed 卡片与 visibleModels 选择。安装：dsh plugin --profile web add dsh-zed-provider。凭据可来自登录、配置或 Zed 本地 keyring。配额由 Zed 返回 402。MIT 0.1.4，兼容 DSH ^0.2.1-alpha.2。
Read Moredsh-compaction-effort-inherit：压缩时沿用会话思考档位，保住 DeepSeek 前缀缓存
DSH 官方 compaction-basic 压缩摘要时未继承会话 reasoningEffort，DeepSeek 前缀缓存按思考档位分桶，换档会导致长上下文每次压缩全价重算。dsh-compaction-effort-inherit 在 llm/stream 上仅处理 purpose: compaction，从 session.requestHeader 继承档位，零依赖、fail-open、不修改官方插件。安装：dsh plugin --profile web add github:rezon-aki/dsh-compaction-effort-inherit，需 DSH >=0.1.7-rc.2 与 web profile。
Read MoreDSH 腾讯混元生图插件：Hy-Image 接入、本地图库与对话内出图
dsh-plugin-hunyuan-image（v1.0.0，rayyugz）是 DSH 原生混元生图插件，支持 Hy-Image-3.5-preview（最高 4K）与 Hy-Image-3.0、TokenHub 同款分辨率×宽高比选择、单次 1/2/4 张批量生成、本地图库搜索与再次生成、可配置输出目录及官方高级参数，并注册 hunyuan_generate_image 供对话调用。无原生依赖，跨 macOS/Windows/Linux。安装推荐从 Release 解压 tgz 后通过 plugin_manager install_bundle 启用，需腾讯云 TokenHub API Key。
Read Moredsh-go-game：在 DeepSeek Harness 里下圍棋
dsh-go-game 是 raymondkm2025 为 DeepSeek Harness 编写的圍棋插件：host 侧纯 JS 规则引擎支持 19/13/9 路与中国数子法，三档内置 AI 作练习对手，形勢判斷用 playout 估死子与 ownership，可选调用 Harness llm 做粤语覆盤，并支持 SGF 与 /dsh-go-game/ 独立页及 shell.overlay 面板。可选 KataGo GTP 桥与内置 AI 并列，异常时自动回退。安装需在 web profile 的 package.json 与 node_modules 手动 link 后重启 DSH；MIT 许可，Node >= 20。
Read Moredsh-jev-plugin：在 DeepSeek Harness 里批量向 Jev 提 typed 问题
dsh-jev-plugin 由 RaulLazaro 维护，将 TypeSafe Jev（System One）接入 DeepSeek Harness。Agent 通过 jev 工具对同一段 state 并行提交 noul、choice、score 等问题，获得带概率的结构化 judgement，而非生成 prose。插件在 Settings → Plugins → Jev 配置多 Provider 与 API 密钥，安装需 dsh plugin --profile web add 并重启 Harness。适合对日志、测试输出、候选文件等大量条目做语义筛选；机械匹配仍应使用 grep。MIT 许可，版本 0.2.2。
Read Moredsh-freeapi-panel：DSH 免费 LLM API 聚合面板
dsh-freeapi-panel 是 DSH Web 插件，整合 29 家正规免费 OpenAI 兼容 LLM provider，数据来自 awesome-freellm-apis。支持卡片领 key、chat/completions 测试、免信用卡筛选与按优先级故障转移；key 存本地 keys.json 且接口仅返回掩码。安装：dsh plugin --profile web add github:qianai05233/dsh-freeapi-panel，需 DSH >= 0.1.7-rc.1，重启 dsh web 后使用右下角「免费 API」入口。
Read Moredsh-token-gobbler：用 DSH 会话轨迹审计 Token 与模型表现
token-gobbler 是 Psynosaur 维护的 DSH composition 插件，读取 ~/.dsh/sessions 下 session.jsonl.zstd 与 projection 数据，按步归属模型并统计未缓存输入、输出与缓存读写等 Token 分桶，估算解码与预填充 tok/s，并对照企业与本地价目表。提供 Web 设置页与 Activity 模态（Runs、Compare、Daily 等）及 CLI 报告。安装：npx @deepseek-ai/dsh plugin --profile web add github:Psynosaur/dsh-token-gobbler 后重启 harness。MIT 许可，需 Node >= 24。
Read Moredsh-zh-thinking：用常驻 system prompt 让 DSH 模型用简体中文思考与作答
dsh-zh-thinking 是 programerror233 维护的 DSH 模型推理类插件。它向系统提示注册 zh:reasoning-language 常驻 section，使简体中文要求在每步 renderPrompt 时生效，不进入对话历史、不被 compaction 丢弃，并对所有会话、工作区与子代理生效。文章说明 AGENTS.md 方案衰减原因、安装命令 dsh plugin add dsh-zh-thinking、可选配置与验证方式，以及成本与已知边界。MIT 许可，npm 包版本 0.1.0。
Read Moredsh-reasoning-plus：为非官方 API 提供方补齐 DSH 推理等级
dsh-reasoning-plus 是 polan-prologue 维护的 DSH 模型推理类插件，面向 API 地址非官方的自定义提供方，通过实测补齐与官方同形的推理等级，并强调每一档在上游真实生效。SkillHub 标注 v0.2.0、可安装；第三方索引显示未发布 npm、无 Web client bundle。本文基于目录页与 DSH Insights 可核对信息撰写；GitHub README 暂不可读，故不编造安装命令与配置示例，安装前请查阅 SkillHub 与仓库文档，并注意插件以 dsh 进程权限运行。
Read Moredsh-llm-fidelity：参数被拒自动修正与用量统计保真
dsh-llm-fidelity 是 philipho01 维护的 DeepSeek Harness 模型推理类插件。它在 llm/stream 上监听请求：当模型服务以 400/422 明确拒绝 temperature、stop、maxTokens、reasoningEffort 等可选参数时，同请求内解析错误、修正并重发，进程内按 provider+model 缓存偏好约 1 小时；对 usage 流防清零、防重复合并，且不编造未上报数字。安装命令为 dsh plugin add dsh-llm-fidelity，兼容 DSH 0.1.0-rc.6 及 0.2.0-rc 线，可在 cordis.patch.yml 中按 id llm-fidelity 调整 maxAdaptations 等项。
Read Moredsh-prompt-file-injector：把指定文件注入 DSH 模型上下文
dsh-prompt-file-injector 是维护者 pgnqukezrdxmhjso 发布的 DSH 模型推理类插件，通过 Plugins 页配置将选定文件注入模型上下文。支持 System prompt、Every turn、AGENTS 三种模式，可设 workspace 范围、reminder 与单文件大小上限。安装命令为 dsh plugin --profile web add dsh-prompt-file-injector，不支持从 GitHub 直接安装。相对路径按对话工作目录解析，不可读或超限文件会静默跳过。使用前请审阅源码与 MIT 许可证，并注意插件以 dsh 进程权限读文件。
Read Moredsh-llm-codebuddy-power：在 DSH 里接入 CodeBuddy 全量模型与账号
dsh-llm-codebuddy-power 是 pgnqukezrdxmhjso 维护的 DSH 模型推理类插件，将 CodeBuddy（WorkBuddy）的模型切换、浏览器登录、多账号、额度环与可选会话活动上报接入 Harness Web。安装命令为 dsh plugin --profile web add dsh-llm-codebuddy-power，不支持 GitHub 直装。支持图像输入与对话内出图，并提供 codebuddy-power CLI 管理账号。使用前请阅读源码与 MIT 许可证，并知悉活动上报会向 CodeBuddy 发送 token 等元数据。
Read Moredsh-model-kit：DSH 模型参数补全、导入与定价编辑
dsh-model-kit 是 PerseusNana7mi 维护的 DSH 模型推理类插件，面向 Desktop 0.2.x，在「设置 → 模型信息助手」中从 Models.dev 与 basellm 补全模型参数与参考单价，支持预览比较、上游导入、内置覆盖与最多 100 个模型批量保存。安装命令为 dsh plugin --profile desktop add dsh-model-kit；配置经 Settings 写入 cordis.patch.yml。本文基于 GitHub README 介绍功能、安装步骤与使用注意。
Read Moredsh-typesafe-ask：在 DSH 里用 TypeSafe Jev 做结构化决策
dsh-typesafe-ask 是 DeepSeek Harness 社区插件，由 onepayzk-glitch 维护。它注册工具 typesafe_ask，将 state 与 choice、score、noul 等类型化问题发往 TypeSafe System One 模型 Jev，返回带校准概率的结构化答案，用于路由、分类、打分与置信度门禁。安装可通过 Settings → Plugin Market 搜索 typesafe-ask，或执行 dsh plugin --profile <your-profile> add dsh-typesafe-ask；API Key 在 Settings → TypeSafe / Jev 配置并保存在凭据存储。Jev 不替代主聊天模型，需访问 api.typesafe.ai。
Read Moredsh-auto-review-model-router：把 auto-review 审阅请求路由到指定模型
DSH 实验性 auto-review 默认用会话同款 provider/model 做安全审阅，高价 agent 会为每次 gated 调用多付一整笔审阅费用。OldLigant 的 dsh-auto-review-model-router 在 LlmRuntime 层识别审阅请求并改道到 reviewModel，可选 transient 重试；agent 主路由不变。本文基于 README 说明安装（profile 内 pnpm add）、cordis.patch.yml 配置与各字段含义，并提醒 API pre-stable、进程权限与源码审查。
Read Moredsh-blank-handoff：空白会话被占用时的静默转交
dsh-blank-handoff 是 OahzOb 维护的 DeepSeek Harness 插件。浏览器连接 dsh web 时会优先复用磁盘上的空白 Session；若上一次连接的进程仍持有写租约，Host 返回 session/writer-held，官方客户端虽会新建会话但先抛错，对话层会显示「当前会话已被占用」。本插件在符合接管空白 Session 且非当前选中会话等条件时，去掉 sessionId 重试并静默转交新 Session。安装使用 plugin_manager install_bundle target: github:OahzOb/dsh-blank-handoff，可通过 globalThis.__dshBlankHandoff 自检。
Read Moredsh-typewriter-blip：模型流式回答时的逐字说话音效
dsh-typewriter-blip（说话音效）是 nsswkj 维护的 DSH 模型推理类插件：在助手回答流式输出时为每个新字符播放短促音效，窗口最小化或被遮挡时仍通过 Web Audio 与双时钟排程尽量保持发声。支持柔和/打字机预设与自定义合成或音频样本，面板可调间隔、说完即停、仅回答发声等，设置存 localStorage。desktop profile 下可通过 dshmarket 或 dsh plugin add 安装，刷新页面即用。
Read Moredsh-anthropic-image-clamp：多图 Claude 请求前的图片尺寸钳制
在 DeepSeek Harness 中，会话内联图片超过 20 张时 Anthropic 会将单边最大像素限制为 2000px，易触发 messages 图片尺寸错误。host 插件 dsh-anthropic-image-clamp 包装对 api.anthropic.com/v1/messages 的 fetch，在出站前按 maxDimension（默认 2000）缩小超长边的 base64 内联图，结果确定性并可按内容哈希缓存。安装：dsh plugin --profile desktop add github:noahchalifour/dsh-anthropic-image-clamp，重启 DSH 后生效；可在 cordis.patch.yml 中配置 maxDimension 与 hosts。经 subscriptions-plugin HTTP 代理的请求可能绕过包装。MIT，Node >= 22。
Read Moredsh-shuorenhua：在 DSH 里一键把 AI 回答润成大白话
dsh-shuorenhua（说人话）是 neuneed 维护的 DeepSeek Harness 插件，在 Web 助手消息操作栏提供一键润色，通过同源 SSE 流式输出简化寒暄、套话与公文腔，并用占位符保护代码块与公式。支持 Agent 工具 shuorenhua_simplify、润色结果 Host 缓存及 cordis.patch.yml 配置 provider/model。安装命令为 dsh plugin --profile web add github:neuneed/dsh-shuorenhua，MIT 许可，版本 0.1.0。
Read Moredsh-agnes-image：在 DeepSeek Harness 里用 Agnes AI 做文生图与图生图
dsh-agnes-image 是 Naskete 为 DeepSeek Harness 开发的社区插件，通过 Agnes AI（agnes-image-2.5-flash）提供文生图、图生图与多图合成。模型调用工具 agnes_generate_image，图片保存到工作目录 generated-images 并在对话中显示。零依赖纯 JavaScript，需自备 Agnes API Key，支持命令行或图形界面安装。MIT 许可，非官方插件。
Read Moredsh-thinking-slider：Composer 里的模型与推理强度选择器
dsh-thinking-slider 由 MuFeng262 维护，将 DeepSeek Harness Composer 的模型选择器替换为「模型名·等级」展示，展开后可拖动渐变粒子或能量充能滑块切换推理强度，并支持键盘操作。等级来自 DSH 的 model.reasoning.efforts，不写死档位。插件提供官方/粒子/能量三种 UI 模式，并在设置中可为缺省 reasoningEfforts 的代理商逐个点「补四档」。要求 DSH 0.2.x，安装命令为 dsh plugin --profile desktop add github:MuFeng262/dsh-thinking-slider，装后需完全重启 Desktop。MIT 许可，无第三方 npm 依赖。
Read Moredsh-fluent-korean：让 Harness 写出更自然的韩语
dsh-fluent-korean 将 snflkd/fluent-korean 的写作规则移植为 DSH 插件，通过 Settings 在 Fluent Korean 与 non-coding 两种样式间切换，改善 agent 韩语输出的句法完整度与工程师常用表述。安装命令为 dsh plugin add mubaid/dsh-fluent-korean，变更从下一条消息生效。适用于韩语文档、注释与说明场景；略增 token，属软约束且仅影响韩语输出。MIT 许可，要求 dsh 0.2.0-rc.2。
Read Moredsh-model-ultra：为 DSH 0.2.0 提供 llm-pi-ai 提供商全生命周期管理
dsh-model-ultra 是 MrCashmere 维护的 DSH 静态 bundle 插件，兼容 >=0.2.0-rc.1，在设置页提供「模型 Ultra」界面，覆盖 llm-pi-ai 提供商 CRUD、加密密钥、远端模型发现、逐模型思考强度、OpenRouter 路由、智能路由与组合、探活观测及连通性测试，卸载时可还原 providers。本文基于官方 README 介绍安装命令、典型步骤与升级注意事项。
Read Moredsh-short-tool-ids：为 OpenAI 兼容接口缩短超长 tool-call ID
dsh-short-tool-ids 是 mrbeandev 维护的 DSH 实验性插件，针对 OpenAI 兼容 Chat Completions 拒绝超过 64 字符的 tool-call ID 导致对话失败的问题。它仅改写 pi-ai 的 openai-completions 出站请求，将超长 ID 规范为 call_ 加 SHA-256 hex，并按 provider 在 Settings → Short tool-call IDs 中默认关闭地开关；不修改已保存 session。支持 DSH 0.1.5-rc.1 至不含 0.3.0，插件 0.4.0+，Node ^22.19.0 或 >=24。安装：dsh plugin --profile web add dsh-short-tool-ids，重启 Web 后为出错 provider 开开关。
Read Moredsh-reconnect：DSH 模型请求安全重试与网关恢复
dsh-reconnect 是 mrbeandev 维护的 DeepSeek Harness 模型推理插件。它在 agent/request-error 瀑布流上，于 Provider 内置重试用尽后对 EMPTY_RESPONSE、RATE_LIMIT、TIMEOUT 等可恢复错误做指数退避重试，支持 Retry-After、会话事件与 UI 倒计时。可通过 dsh plugin --profile web add dsh-reconnect 安装，并在设置面板或 YAML 中配置 maxDelayMs、retryQuota、retryUnknown 等项。适用于 web profile 下的长任务与智能体场景。
Read Moredsh-image-edge-cap：按最长边限制发往指定模型的图片尺寸
dsh-image-edge-cap 是 mrbeandev 维护的 DSH 社区插件，在保持宽高比的前提下对所选 provider 或 model 的请求图片做最长边封顶（默认 2000 px），仅影响发往模型的副本。它解决 Anthropic 多图请求下边长超 2000 即 400、而 DSH 总像素预算无法压细长截图的问题，并提供 Settings → Image size 配置。支持 DSH 0.2.0-rc.1 至 0.3.0 前、Node ^22.19.0 或 >=24。安装：dsh plugin --profile web add dsh-image-edge-cap。
Read Moredsh-flowglass-en：DeepSeek Harness 会话执行流可视化
dsh-flowglass-en（mrbeandev/dsh-flowglass）是基于 Iwctwbh/dsh-flowglass 的纯英文 Flowglass 插件，为 DeepSeek Harness Web 客户端提供实时三泳道执行图，支持工具调用、子代理与并发会话的 Flow Zoom。包名与 Bundle ID 独立于原版中文包。要求 Node.js ≥22.19、DSH ≥0.1.5-rc.1；安装命令为 dsh plugin --profile web add dsh-flowglass-en@0.6.1，安装后需重启 Harness。插件与 Harness 同进程权限，安装前请查阅源码与 MIT 许可证。
Read Moredsh-model-sync：把推理端模型列表自动写进 DSH 设置
dsh-model-sync 是 mpetruc 维护的 DSH 模型推理类插件（v0.7.1，MIT）。它通过 llm 服务的 discoverModels，把各 llm-pi-ai provider 公布的模型镜像到 Settings → Models，减少手填 model id。激活或 Plugins 页重开 model-sync 约 1.5 s 后同步；可配置 intervalMinutes 与 prune，并用 owner 标记区分手填与自动项。同步摘要写入 lastSync 并在插件卡片展示；rejected 状态存于 storages/model-sync/state.json。endpoint 异常时跳过并告警，不拖垮整插件。
Read Moredsh-xiaoyiwork-connect：在 DSH 里复用小艺Work 桌面端本地模型代理
dsh-xiaoyiwork-connect 是 moxiaoren 维护的 DSH 模型推理插件（0.2.0，MIT，仅 Windows）。它读取已登录小艺Work 桌面端的本地 OpenAI 兼容代理（127.0.0.1:19693），在进程内通过注册表与 DPAPI 解出 proxyKey、不落盘明文，将 DeepSeek-V4、GLM、Kimi 等模型注册到 DSH 对话窗口，直接消耗小艺Work 额度而无需单独 API Key。安装：dsh plugin add dsh-xiaoyiwork-connect，需 DSH≥2.0 与小艺Work App 保持运行。
Read Moredsh-bio-galatea：在 DSH 里跑通蛋白质序列设计与结构验证
dsh-bio-galatea（@dsh-bio/dsh-bio-galatea）是 moonbowterfly 维护的 dsh-bio-genie 生态插件，提供 19 个 galatea_* 工具：ProteinMPNN/SolubleMPNN/LigandMPNN 序列设计、区域重设计、ESMFold 折叠与复折叠、界面分析与理化打分、共识排序、接触聚类、portfolio 组合选择、跨靶 budget 与 coverage 审计。支持 npm 与桌面端安装、私有 venv 自举及可配置模型目录。本文基于官方 README 说明安装命令与典型 workflow，并提醒本机能力边界与插件运行权限。
Read Moredsh-voice-input：在 DSH Web 对话里本地离线语音转文字
dsh-voice-input 是 moazzamak 维护的 DSH 社区插件，面向 web profile 的 Web GUI 对话输入框。浏览器录音后把音频交给本机 harness，用 faster-whisper（默认 CPU）或 whisper.cpp/Vulkan（可选 GPU）离线识别，文本追加到草稿且不自动发送。支持 warm 模型、静音切分下的边说边出字，以及可选的 cleanup 去口语 filler。安装需 dsh plugin --profile web add github:moazzamak/dsh-voice-input 并运行包内 python/setup.py。适合重视隐私、在本机跑 Whisper 的 DSH 用户；使用前请核对源码与 MIT 许可证。
Read Moredsh-model-latency-test：在设置里一键测各模型首 token 延迟（TTFT）
mdl-latency-test 是 Mo3ter 维护的 DSH 模型推理类插件，在设置中增加「模型测试」页，通过 llm.stream 测量每个已配置模型的首 token 延迟（TTFT，毫秒）。按提供方折叠卡片串行测试，支持 Web GUI 与 Desktop。安装：dsh plugin add github:Mo3ter/dsh-model-latency-test。每次测试会发最小真实请求并消耗少量 token，结果反映当前网络与提供方状态，宜跨时段对比。MIT 许可，Node >= 20。
Read Moredsh-vertex-gemini：在 DSH 里用 Google 账号走 Vertex AI 调用 Gemini
dsh-vertex-gemini 是 mn288 维护的 DeepSeek Harness 社区插件，在官方 npx @deepseek-ai/dsh 上加载，经 Vertex AI OpenAI 兼容端点流式调用 Gemini 3.8 Flash 等模型。认证使用 Application Default Credentials（gcloud auth application-default login），不在 DSH 中存储 API Key，费用计入配置的 GCP 项目。插件处理 Gemini 3 工具调用的 thought signature 回放，仅支持文本；通过 gh release 下载 0.1.0 tarball 安装，并在 settings.yaml 的 llm-vertex 下配置 project 与可选模型列表。
Read Moredsh-all-notify：DSH 宿主事件的原生 Windows 系统通知
dsh-all-notify（@mkasoy/dsh-all-notify）是 DeepSeek Harness 的 Windows 原生 toast 插件：宿主监听审批、提问、任务结束与 Agent 错误，经 PowerShell 与 WinRT 弹系统通知；浏览器上报当前会话与页面可见性，在前台看该会话时静音。零运行时依赖，web profile 下用官方 dsh plugin 命令安装，可通过 cordis.patch.yml 分项开关。仅 Windows，实测 DSH 0.1.7-alpha.1。
Read MoreMissher Enhance：DeepSeek Harness 独立增强包
Missher Enhance（dsh-missher-enhance 0.1.0-rc.2）是 Missher12 维护的 DeepSeek Harness 独立增强插件包，面向 Harness 0.1.5-rc.2 / Cordis 4.0.2。提供用量与费用统计、个人指令、会话管理、Piano 轮次导航、模型元数据与推理辅助、文档与会话消息、无项目会话等能力，设置入口集中在 Settings 与对话界面。安装需使用官方 release tgz 命令写入 web profile，安装前停宿主并校验 SHA-256；本预览不含 Workbench 与 Archive 原生 Desktop 桥。卸载不删用户数据。
Read Moredsh-pixmart：在 DeepSeek Harness 里做电商生图
dsh-pixmart 是 mirror9933 维护的 DSH 插件，在对话中通过 Agent 工具生成电商主图、详情图、广告图，支持风格复刻与白底图，结果落盘至 PixMart 作品库。本文基于官方 README 介绍安装命令、首次配置厂商与模型、8 个工具的分层与计费规则，以及保真需 pixmart_edit 加参考图等注意事项。版本 0.1.2，MIT 许可。
Read Moredsh-image-gen-plugin：在 DeepSeek Harness 里接入 OpenAI 兼容出图
dsh-image-gen-plugin 为 DeepSeek Harness 提供 generate_image 工具与「图片生成」设置页，对接 OpenAI 兼容的 /images/generations 与 /images/edits。支持文生图、图生图、分辨率与画质参数、文件名模板及聊天卡片预览。安装使用 dsh plugin --profile web add dsh-image-gen-plugin。维护者 miller-owo，MIT 许可。
Read Moredsh-codex-plus：把本机 Codex 订阅接入 DeepSeek Harness
dsh-codex-plus 是 mia-wong1016 维护的 DSH Web 插件，读取本机 Codex CLI 登录态，在侧边栏展示 5 小时与一周窗口额度、重置时间与重置券，支持选券一键重置。读额度时自动注册 openai-codex 路由，模型选择器出现 Codex (ChatGPT Plus/Pro)；token 不足 24 小时时向 auth.openai.com 续期并原子写回 auth.json，密钥进 DSH credential store，无需 API key。安装：dsh plugin --profile web add dsh-codex-plus。支持 Linux/Windows，Node 20+，MIT。
Read Moredsh-plugin-clm：把 DSH 上下文窗口镜像到磁盘并可编辑
dsh-plugin-clm 为 DeepSeek Harness 提供 Context Language Models 式上下文管理：将会话镜像到 .dsh/clm/LIVE_CONTEXT.md，支持裁剪压缩与 budgetTokens/reserveTokens 防护，并提供 /clm 斜杠命令与 UI 开关。本文基于官方 README 介绍安装（profile package.json 或 dsh plugin add dsh-plugin-clm）、用法与适用场景，维护者 maxeone，MIT 许可证。
Read Moredsh-entry-shaper：在入口定型 LLM 出网 payload，减轻推理块对上下文的负担
dsh-entry-shaper 是 MaudieHakimi 维护的 DSH 实验性插件（v1.0.6，MIT），在会话元素首次出网前按类别执行 op 链，默认塑形 reasoning，通过代理 wire 层保持前缀单调、减轻出口小块压缩带来的断点税。配置位于 ~/.dsh-entry-shaper.config.json，支持 inline/exec/http 扩展。文章基于 GitHub README 介绍安装、典型 policy、适用场景及 extract 偏差、日志污染等已知限制；安装前需审阅源码，插件以 dsh 进程权限运行。
Read Morejev-harness-router：带预算的 Jev 多厂家 LLM 路由
jev-harness-router（v0.2.1，MIT）是 mattlennon 为 DeepSeek Harness 0.1.7-rc.2 编写的 TypeScript 路由插件，在 economy/strong 两档下对接 DeepSeek、Kimi、智谱等 Chat Completions，支持工具调用、厂家推理回放、增量 SSE 与取消对账。路由模式含 off/shadow/auto 与固定模型，SQLite 做预算原子预留。未发 npm，需按 INSTALL.md 用 bundle 或 tgz 集成。本文基于 README 介绍依赖、init/doctor/smoke、宿主挂载与 jevRouter 任务 API，并说明验证边界与启用 auto 前校准要求。
Read Moredsh-trae-connect：把 Trae 桌面 App 模型接入 DeepSeek Harness
dsh-trae-connect 是 marsma-101 维护的 DSH 模型推理插件（v0.1.0，MIT）。它自动发现本机 Trae 国内版/国际版安装与数据目录，解密 storage.json 中的 Cloud-IDE-JWT，在本机 loopback 提供 OpenAI 兼容小门，并把 Trae remote 会话协议翻译为 DSH 可用的 trae 与 trae-intl 模型分组。支持动态模型目录、积分倍率与国际版锁定标注、1M 上下文与思考档位、令牌自动续期。未实现 function calling；安装命令以 SkillHub 与 DSH 文档为准。仅供个人学习研究，请遵守 Trae 服务条款。
Read Moredsh-qoder-connect：在 DeepSeek Harness 中复用 Qoder CN 桌面版登录与模型
dsh-qoder-connect 是 marsma-101 维护的 DSH 模型推理插件（v0.1.0，MIT）。它解密 Qoder CN 桌面 App 在 %APPDATA% 下的 Chromium OSCrypt 凭据，按 COSY 方案签名并注册 qoder 模型组，无需 OAuth。支持 dataDir/QODER_DATA_DIR、动态模型目录、tool_calls 与推理强度透传。仅支持国内版；安装请以 SkillHub 目录页为准。插件以 DSH 进程权限访问本地登录数据，使用前请审阅源码与免责声明。
Read Moredsh-session-title：让模型按你的规则改会话标题
DSH Host 已有 sessionTitle 重命名能力，但模型工具目录原先无改名工具。dsh-session-title 向模型暴露 set_session_title，规则在「设置→会话标题」或 cordis.patch.yml 中配置（正则、字节上限等），校验通过后钉住标题。默认关闭，需勾选启用；安装命令为 dsh plugin --profile web add dsh-session-title，装后需重启 dsh。维护者 ly6170，MIT，v0.1.0。
Read Moredsh-openai-account-connector：把 OpenAI 账号登录接到 DSH 统一模型目录
dsh-openai-account-connector 是 DSH Desktop 社区插件，通过官方浏览器完成 OpenAI 登录，在 ctx.llm 注册 openai-codex 路由，刷新 GPT 模型列表并支持账号侧生图导入 Harness 附件。需先安装 dsh-account-authorization@0.1.0，再安装本插件 0.1.4；不读不写 OAuth Token。本文基于 GitHub README 介绍安装命令、能力边界与已知限制。
Read Moredsh-imagegen：在 DSH 对话里用 generate_image 生图与改图
dsh-imagegen（Lujinan991）是 DeepSeek Harness 原生 Cordis 插件，非 MCP 包装。智能体通过 generate_image 完成文生图、参考改图与 PNG 蒙版局部编辑，结果写入 DSH 持久附件并在对话预览。支持 OpenAI 兼容 API、异步 task_id 轮询、adaptive 尺寸与 fallbackEndpoints。安装地址 git+https://github.com/Lujinan991/dsh-imagegen.git；兼容 DSH >=0.2.0-rc.2 <0.3.0。配置保存在 cordis.patch.yml，保存后下次调用即生效。
Read Moredsh-opencode-live-models：在 DSH 中同步 OpenCode 与 Codex 的实时模型目录
DSH 内置 pi-ai 模型目录可能滞后于 OpenCode Go、OpenCode Zen 与 OpenAI Codex。dsh-opencode-live-models 在运行时刷新三条路线的 live 目录并保留 OpenCode 会话头与 prompt-cache 策略；集成 Settings → Models 拉取与选择；在插件详情页提供 ChatGPT 与 OpenCode Console 登录，在 Settings → Account & balance 只读展示 OpenAI、Go 订阅用量与 Zen 余额。要求 DSH ≥0.1.5-rc.1，插件 ≥0.3.0。安装 spec 为 github:Luca4Don3/dsh-opencode-live-models，需主机可访问 GitHub。
Read Moredsh-cost-meter：在 Composer 下方常驻显示本次回答与会话花费
dsh-cost-meter 是 Lopncod 维护的 DeepSeek Harness 插件，在输入框下方显示「本次回答」与「本会话消耗」人民币估算，按官方峰谷价目与 assistant/message 真实 usage 四桶计价。支持 Plugin Manager 安装、可选 .dsh-cost.json 自定义费率、本地账本与只读 HTTP 状态接口。安装后需重启 DSH；仅统计启用后的轮次。MIT 许可。
Read Moredsh-work-list：DSH Web 侧边栏本地工作清单
dsh-work-list 是 liweidong1722 维护的 DSH Web 侧边栏工作清单插件（v0.2.0，MIT）。数据写入 ~/.dsh/work-list.json，支持日常/工作/学习/生活及自定义分类、富文本（ProseMirror）、拖拽排序、回收站与 JSON 导入导出。安装：dsh plugin --profile web add github:liweidong1722/dsh-work-list#v0.2.0，重启 dsh web 后在左侧栏打开「工作清单」。适合本地优先、可备份的待办场景；安装前请审阅源码与权限。
Read Moredsh-llm-auto：在 DSH 里用一条 auto 模型合并多条 provider 回退链
dsh-llm-auto 是 liuyun847 维护的 DSH 宿主插件（MIT，当前版本 0.11.1），在模型选择器注册 Auto/auto，按 routes 有序尝试多条 provider+model；单路由内在可重试错误上原地退避重试，耗尽后静默切换，对上层仍为普通模型。支持 compactWindow 定压缩点、keepThinking/breakToolLoop、ordering.mode 额度感知排序与冷却落盘，插件页可编辑回退链。安装通过 dshpm 写入 profile bundles，需按本机改 routes 并重启 DSH 使浏览器半侧生效。
Read Moredsh-freellmapi：在 DSH 里用一条 `auto` 路由免费多厂商模型
dsh-freellmapi 是 DeepSeek Harness 社区插件，在 direct 模式下无需外部网关即可通过 provider FreeLLMAPI (Auto) 与模型 auto 使用内置免费模型池。插件按请求复杂度分类并排序候选，在首 chunk 前对 429/5xx 等自动 failover；Kilo、LLM7 等 keyless 线路默认启用，可无 API Key 起步。安装可在 DSH Plugins 管理器粘贴 github:lion231226/dsh-freellmapi 或固定标签 v0.7.18。固定 model id 时不会静默换模型。非官方插件，需自行遵守各 provider 条款并在安装前审阅源码。
Read Moredsh-workspace-prompt：按工作区挂载自定义提示词
dsh-workspace-prompt 是 linbin-mk 维护的第三方 DSH 插件，以会话 cwd 为 key 保存提示词与开关，默认不注入；Web 端通过 /workspace-prompt 配置，打开工具行开关后在 agent/pre-step 注入，每会话首次经 userQuestions 确认。与 AGENTS.md 同为上下文注入形态，Config 持久化到 profile。需 DSH 0.2.1-rc.2 级 Web profile 与 Node ^22.19。安装：dsh plugin add @linbin-mk/dsh-workspace-prompt。
Read Moredsh-undici-timeout-fix：去掉 undici 300 秒 body 超时，避免本地模型 prefill 被掐断
DSH 调用本地大模型时，prefill 超过约五分钟可能因 Node undici 默认 bodyTimeout（300s）报 terminated。插件 dsh-undici-timeout-fix 在加载时 setGlobalDispatcher 禁用 undici 计时，并说明与 streamIdleTimeoutMs 的区别、安装命令、重启要求、本地 provider 看门狗配置、代理环境变量冲突及验证方式。
Read Moredsh-agnes：在 DeepSeek Harness 里接入 Agnes 图像与视频生成
dsh-agnes 是 DSH 0.2.x 社区插件，由 Lennon-lylt 维护，通过设置页配置 API Key 与模型参数，并以 bundled skill 加 lib/cli.mjs 让 agent 用 shell 完成 Agnes 文生图、图生图与文生视频。宿主侧不注册模型工具；配置存于 DSH_HOME/storages/agnes/config.json，CLI 独立读 Key 并立即下载落盘。安装请用 dsh plugin --profile web add dsh-agnes 或 file: 本地路径，勿手工拷贝 node_modules。
Read Moredsh-mimo-tts：在 DeepSeek Harness 里用小米 MiMo 朗读回复与语音输入
dsh-mimo-tts 是 lengmodkx 为 DeepSeek Harness 开发的 MiMo 语音插件（0.1.0，MIT）。TTS 与 ASR 均走小米 token-plan 端点，密钥可与 LLM 共用 credentials。支持回答自动朗读（去 Markdown 并截断长文）、Composer 点按说话、Esc/停止键打断、九种内置音色及本地 5–15 秒样本克隆。需 DSH ≥0.2.0-rc.1，安装命令为 dsh plugin --profile web|desktop add dsh-mimo-tts 并重启宿主。配置位于 $DSH_HOME/.dsh-mimo-tts/config.json。
Read Moredsh-liuyao：在 DeepSeek Harness 里一键六爻起卦与 AI 解卦
dsh-liuyao（lctfwyt）是 DeepSeek Harness 社区插件，在输入框上方提供六爻起卦面板：三钱法一键取爻、八宫纳甲装卦、卦盘卡片与本地卦例库；解卦通过「请教解卦」或对话调用模型。支持 npm 包 dsh-liuyao 或 github:lctfwyt/dsh-liuyao 安装，须匹配 desktop/web profile 并重启客户端。起卦本地完成，解卦需模型连接；卦例存于 liuyao/cases.json。传统文化参考用途，MIT 许可。
Read Moredsh-think-zh：用系统提示词固定 Agent 的中文推理
dsh-think-zh 是 DeepSeek Harness 的零依赖 bundle 插件，通过 systemPrompt section 注入高优先级语言规则，默认固定推理通道为简体中文，并可配置 target、verbatim 与自定义 instruction。可在 Harness Web 插件页或 plugin_manager install_bundle 安装，配置写在 cordis.patch.yml。本文介绍安装、配置项、日志量化验证方式及工具输出导致的已知边界。
Read Moredsh-model-refresh：从 pi-ai 上游发现与管理模型目录
dsh-model-refresh（v0.7.2）是 KouzakiUmi 维护的 DSH 模型推理插件，以 npm 最新 pi-ai 为主源补缺模型目录，支持待确认、按 provider 开关、可选 catalog 安全写入与回滚，以及通过 configEditor 手动增删模型。默认不写安装目录；写盘后需重启消费进程。本文基于仓库 README 介绍安装接入、配置与环境变量及典型使用顺序。
Read Moredsh-prompt-profiles：为 DSH 会话按名组装并封存 system prompt
@knopki/dsh-prompt-profiles 为 DeepSeek Harness 增加按会话选择的 prompt profile：将可复用 system prompt 片段（section）按顺序与 scope 组成命名 profile，在会话启动时封存进 prompt。支持 Web 编辑器、Composer chip、cordis.patch.yml 存储与 Preview。安装：dsh plugin --profile web add @knopki/dsh-prompt-profiles。要求 DSH >=0.1.7-rc.2 <0.2.2-0。MIT 许可证，维护者 knopki。
Read Moredsh-lossless-context：可溯源的 DSH 上下文压缩与检索
Kerberos255 的 dsh-lossless-context 为 DeepSeek Harness 提供分层上下文压缩与检索后端：沿用原生 compaction 事务，不删除源会话日志；将受界事件组整理为可追溯摘要 DAG 与本地 SQLite 索引，并通过会话工具 lcm_recall 支持 search、describe、expand。安装后于 Settings → Plugins → Lossless Context 启用，并在 preset 选用 dsh-lossless-context/agent。含 idle/压力压缩、模型自适应预算、可选 Dream & Memory 后 DM 轮换与 MIT 许可。
Read Moredsh-instruction-files：用工作区 Markdown 管理 DSH 指令与 Agent 配置
Kerberos255 的 DSH 插件 dsh-instruction-files（0.2.19，MIT）在不替换宿主原生 Agent 的前提下，用 IDENTITY/SOUL/AGENTS/TOOLS/USER/MEMORY 等可选 Markdown 段按工作区注入提示，下一轮模型步进热重载；支持新会话 Agent Profile 默认模型与权限。安装：dsh plugin --profile desktop add github:Kerberos255/dsh-instruction-files。MEMORY 段在私有通道有边界校验，可与 Dream & Memory、Channel Core 协作。
Read Moredsh-model-show：Composer 模型选择器显示「供应商/模型」
dsh-model-show 是 ken95573 维护的 DSH web 客户端插件（v1.0.1，MIT），在 Composer 模型选择器触发器上将展示从纯模型名改为 Provider/Model，悬停与 ARIA 同步，下拉菜单与按模型名搜索不变。需保持内置 dsh-client-ui-model-selection 启用，DSH 版本见 engines.dsh。可通过插件市场或 Desktop CLI 安装，安装后刷新页面。无配置项；长标签会省略显示。
Read Moretrae-cn-provider：在 DeepSeek Harness 中原生接入 Trae CN 模型
trae-cn-provider 是 JsonLing 维护的 DSH 模型推理插件（0.1.6，MIT），以原生 LlmAdapter 注册 trae-cn 路由，默认浏览器 OAuth 凭据留本机，可选读本机 Trae 客户端，并在宿主 web 上挂载 /trae-cn/v1 OpenAI 兼容接口。文章涵盖安装命令、凭据三档、配置项与 CLI 用法，并强调违反 Trae 条款与封号风险，安装前需读源码与 README 风险声明。
Read Moredsh-inject：DSH 请求侧上下文注入与消息重写
dsh-inject 是 john-walks-slow 维护的 DSH 插件，在模型请求前对指定角色做关键词/正则替换，按会话启用多组预设，并支持构造回合与 /inject、/inject-turn 命令。改写仅作用于请求副本，会话 UI 保持原文；原生兼容 Keysmith 规则导入，内置 unrestricted/nsfw 与 agent preset 联动。安装：npm install dsh-inject 并加入 dsh.profile.bundles。MIT 0.1.0。
Read Moredsh-ujn-image：在 DSH 里调用济大词元做文生图与图生图
dsh-ujn-image 是 Jia-HuaWu 维护的 DeepSeek Harness 社区插件，通过济南大学大模型平台 llm.ujn.edu.cn（济大词元）调用 Qwen-Image，提供 ujn_generate_image 与 ujn_edit_image 工具。在设置页填写 ujn- 开头的 API Key 后，对话内可中文文生图，或附 1–4 张参考图做图生图；结果解码落盘并写入持久图片附件。含连接测试、试跑与凭据域存储。安装需 dsh plugin add 后重启 DSH，可选配置 llm-pi-ai 的 defaultInput 以便模型读图。MIT 许可，面向 DSH 0.2.0-rc.2。
Read Moredsh-time-context：为 DSH 模型注入当前时间与上条消息时效
dsh-time-context 是 JessenReinhart 维护的 DSH 模型推理类插件，在每次 prompt 组装时通过 systemPrompt 注册表自动写入 runtime-context：含当前本地时间（日期、钟点、星期、时区、时段）及上一条用户消息的时间戳、日关系与 recency gap，并支持跨会话与重启后的 ledger 持久化。相较依赖 Skill 的用法，事实随请求直达模型，便于解析「昨晚」「刚才」等相对时间。安装侧已接入 DSH Desktop 的 dsh.profile.bundles，改配置后需重启 Desktop；可用 /time 查看当前注入块。使用前请审阅源码与 MIT 许可证，并注意插件以 DSH 进程权限运行。
Read MoreDSH Connect MiniMax Code：把本机 MiniMax Code 桌面 App 模型接入 DSH
DSH Connect MiniMax Code（dsh-connect-minimaxcode）由 JeffTsuiCUG 维护，MIT 非官方插件。它读取本机 MiniMax Code 桌面 App 登录态或通过 OAuth 设备码独立登录，在 DeepSeek Harness 注册 minimaxcode provider，动态拉取模型目录并以 Anthropic Messages 对接网关，无需在 DSH 配置 API Key。插件支持积分胶囊、用量统计与每日签到，并说明 JWT 与 OAuth 双凭据、重启 DSH 显示 UI、以及学习研究用途与接口变更风险。安装命令见 README：github 或 npm 方式，Web 需替换 profile。
Read Moredsh-gateway-models：为 OpenCode Go 与 Command Code 网关路由补齐协议、地址与模型清单
itchenshi 维护的 DSH 宿主插件 dsh-gateway-models，为 opencode-go 与 commandcode 声明 wire 协议与 baseURL，将 DeepSeek V4.1 置顶 OpenCode Go 模型列表，并从 Command Code 公开目录幂等补齐模型；为 OpenCode 请求附加 x-opencode-session（默认每会话 UUID）。支持 GUI 勾选或 CLI 从 GitHub/Release 安装，需 DSH >= 0.2.0-rc.2、Node >= 20，MIT 许可。
Read Moredsh-jev-judge：用小型 Jev 裁判分担编码智能体的例行判断
dsh-jev-judge（HunterXing）是 DeepSeek Harness 下的模型推理类插件：在 tool.admission、turn.completion 等决策点用小型 Jev 裁判回答 bounded 问题，主模型保留 token 做推理，裁决写入 ledger。本文基于官方 README 介绍安装命令 dsh plugin add dsh-jev-judge、provider 与 kernel.json 配置、决策点默认模式及 fail open 设计，并说明适用上下文紧张与扫描型输出的场景及安装前审阅源码注意事项。
Read Moredsh-plugin-screenshot：让 DSH 会话里的模型真正「看见」屏幕
dsh-plugin-screenshot 由 HsgtLgt 维护，为 DeepSeek Harness（Windows）注册 screenshot 工具，支持全屏与按标题截窗，并通过 PrintWindow 与临时前置兜底处理遮挡、最小化及 GPU 空白帧。PNG 存入会话工作区并以原生图像块回传，便于模型内嵌展示。安装方式为 git clone 后在 plugin_manager 执行 install_bundle（file: 路径），需 DSH ≥0.2.0 且模型支持图像输入。MIT 许可，版本 0.0.5。
Read Moredsh-subagent-model-director：为 subagent 按智能分级与并发分派模型
DSH 插件 dsh-subagent-model-director 向父会话注入模型档案（高/中/低智能与并发数）、复杂度判据与分配策略，并提供 subagent_assign_model 做批量并发感知分派。需 DSH≥0.1.7-rc.1，且 subagent 模型须先在全局 Model selection 中允许。含设置页、分派测试与 MIT 开源实现。
Read Moredsh-better-reasoning-effort-fish：在官方「模型」页编辑思考强度与输入模态
DSH 的 llm-pi-ai 适配器支持每模型声明 reasoningEfforts 与 input，但官方「模型」页编辑卡不暴露这两项，第三方模型在 Composer 常缺思考档位。本文介绍 Fish-under-sea 维护的 Fork 插件 dsh-better-reasoning-effort-fish（基于 HaoyueQin 上游 v0.5.2）：在模型卡片内编辑档位与模态、一键自动适配，并让出 provider-card 席位以与 dsh-web 模型能力插件共存。含 npm 安装命令、典型用法与 settings.yaml 配置项。
Read Moredsh-plugin-subscription-oauth：用已有订阅账号接入 DSH，无需 API Key
dsh-plugin-subscription-oauth 由 EricFL1998 维护，将 Kimi Code、Grok、Google Antigravity、GLM、火山 Ark、小米 MiMo、WorkBuddy 等订阅账号通过 OAuth 注册为 DeepSeek Harness 的 LLM Provider。在 Settings → Subscriptions 一次登录、查看套餐余量，模型进入会话选择器，支持 per-model 显隐与拖拽排序、多账号池化及 auto 跨 Provider 回退。凭据存本机 ~/.dsh，请求直连厂商 API。安装：dsh plugin --profile web add dsh-plugin-subscription-oauth。MIT 许可。
Read Moredsh-fal-imagegen：在 DSH 里直连 fal.ai 队列生图
dsh-fal-imagegen（v0.2.1，MIT）为 DeepSeek Harness 提供 fal.ai 原生图像能力：设置卡片配置 FAL_KEY 与默认端点，Agent 工具直接调用 queue.fal.run，支持异步 fal_generate_image、fal_get_image_task 与同步 fal_edit_image。安装命令为 dsh plugin --profile web add dsh-fal-imagegen。结果写入对话附件与本地目录，内置多组已校验的模型别名。
Read Moredsh-prompt-optimizer：在 DSH 输入框里用当前模型优化草稿
dsh-prompt-optimizer 是 EmotionG 维护的 DSH 常驻 bundle 插件，在 conversation.input.right 提供普通模式与优化模式切换。优化模式下基于当前会话上下文与会话所选模型，通过 /prompt-optimizer RPC 流式优化 composer 草稿并 setDraft 回填，支持 Alt+O 与可传播到底层传输的取消链路。本文依据 GitHub README 说明与动态插件版的差异、上下文与截断限制，以及 web profile 下 file: 依赖、dsh.profile.bundles 与 dsh plugin add 的安装步骤。
Read Moredsh-failover-continue：DSH 统一自动续跑与模型故障转移
dsh-failover-continue 是 elanski 维护的 DeepSeek Harness 社区插件，将 auto-continue 与 model-failover 合并为单一 failover-continue 命名空间。支持有序 fallback 链、模型与提供商两级熔断、Retry-After、401/403/402 路由切换、CONTEXT_WINDOW_EXCEEDED 请求级升窗、失败 turn 自动续跑与冷会话扫描，并提供 RU/EN 设置卡片。安装需 dsh plugin --profile web add，并禁用 auto-continue 等旧插件后执行 migrate-settings。MIT 0.1.0。
Read Moredsh-connect-modelscope-token-plan：把魔搭 API-Inference 额度与模型接进 DSH
介绍 eghrhegpe 维护的 DSH 插件 dsh-connect-modelscope-token-plan：在 Plugins 页内联展示魔搭官方魔粒余额与按 profile 的本地调用统计，可选将魔搭注册为 OpenAI 兼容 provider（modelscope-token-plan）及出图工具。说明 ms- 令牌配置、安装命令、三 tab 面板、对话框用量胶囊与 doctor 诊断，并列出默认关闭、本地计数口径及已知限制。
Read MoreDictation Turbo：在 DSH 里双击 Control 听写，中英文不必切输入法
Dictation Turbo 是 DSH 社区插件，在 Harness 窗口内双击 Control 录音，识别文本写入输入框。中文与英文使用 DSH 自带 SenseVoice，无需 VoiceStudio；德语等语言可选对接免费开源 VoiceStudio。安装方式为编辑 ~/.dsh/profiles/desktop/package.json 添加 file: 依赖与 dictation-turbo bundle，并 pnpm install --no-frozen-lockfile 后重启 DSH。插件仅在 DSH 内有效，MIT 许可，维护者 dajie2014。
Read MorePinMe：在 DSH 里收藏「模型 + 思考强度」并一键切换
PinMe（dsh-plugin-pinme）是 bruceyork00-a11y 为 DeepSeek Harness 开发的组合包插件，将常用的「模型 + 思考强度」保存为收藏，在输入栏下方以快捷标签展示，单击切换，并在思考强度菜单与 Model Configuration 处提供心形收藏。数据持久化在 localStorage，兼容 Session 与 ModelDirectoryResolver。推荐安装：dsh plugin --profile <profile> add dsh-plugin-pinme；亦支持 GitHub tag v0.1.5 与 tarball。MIT 许可，安装前请审阅源码。
Read Moredsh-magpie-connect：把 Magpie 网关模型接进 DSH 模型选择器
dsh-magpie-connect 是 bitxeno 维护的 DSH Web 插件，通过原生 LlmAdapter 将 Magpie 网关的 OpenAI 兼容模型注册到 DSH 模型选择器。支持模型目录同步与白名单、图片输入、思考档位及本地缓存与快照兜底。安装命令为 dsh plugin --profile web add dsh-magpie-connect，需在设置中配置带版本段的 API 地址与 Key，并按需勾选可见模型。亦可通过 cordis.patch.yml 覆盖超时与刷新等参数。
Read Moredsh-openjev-browser：用 OpenJev 做模型驱动浏览器操控，agent 不碰 DOM
dsh-openjev-browser 是 openjev-browser MCP 工具的原生 DSH（cordis）移植：本地 OpenJev-4B 读取页面并只返回关键元素，agent 不接触 DOM，上下文开销小。提供 browser_open/pick/click/type 等九项工具，schema 与 MCP 一致。需 pnpm 构建并在 cordis.patch.yml 注册；模型与 Python 需自行部署，设置面板含 agent 部署指引。MIT 0.1.0，维护者 belowthetree。
Read Moredsh-reasoning-setup：为未声明思考能力的模型补全推理档位
DSH 中「思考强度」依赖模型元数据里的 reasoningEfforts 声明，自定义 models 默认不会出现该选项。Ayelsh 的 dsh-reasoning-setup 扫描已注册 Provider，在设置页为可编辑路由勾选档位、填写 wire 值与 thinkingFormat，写入 llm-pi-ai 同一份设置文档并做乐观锁与校验。支持 dshmarket、release tarball 与 github 源码安装，需 DSH 0.1.7-rc.2+。本文介绍功能、安装命令、典型步骤与已知边界。
Read Moredsh-win-native：Windows 上的 DeepSeek Harness 正确性层
dsh-win-native 是 ax2 维护的 DeepSeek Harness 插件，面向 Windows 将编码检测/转换（GBK/GB18030/Big5/Shift-JIS）、路径合法性、netstat 解析及有边界的只读目录统计与重复文件分析封装为六个工具，返回 result/meta 回执。安装命令为 dsh plugin --profile web add dsh-win-native，装后需重启。插件无网络、不写盘，dir_usage 与 file_duplicates 仅只读且拒绝相对路径。
Read Moredsh-toolbox-bundle：一条命令为 DSH 装上 46 个离线确定性工具
dsh-toolbox-bundle 是 ax2 维护的 DSH 合装包，一次写入 dsh-deterministic-tools、dsh-jvm-doctor、dsh-py-doctor、dsh-win-native 四个插件，共 46 个纯本地、无网络工具，覆盖 JSON/YAML/CSV/编码/正则及 JVM、Python、Windows 诊断解析。安装命令为 dsh plugin --profile web add dsh-toolbox-bundle，装后需重启。工具提供结构化回执与明确边界；schema 会增加 prompt token，可按 cordis.patch.yml 裁剪。MIT 许可，Node >= 20.10。
Read Moredsh-provider-hub：在 DSH 里统一配置 BharatCode、OpenAI 兼容端点与订阅模型
dsh-provider-hub 是 arzvaak 维护的 DeepSeek Harness 模型推理类插件，在 Settings → Providers 中集成 BharatCode 预设（含模型发现与 API Key）、OpenAI 兼容端点（Chat Completions / Responses），并通过 dsh-plugin-subscriptions 嵌入 ChatGPT Codex 与 GitHub Copilot 等订阅登录。密钥写入 DSH credential store，使用原生 llm-pi-ai 适配器。安装命令为 dsh plugin --profile desktop add github:arzvaak/dsh-provider-hub，需 DSH >=0.2.0-rc.2；订阅功能需同 profile 安装 dsh-plugin-subscriptions。
Read Moredsh-onedev：让 DSH 通过 MCP 直连 OneDev 管理 REST API
dsh-onedev 是 Altermoe 维护的 DeepSeek Harness 插件，内置 dsh-onedev-mcp，经 dsh-mcp-client 将 OneDev 管理 REST API 暴露为 mcp__onedev__ 工具（61 个，含多环境、文档查询与 onedev_api_request）。支持 Token 或账号密码、只读模式，bundle 安装后可在 Settings→OneDev 配置凭据。需 Node ^22.19.0 与 DSH ^0.2.0-rc.2。本文基于 GitHub README 与 dsh-integration 文档整理安装与典型调用。
Read Moredsh-model：多家模型汇成一个 OpenAI 兼容地址
dsh-model 由 AIcivilization 维护，将 Codex、Kimi、WorkBuddy、Gemini、Groq、Ollama 等多来源模型统一到本机或 VPS 上的 OpenAI 兼容 /v1 入口，一把 key 供 dsh 与任意 OpenAI 客户端使用。支持来源开关、组合 failover、多来源自动切换、每日签到与失败/用量统计。安装：npm install -g dsh-model 后执行 dsh-model setup；在 dsh 中于设置 → dsh-model 管理。需 Node ≥ 20、dsh ≥ 0.2.0-rc.2，MIT 许可。
Read Moredsh-plugin-cost-ledger：把每次模型调用记入月度 JSONL 成本台账
dsh-plugin-cost-ledger 是 DeepSeek Harness 插件，在 accounting: own 下监听 session usage，按 DeepSeek 官方价与 token 桶计价，按月写入 JSONL。提供 /ledger 月报、/ledger-export CSV 与 ledger_query 工具；无价模型丢弃不猜价。安装需 dsh plugin add、合并 cordis.patch.yml 并重启。与 price-aware 分工为持久账本 vs 实时用量。README 注明尚未 live mount 验证；插件以 dsh 进程权限读写 ~/.dsh/cost-ledger。
Read More@ztyss/dsh-llm-provider：接管 DSH 模型服务与设置界面
DSH 插件 @ztyss/dsh-llm-provider 由 Ztyss 维护，接管 llm-pi-ai、llm-deepseek、模型选择器与 Models 设置页，提供供应商就地编辑、模型勾选保存、额度查询及可选 npm 最新 pi-ai 桥接。安装命令为 dsh plugin --profile web add github:Ztyss/dsh-llm-provider，需重启 dsh web。桥接与 Cookie 等状态落在 $DSH_HOME/llm-provider-bridge/。适合需在 Web 侧集中管理多供应商与模型的开发者；安装前请审阅源码与许可证。
Read MoreModelTester：DSH 会话页上的模型结构指纹检测面板
ModelTester（dsh-modeltester）是 Yuer6327 维护的 DeepSeek Harness 网页插件，在会话页右上角提供模型检测面板。通过本地、零网络的结构指纹匹配，对 29 家厂商候选排名并展示证据账本；含 12 项探针与一键全量测试、usage 差分定量指纹及推理健康提示。安装需 dsh CLI ≥ 0.1.0-rc.7，推荐命令为 dsh plugin --profile <profile> add dsh-modeltester。适合网关联调与灰测模型侧写，结论为「像谁」而非身份认证。
Read MoreDSH Qoder CN Connect：把 Qoder 桌面订阅模型接进 DeepSeek Harness
dsh-qoder-connect 是 YSwo-fei 维护的 DSH 模型推理插件（v0.5.0，MIT）。在 Windows 上读取 Qoder CN 桌面 App 本地凭据并自动续期，将约 14 个订阅模型注册为 provider「qoder」，支持 1M 上下文档位、联网搜索与图片上传，无需单独填 API Key。安装通过 profile 的 file 依赖与 bundle 挂载；提供 doctor、models、ask 等 CLI。非官方逆向桥接，存在协议与账号风险，使用前需审阅源码与免责声明。
Read Moredsh-jev-verify：在 DSH 里接入 Jev 决策模型与可审计护栏
dsh-jev-verify 是 xienda 维护的 DeepSeek Harness 插件（v0.7.2），将 TypeSafe Jev System One 决策模型暴露为 jev_decision、jev_verify、jev_guard_status 等工具：一次调用返回带概率的结构化答案与用量审计；jev_verify 对 live API 跑 27 题基准（2026-09-28 复测 96.3%）；可选 auto-guard 在 shell 类工具前后做黑名单与 Jev 风险/循环判断。安装需 Node≥20、dsh≥0.1.5-rc.2，命令为 dsh plugin --profile web add dsh-jev-verify，并配置 cordis.patch.yml 与 TYPESAFE_API_KEY。适合需要快判、路由与可验证护栏的智能体开发者；插件以 dsh 进程权限运行，启用前请审阅源码与 MIT 许可证。
Read Moredsh-model-proxy：按模型名走出站代理，代理不可用时直连回退
dsh-model-proxy 是 windwhiterain 为 DeepSeek Harness 提供的 bundle 插件，通过占用 undici 全局 dispatcher，从请求 body 读取 model 字段，仅将匹配 origins 与 models 模式的调用经指定 HTTP(S) 代理出站；其余请求保持插件加载前路由。支持 fallbackToDirect 在代理 TCP 不可达时回退，避免进程级 HTTP_PROXY 导致全网关走代理。本文基于官方 README 说明安装（profile package.json）、cordis.patch.yml 配置、典型场景与已知限制。
Read Moredsh-llm-quota-retry：账户额度用尽后的长期重试与路由池切换
dsh-llm-quota-retry 是 windwhiterain 维护的 DSH 模型推理类插件。当 @deepseek-ai/dsh-llm-retry 在 QUOTA 等额度失败上耗尽快速重试后，本插件接管：若 route pool 中有仍有余额的路由则按比例切换，否则在同一 turn 内按间隔（默认一小时）无上限重试直至成功或取消。支持 QUOTA/AUTH/RATE_LIMIT 等等价场景，可配置 nativeRetryFirstProviders 避免误把限流当额度用尽。安装需在仓库 checkout 旁执行 dsh plugin --profile web add；compose 须排在 dsh-llm-retry 之后，首次需重启 dsh web 以显示 composer chip。
Read Moredsh-wb2api：本机 WorkBuddy / CodeBuddy 账号池接入 DeepSeek Harness
dsh-wb2api（dsh-plugin-wb2api-ui）是 Wei894348 维护的 DeepSeek Harness 社区插件：自托管本地网关，将 WorkBuddy / CodeBuddy 账号池注册为 OpenAI 兼容 LLM provider，并提供 Web 设置面板与成长任务自动化。本文基于 GitHub README 介绍定位、核心能力，以及 dshmarket、桌面端 URL 与 web profile 命令行三种安装方式，并说明重启 dsh、数据目录与权限等注意事项。MIT 许可，Node >= 22。
Read Moredsh-jev-subagent-dispatch：用 Jev 把例行任务路由到廉价子代理模型
dsh-jev-subagent-dispatch 是 vitas 维护的 DeepSeek Harness 插件（MIT，0.8.0）。它调用 TypeSafe Jev 对任务做结构化分类，按 auto/careful 等 profile 注入 subagent 模型路由建议，主模型仍通过 harness 原生 subagent 工具交接。插件不自带委派能力，依赖 Subagent 插件与 allowlist；默认 mode 为 off，支持 /route、/jev 触发与 preview 评估。安装使用 dsh plugin --profile web add。适合希望把 mechanical、bugfix 等例行工作导向廉价子模型并可用 NDJSON verdict 校准策略的开发者。
Read Moredsh-minimax-connect：把本机 MiniMax Code 订阅接入 DeepSeek Harness
dsh-minimax-connect 由 VICqiufeeng 维护，将本机 MiniMax Code（Mavis）已登录的积分制订阅接入 DeepSeek Harness，作为 Anthropic Messages 协议的 model provider。插件读取 ~/.minimax 凭据，在客户端退出后仍可通过 OAuth refresh 自续期，并提供设置页积分、签到与模型清单。安装命令为 dsh plugin --profile desktop add "github:VICqiufeeng/dsh-minimax-connect"，重启 DSH 生效。适用已登录 MiniMax Code 的开发者；需注意进程权限、cn/Windows 验证范围及非公开 API 依赖。
Read Moredsh-plugin-smart-config：DSH 下的模型智能配置与参数适配
dsh-plugin-smart-config 由 TACONailoong 维护，将 ZCode 的推荐配置引擎逆向移植为 DeepSeek Harness Cordis 插件。通过五层级联规则、字段级覆写继承、工具与多模态守卫及推理参数映射，把 contextWindow、input、reasoningEfforts 等写入 llm-pi-ai 档案，适配 dsh 0.2.0-rc.2。内置 Revision 30 离线规则并支持云端热更新。安装命令为 dsh plugin --profile web add dsh-plugin-smart-config，亦可 npm 安装后以 SmartConfigService 独立解析模型推荐配置。
Read Moredsh-i-am-rich：在 DSH 里把每次模型请求真的发两份，并如实记浪费的 Token
介绍 DeepSeek Harness 社区插件 @studyzy/dsh-i-am-rich：每次模型请求额外派发可配置份数并丢弃结果，第二份为真实计费调用；侧边栏按 provider usage 展示今日、本月、累计浪费 Token，账本存于 ~/.dsh/i-am-rich/。涵盖 npm 安装命令、cordis.patch.yml（id 须为 i-am-rich）、enabled/discardedCopies/fortune 配置，以及千万富翁与亿万富翁两档与缓存计费差异。适合在真实 harness 中实验双份调用与用量统计的开发者，安装前需知会真实增加 API 费用。
Read Moredsh-token-gauge：DSH Web 界面上的实时 Token 仪表盘
dsh-token-gauge（@stone100010/dsh-token-gauge）是 stone100010 维护的 DSH Web 客户端插件，在 shell.overlay 上显示 Token 累计、tok/s 与估算费用三块可拖拽表盘。数据仅来自 tokenUsage、sessionStats 等已有 projection，只读且不改变对话。适用 DSH 0.1.5-rc.2 与 web profile，推荐安装命令为 dsh plugin --profile web add @stone100010/dsh-token-gauge。费用刻度需在 lib/client.js 中按网关单价调整。MIT 许可，npm 版本 0.1.1。
Read Moredsh-adhd-session-mode：按会话切换的 ADHD 友好输出形态
dsh-adhd-session-mode 由 StarterMonk 维护，将 ayghri/i-have-adhd 规则移植到 DeepSeek Harness，通过 system prompt 与 /i-have-adhd 命令在单会话内开关输出形态（动作优先、编号步骤、估时），状态存于会话日志投影而非全局文件。支持 composer 徽章、按需 Skill 与 alwaysOn 配置。适合多会话并行时「执行态」与「细讲态」分开；子 agent 不继承模式。MIT 1.0.0，Node >=20。
Read Moredsh-our-free-model-fish：在 DSH 里接入多路免费模型与本地网关
dsh-our-free-model-fish 是上游 dsh-our-free-model 的社区 fork（v2.2.0，MIT），由 Fish-under-sea 维护。插件将匿名免费通道、Kilo 免费池及十三个账号渠道并入 DSH 模型选择器，并提供 OpenAI/Anthropic 兼容本地转发。本改版新增渠道分组可见性总开关与按像素面积的视觉计价。安装可用 dsh plugin add 本地目录；桌面端宜走应用内插件管理。升级须用 npm i dsh-our-free-model-fish@latest，勿点应用内上游升级。
Read Moredsh-pdf-reader：在 DSH 里按页读 PDF、本地 OCR 与可视化裁剪
dsh-pdf-reader 是 SkanderBog 维护的 DeepSeek Harness 功能插件（v0.1.4），在 DSH Desktop 中通过 pdf_status、pdf_inspect、pdf_read、pdf_search、pdf_render 实现页级阅读、字面量搜索、选择性本地 OCR 与整页/裁剪图附件。依赖系统 Poppler 与可选 Tesseract，要求 Node 22.19+ 或 24+、Harness ≥0.2.0-rc.2。安装需从 GitHub Release 下载 dsh-pdf-reader-0.1.4.tgz，经桌面 Plugins 页或 dsh plugin add 启用。插件以 DSH 进程权限运行，不修改源 PDF；DeepSeek 官网聊天不加载此插件。
Read Moredsh-context-pilot：长会话上下文用量可见，压缩与思考档位交给模型
dsh-context-pilot 是 Shuffle-1992 维护的 DSH 宿主插件（面板名「上下文智能压缩」），面向长 Agent 会话的窗口压力：每轮注入上下文用量，模型通过工具在智能压缩线自主决定是否压缩并自动续跑，强制压缩线在 pre-step 无条件执行；压缩在下一步前落地，本轮不中断、不伪造用户消息。可选智能思考向 Agent 暴露思考强度档位并允许工具换档。安装命令为 dsh plugin --profile web add github:Shuffle-1992/dsh-context-pilot，需重启 DSH，无 npm 包。当前 v0.1.61，MIT 许可。
Read Moredsh-code-review：在 DeepSeek Harness 里按需审查未提交改动
dsh-code-review 是 SeanTolstoyevski 维护的 DSH bundle 插件，在会话中用 /review 让 reviewer 模型审计工作区未提交改动，并以对话卡片呈现报告。内置 cr、arc 模式，可自定义 modes；默认与主 agent 隔离，支持 session 范围、忽略路径与 focus 消息。安装通过 plugin_manager install_bundle 指向 github:SeanTolstoyevski/dsh-code-review，重启 dsh web 后使用。MIT 许可，版本 1.5.0。
Read Moredsh-plugin-chatgpt：用 ChatGPT 订阅在 DSH 里当模型提供方
dsh-plugin-chatgpt 是 ckanner 维护的 DSH 社区插件，通过 Sign in with ChatGPT 完成 OAuth 登录，无需 API Key，在 harness 中注册 chatgpt 提供方路由。模型列表、上下文与 reasoning 档位来自账号 listing；可在 Settings → Models 登录并启用模型，可选 contextWindowLimit 与 modelContextWindowLimits 调整上下文预算。安装命令为 dsh plugin --profile <name> add dsh-plugin-chatgpt，安装后需重启 profile。适合已有 ChatGPT 订阅、希望在 DSH 内用订阅额度推理的开发者；使用前请查阅 GitHub 源码、MIT 许可证及已知限制。
Read MoreCture Image：在 DSH 对话里直连 OpenAI 兼容生图与修图
Cture Image（@cikian/dsh-cture-image）是 Cikian 维护的 DSH 0.2.x 插件，通过 OpenAI 兼容的 /images/generations 与 /images/edits 在对话中完成文生图、本地图编辑与批量出图，结果内联展示并写入工作区 saved-images。支持 Plugins 页配置与 imagegen_config、generate_image 工具。安装使用 dsh plugin add @cikian/dsh-cture-image，MIT 许可，版本 1.0.0。
Read Moredsh-prompt-lab：在 DSH 输入栏把草稿改写成 Agent 结构化提示词
dsh-prompt-lab（提示词工坊）是 xueccci 维护的 DeepSeek Harness 社区插件，在 Web composer 提供「优化」入口，将草稿改写成 role/task/context/constraints/format 结构。支持普通优化、精简、扩写三种模式，以及仅本地、hybrid、仅 DSH、云端 API 四种引擎；确认弹窗含五层诊断与输出分，支持继续优化与显式写回。需 DSH >= 0.2.0-rc.2，安装命令为 dsh plugin --profile web add github:xueccci/dsh-prompt-lab#v0.2.0。MIT 许可，与 DeepSeek AI 无隶属关系。
Read Moredsh-opencode-free：在 DSH 中直接使用 OpenCode Zen 免费模型
dsh-opencode-free 是社区维护的 DSH 插件，通过 PiAiAdapter 直接调用 OpenCode Zen 推理端点，在模型选择器的 OpenCode Zen Free 分组中提供流式、reasoning 与 tool calls。默认匿名访问，可选 config.apiKey 或 OPENCODE_API_KEY。安装需 DSH 0.2.1-rc.2 与 dsh-opencode-free@0.3.6 精确配对。插件非官方，会做进程级 HTTP 拦截并发送 CLI 格式身份头；详情页可探测可用性但消耗配额。本文基于 GitHub README 整理安装、配置与安全注意事项。
Read Moredsh-set-model：让 Agent 按任务自主切换模型与思考深度
dsh-set-model 是 DeepSeek Harness 下的模型推理类插件，由 john-walks-slow 维护。它通过可选的 TypeSafe Jev 每轮决策或 set_model、switch_preset 等工具，让根 Agent 在声明的模型档位与 reasoning 强度间切换，并把当前模型注入运行时上下文。支持 policyPrompt 策略、Plan 模式自动升降级、Provider 白名单与上下文窗口护栏，以及 Web 设置页「动态模型」。启用方式为 profile 中 id 为 set-model 的 loader 配置，README 未给出单独 dsh plugin add 命令。
Read Moredsh-hyperworkbuddy: DSH 设置里直接管理 WorkBuddy 多账号并对接 OpenAI 兼容接口
`dsh-hyperworkbuddy` 插件旨在解决 DeepSeek Harness (DSH) 使用 WorkBuddy 时多账号管理与模型集成的痛点。该插件允许在 DSH 中直接管理 WorkBuddy 账号，并支持将模型对接至任意 OpenAI 兼容客户端。 插件核心功能丰富，包括对外接口独立开关、API Key 管理与自动续期、国内与国际版隔离、故障转移与指数退避、会话亲和以保留缓存，以及详细的用量看板。安装后，用户可在设置面板管理账号、端点与模型，并通过本地回环地址在各类客户端中使用。插件复用 WorkBuddy 登录状态，零配置即可启用。
Read MoreDeepSeek Harness 模型分发插件
该插件为 DeepSeek Harness (DSH) 增强了模型管理能力，允许用户在统一界面中为主会话、子代理及计划模式分别配置独立的模型和推理强度。核心功能包括输入栏右侧的芯片控制、统一的模型搜索菜单、实时生效的推理滑块，以及计划模式下的自动切换与会话覆盖机制。该插件需配合 DSH 0.2.0-rc.2 使用，安装后需完全重启应用。
Read Moredsh-model-retry-settings：DeepSeek Harness 模型请求重试次数控制
`icrefin/dsh-model-retry-settings` 是 DeepSeek Harness 的一个可视化插件，旨在解决手动编辑配置文件调整重试次数的痛点。该插件在 Web 客户端的 Settings 页面新增了“Request retries”卡片，允许用户直接设置每个提供商的 `maxRetries`（0-50）。 安装时需注意必须使用打包的 `.tgz` 文件，并依赖 `dsh-llm-retry` 插件才能生效。设置会动态更新且仅影响 normal 模式，无需重启即可生效。
Read MoreIntroduction to the DeepSeek Harness Kilo Gateway Plugin
dsh-llm-kilo-gateway is a DeepSeek Harness (DSH) plugin designed to automatically discover and integrate free models available on Kilo Gateway, eliminating the need for manual model list maintenance. It supports automatic refreshing, change notifications, tool calling, multimodal capabilities, and token usage statistics, while providing secure key management. Users can install it via the CLI and configure the API Key, retry strategy, and refresh interval on the settings page. Currently, it supports 16 free models, but users should be mindful of data privacy, rate limiting, and risks associated with dynamic model changes.
Read Moredsh-jev-thinking: Automatically Determine Thinking Depth for Prompts
The `dsh-jev-thinking` plugin is designed to automatically set an appropriate `reasoningEffort` parameter for DeepSeek Harness (DSH) agents. It determines the reasoning intensity by querying the Jev model and is licensed under the MIT License. Core features include reading routing options, single-turn prompting, and context completion that incorporates the previous response. After installation, a TypeSafe API Key must be configured, with a default timeout of 2500 ms. When using it, note limitations such as context dependency, relatively high token consumption, potential increased costs due to routing support, and the absence of a fallback strategy.
Read Moredsh-pi-catalog-sync
The `dsh-pi-catalog-sync` plugin is designed to address the outdated model list for the `llm-pi-ai` provider in DeepSeek Harness (DSH). It synchronizes the real-time model catalog from pi.dev to DSH via the official API, with control through both the CLI and Web UI. It supports intelligent filtering, hybrid protocol handling (such as companion routing), and scheduled refreshes. The configuration is flexible, allowing customization of synchronization strategies and companion routing. The plugin runs only at startup or refreshes at defined intervals, and automatically skips when conflicts or specific configurations are encountered.
Read MoreDeepSeek Harness (DSH) Plugin: dsh-vertex-argon
`dsh-vertex-argon` is a DeepSeek Harness plugin tailored for the Google Gemini 4 Argon model, supporting dual-channel access via Google Vertex AI and the Gemini API. The plugin resolves the HTTP 400 error in the built-in adapter, natively supports three thought levels, thought summaries, v1beta1 long-text continuation, and multi-turn tool-call signing, and automatically sanitizes invalid parameters. Users can install it via CLI or a command, configure the Vertex AI or Gemini API channel as needed, and adjust parameters such as thought level, long decoding, and context window on the settings page.
Read Moredsh-compact-manager: Hierarchical Compaction Threshold Manager for DeepSeek Harness
dsh-compact-manager is a DeepSeek Harness plugin designed to address premature triggering caused by an overly low default compression threshold. It provides a three-level override strategy spanning global, context-length tier, and model scopes, allowing users to customize the trigger formula (such as a ratio, output reserve, or fixed value). Through the sidebar, the plugin enables real-time viewing and editing of configurations and supports applying presets, primarily for finely controlling the timing of context compression and preventing premature loss of historical information.
Read Moredsh-commandcode-goat: Integrate Command Code Subscriptions as an llm-pi-ai Route
The DeepSeek Harness plugin dsh-commandcode-goat unifies Command Code's Go/GOAT/Pro/Max subscriptions under DSH's `llm-pi-ai` namespace, supporting single-credential management, web search, and usage display across multiple tiers. It is compatible with DSH >= 0.1.7-alpha.2, installs via the `link:` command, automatically syncs routing, and preserves user manual configuration. After installing and adding an API Key, all tiers can be managed uniformly in DSH.
Read MoreDSH ZCode Connector: Driving the ZCode Desktop App as a Model Channel via Chrome DevTools Protocol
DSH ZCode Connector is a plugin that drives the ZCode desktop application GUI via the Chrome DevTools Protocol (CDP) to address the issue that API keys cannot be directly extracted from account-bound quotas. It provides a local OpenAI-compatible endpoint and simulates real user interactions to have ZCode perform request signing and authentication, enabling tools such as DeepSeek Harness (DSH) to use these quotas through a local proxy. The plugin includes a status monitoring panel and MCP server functionality, supporting views for quota usage, subagent status, and more. Its design emphasizes security (tokens are read locally only) and a single-conversation session model.
Read Moredsh-opencode-go-model-list: Fix opencode-go Model List Issue on DeepSeek Harness
The `opencode-go` routing model list in DeepSeek Harness (DSH) often lags behind upstream. The `dsh-opencode-go-model-list` plugin is designed to address this by directly contributing models to the built-in catalog, keeping the list synchronized. It supports extending the routing list, background cache refresh, and offline access. After installation, DSH must be restarted for the plugin to take effect, and it requires a pnpm or PowerShell environment. This tool eliminates cumbersome manual configuration and makes it easier for users to access the latest models.
Read MoreDSH Qoder Connect: Integrating Qoder’s Subscription Model
`masknull/dsh-qoder-connect` is a DeepSeek Harness (DSH) plugin designed to unify access to both the domestic and international product lines of Qoder. It features independent dual-variant configuration, real-time PAT validation, three-tier model catalog fallback, one-click context window switching, batch model toggle management, daily automatic check-ins, and sidebar display of quota and price multipliers. Installation requires DSH version 0.1.7-rc.1 or later, and users assume the risks associated with their own accounts.
Read Moredsh-bottom-info-bar: Replaces the bottom status bar with real-time balance and spending view
Based on the native statistics line feature in DeepSeek Harness (DSH), the `dsh-bottom-info-bar` plugin replaces it with a detailed data bar showing provider, model, balance, subscription quota, peak/off-peak pricing, and session cost. It supports switching between full and compact modes, as well as multilingual support. The installation command is `dsh plugin --profile web add dsh-bottom-info-bar`; a restart is required for the changes to take effect. Data is stored locally, and uninstalling does not automatically clean up the ledger.
Read Moredsh-xingtu-skills: DSH Plugin Pack with 26 Production-Grade AI Agent Skills
The DeepSeek Harness (DSH) plugin `dsh-xingtu-skills` is maintained by user `xingtu1996` and aims to address the standardization of skill distribution and management. The plugin encapsulates 26 production-grade AI agent skills into a standard plugin, featuring four core modules: token compression, minimalism and debt management, engineering practices, and content and media review. It adopts a zero-dependency design, can be installed with a single `dsh plugin add` command, and automatically loads skills through a progressive disclosure mechanism. The plugin has passed rigorous static checks and functional validation. It is suitable for DSH developers who need unified management of development workflows, code refactoring, and content review.
Read Morexby-detect: DeepSeek Harness (DSH) Detection Toolkit
xby-detect is a DSH plugin maintained by xby-skill, providing rich visual detection and recognition capabilities under the MIT License. To install it, run `dsh plugin --profile web add github:xby_skill/xby-detect` and configure your API key. The plugin covers a broad range of capabilities, including detection of general objects, vehicles, and pedestrians, as well as safety equipment such as reflective vests and hard hats; recognition in vertical domains such as insects, plants, animals, and birds; and scenario detection for gestures, smoking, and flames. It also provides YOLO26-based models for object detection, oriented object detection, instance segmentation, human pose estimation, and image classification. Inputs support image URLs, BASE64, or file paths, and calling the API outputs detection boxes,
Read Moredsh-model-extended: Configure Thinking Intensity and Input Modality Independently per Model Row
The DSH (DeepSeek Harness) plugin `dsh-model-extended` moves thinking intensity and input modality configuration from the connection level to the model row level, allowing users to configure each model individually. The plugin supports UI checkboxes, explicit declaration of no thinking capability, pre-request validation, and custom provider field mapping. After installation, restart `dsh` and refresh the page for the changes to take effect. The configuration uses YAML; note that YAML 1.2 parsers may handle boolean values (such as `off`) differently.
Read Moredsh-submodel-change: Subagent Model Selector
The DeepSeek Harness plugin `dsh-submodel-change` lets users manually choose a model when the AI invokes subagents, addressing the need to distinguish between reasoning and summarization models in complex tasks. The plugin has zero dependencies and works only for `provider: spawn`. It supports GUI popup selection, reusing the previous selection, and nested inheritance. Installation requires restarting Desktop and supports automatic scripts or bundle methods. Configuration options allow customizing the reuse strategy and adding models. By using official extension points to intercept requests and inject model routing into the root session popup, the plugin enables fine-grained control over subtask execution under multi-model strategies.
Read Moredsh-oh-my-claude: Using Claude Code CLI as the LLM Provider for DSH
The DSH platform plugin `dsh-oh-my-claude` allows users to directly reuse an existing local Claude Code CLI login, using it as a DSH backend model provider without needing to apply for a separate API key. By invoking the `claude -p` command, it inherits all of the CLI's configurations, hook mechanisms, and rate-limiting policies. It supports live model lists, session resume, permission-approval relay, multimodal support, interactive panels, and remote SSH workspaces. Installation requires DSH 0.2 and a logged-in CLI state, making it well suited for developers accustomed to using the CLI and enabling seamless integration between DSH and Claude Code.
Read MoreDeepSeek Harness Subscription OAuth Plugin
dsh-plugin-oauth-subs is a DeepSeek Harness plugin designed to unify subscriptions for multiple AI services, including ChatGPT Codex, Grok, and GLM, into a single interface. It supports various OAuth login methods, local proxy, and quota tracking, allowing users to directly invoke these paid models in DSH after installation for local management and cost estimation.
Read Morerelay-dsh-plugin-codex: Using Codex in DeepSeek Harness
`relay-dsh-plugin-codex` is a plugin that provides a native Codex backend for DeepSeek Harness (DSH). It runs through the Codex App Server, allowing users to invoke Codex models seamlessly while maintaining a smooth DSH interface, and preserves DSH's session management, workspace, editor, and tooling capabilities. Installation requires Node.js 22.13 and pnpm, and users must log in via the Codex client before use. The plugin is currently in developer preview, with known issues including feature discrepancies and occasional empty local command events.
Read Morexby-ocr-vehicle-license: Vehicle Driving License Recognition OCR Plugin
The `xby-ocr-vehicle-license` plugin provided by DeepSeek Harness (DSH) is designed to meet the need for vehicle license recognition in agent-based document processing. It supports automatic orientation detection and filtering of primary and secondary pages, and can identify key information such as the license plate number and owner. It provides three standardized interfaces (file URL, Base64, and file path) to support different use cases. Developers must first obtain and configure an API key, after which they can call the relevant interfaces in the DSH process to perform recognition. The plugin is released under the MIT License, relies on external services, and requires stable network access and proper file read permissions.
Read Morexby-insect-recognition: Insect and Arthropod Recognition
The `xby-insect-recognition` plugin of the DeepSeek Harness (DSH) provides insect recognition capabilities for agents, supporting the identification of arthropod names and taxonomic information. Installation requires adding a GitHub repository via the CLI, and an API key must be obtained by registering on the Xiaobenyang official website before use. The plugin offers three recognition methods: through image URL, Base64 encoding, or local file path. Note that this plugin depends on an external API, runs with the permissions of the current process, and is licensed under the MIT License.
Read MoreDeepSeek Harness Plugin: Self-Evolving Deterministic Rule Objects Before Model Calls
`sra-research/self-evolving-router-dsh` is a plugin in the DeepSeek Harness (DSH) ecosystem that provides a mechanism for evolving deterministic rule objects before model calls. It runs with `dsh` process privileges, so permission isolation and security should be carefully considered. As the license is not explicitly stated, the source code should be reviewed before installation to ensure compliance. The plugin is listed in the SkillHub community and is not officially affiliated with DeepSeek.
Read Morevision-exp-tile: DSH Lossless Tiling and Intelligent Recognition Plugin for Large Images
This plugin is a DeepSeek vision enhancement tool for the DSH environment, designed to address detail loss caused by large image inputs. It losslessly splits large images into 800×800 tiles with annotated coordinates and directly calls the DeepSeek API for aggregated recognition. The plugin supports strategies such as intelligent orchestration, local OCR, batch processing, and region cropping, and includes an optimized configuration interface. Installation requires manually mounting the source code and depends on the current Profile, with no dependency on third-party plugins.
Read Moredsh-advisor
`dsh-advisor` is a DeepSeek Harness plugin ported from `rpiv-advisor`, designed to provide agents with a zero-argument `advisor()` tool that assists the primary model in decision-making. It forwards the full conversation history to a review model and returns either a plan or a stop signal. The plugin is disabled by default and must be enabled by configuring `provider` and `model`. It supports three levels of configuration through the plugin line, Web settings, and the `/advisor` command, allowing the primary model to invoke the tool on demand without altering the existing control flow.
Read Morehivanpan-star/dsh-plugin-chat-enhance Plugin Introduction
The DeepSeek Harness (DSH) plugin hivanpan-star/dsh-plugin-chat-enhance is designed to provide conversation enhancement features for DSH. The plugin belongs to the SupaNexus ecosystem, does not currently list specific features, and no official installation command is provided. Users are advised to refer to the catalog page or GitHub for guidance. Because no license information is provided, users should independently review the source code and permission details before use.
Read Moredsh-resume-turn: Resume Interrupted Replies
`dsh-resume-turn` is a plugin for DeepSeek Harness designed to address the loss of already-generated content when an endpoint is interrupted. By automatically collecting partial output and injecting a visible guiding instruction, it enables the model to resume from the point of interruption rather than restarting from scratch. The plugin supports multiple failure codes and offers flexible configuration, significantly reducing compute waste and making it suitable for agent deployment in slow or unstable network environments.
Read MoreIntroducing the DeepSeek Harness (DSH) Plugin: dsh-clinebot
`dsh-clinebot` is a native plugin specifically designed for DeepSeek Harness (DSH), aimed at addressing pain points in ClinePass integration, including dynamic model discovery, complex configuration, and quota management. It automatically synchronizes model lists, supports secure API key storage, visual management, and automatic switching between multiple accounts. The plugin provides quota monitoring and alerts, the session telemetry command `/cline`, and intelligent failover, helping developers manage third-party API resources reliably and intuitively.
Read Moredsh-token: DSH Local Token Usage Statistics Plugin
dsh-token is a local token usage dashboard plugin for DeepSeek Harness, designed to track costs and consumption. By reading local session logs, it provides four-part token composition analysis (including cache savings), supports global filtering, session drill-down, and budget monitoring. The interface offers multi-dimensional charts, and the plugin runs entirely locally without relying on online services, helping users precisely manage inference costs.
Read MoreDSH Codex Adapter: Leveraging ChatGPT’s Coding and Image Generation Capabilities in DeepSeek Harness
The DeepSeek Harness (DSH) Codex Adapter is a plugin that seamlessly integrates ChatGPT Codex capabilities into the DSH environment. By adopting the adapter pattern, it reuses the ChatGPT backend and supports streaming responses, image generation and editing, web search, and account management. The plugin aims to provide a unified entry point and reduce the barrier for developers accessing ChatGPT capabilities. It requires DSH 0.1.5 and Node.js 22.19+.
Read MoreDSH Codex Subscription: Use Your ChatGPT / Codex Subscription Directly in DeepSeek Harness
The DSH Codex Subscription plugin allows users to sign in to their ChatGPT account via OAuth and use Codex subscription models directly in DeepSeek Harness without manually managing API keys. The plugin supports quota monitoring, search, image generation, and custom context window settings. To use it, users must install the plugin via the terminal and restart DSH, then sign in with an account that has Codex access in the settings.
Read Moredsh-llm-gate: DeepSeek Harness Single-Model Provider Concurrency Gating Plugin
`dsh-llm-gate` is a plugin for DeepSeek Harness (DSH), designed to address the issue that underlying single-slot model services (such as `llama.cpp`) cannot handle concurrent requests, causing requests to time out and be dropped. The plugin stores excess requests in an internal FIFO queue to ensure orderly processing. It supports provider-level concurrency control, queue management, and end-to-end coverage, and can handle errors such as a full queue or timeout. Users need to configure concurrency limits, queue length, and wait timeout via a configuration file, while being mindful of serialization constraints and dependencies.
Read Moredsh-model-center: DSH Model Center Plugin
dsh-model-center is a plugin for DeepSeek Harness (DSH), designed to simplify the management of model inference. It consolidates scattered model catalogs, pricing tables, balance checks, and true free testing. The plugin aggregates free models from OpenRouter and provides capability ratings, supports one-click switching of routed models, and uses a “true free test” feature to verify model availability while eliminating any risk of incurring charges. It also supports TeamoRouter pricing and balance queries, with comprehensive caching and security mechanisms. Once installed, developers can efficiently manage model routing through the “Model Center” card in the Web interface.
Read Moregrloper/dsh-claude-oauth: Claude OAuth Integration for DSH
`grloper/dsh-claude-oauth` is a plugin for DeepSeek Harness (DSH), designed to simplify the process of integrating the Claude model. It integrates OAuth authentication into the web interface, supporting one-click sign-in with Google/Gmail accounts, automatic token management, refresh mechanisms, and quota tracking. The plugin is IPv6-compatible and can sync credentials to the official Claude CLI, greatly reducing the configuration overhead of using Claude Pro/Max in DSH.
Read MoreIntroduction to the Model Catalog Plugin
The Model Catalog plugin for DeepSeek Harness (DSH) aims to eliminate the manual burden of maintaining model configurations. By probing host endpoints, it automatically retrieves and normalizes model information, pricing strategies, and inference parameters. It supports multiple host types (such as Ollama and vLLM) and provides capability probing and validation, pricing source-chain population, and interactive configuration generation. Users can easily discover official APIs or local services via the command line and generate ready-to-load configuration files, significantly simplifying the model configuration workflow within the DSH ecosystem.
Read Morehpyperry/dsh-ref-lib: DeepSeek Harness Reference Library Plugin
The DeepSeek Harness plugin hpyperry/dsh-ref-lib focuses on model inference capabilities. The project aims to provide reusable reference library components. Given that public information does not include specific features, installation guides, or license details, users are advised to review the source code and license themselves before deployment to ensure compliance with their usage requirements.
Read Moredsh-variance: A Pure Node.js-Based Model Inference Tool
`dsh-variance` is a pure Node.js model inference plugin for DeepSeek Harness (DSH), maintained by uckkk. It makes no network requests and focuses on local computation, aiming to address localized task requirements. Built on DSH's "everything is a plugin" philosophy, it registers tools in a session after installation for developers to call directly. It runs with the permissions of the current process, with dependencies including `@deepseek-ai/cordis` and `@deepseek-ai/dsh-tools`, making it well suited to scenarios that require local execution and do not involve external API calls.
Read Moredsh-hero-product: Traffic-Driving Bestseller Pricing Tool
`dsh-hero-product` is a pure Node.js plugin for DeepSeek Harness (DSH) used to calculate pricing for traffic-driving bestsellers locally without network dependencies. It supports precise calculations and returns detailed results, making it suitable for scenarios requiring custom calculation logic. Installation requires specific dependency versions. Once enabled, it can be invoked through the `hero_product` tool. It is recommended to review the source code before use to confirm the logic.
Read Moredsh-allrouter: Unified Integration of AllRouter's Model Providers
The `dsh-allrouter` plugin achieves one-click switching across multiple models by registering AllRouter as a unified provider for DeepSeek Harness (DSH). By maintaining only one API key, it enables the use of multiple models such as Kimi K3, Claude, and GPT in DSH, with real token-based billing. After installation, configure the key via an environment variable and switch directly in DSH; no local tool code is required. It offers security features including zero-permission access, direct pricing, and no data storage, making it especially suitable for agent development scenarios that require frequent model switching.
Read Moredsh-llm-chatgpt-oauth: Using OpenAI Codex Models with a ChatGPT Plus/Pro Subscription
`dsh-llm-chatgpt-oauth` is a community open-source plugin that enables DeepSeek Harness (DSH) to directly invoke ChatGPT subscription models without an API key. It integrates with pi-ai's Codex routing via OAuth authentication and supports automatic token refresh, message conversion, and CLI login. The plugin requires Node.js, pnpm, and specific dependency versions, and is currently a private package. After installation, the service must be restarted and configured in the settings. Model availability is subject to the limitations of the ChatGPT account.
Read Moredsh-model-retry-settings: DSH Model Request Retry Count Configuration
`dsh-model-retry-settings` is a plugin that allows users to directly adjust the maximum number of retries for model requests through the DSH Desktop settings interface. It requires no core code modifications, supports 0 to 10 retries, and preserves the existing error classification and exponential backoff logic. After installation, users can configure the value in “Settings → General” to achieve fine-grained control over retry policies.
Read Moredsh-weekly-report: A Weekly Report Generation Plugin for DeepSeek Harness
The `dsh-weekly-report` plugin in DeepSeek Harness (DSH) is a lightweight plugin designed to address the cumbersome process of generating periodic weekly reports during model inference. Guided by the “everything is a plugin” philosophy, it provides a structured weekly report generation feature with a pure Node.js implementation and no network dependencies. Through the `weekly_report` tool, it allows users to easily generate structured reports without modifying core code. Users can install it via official commands. It depends on specific versions of DeepSeek-related libraries and is suitable for agent development workflows.
Read Moredsh-water-resist: DSH Plugin Water Resistance Level Assessment
DSH’s core design philosophy is “everything is a plugin,” and `dsh-water-resist` is designed for such vertical scenarios. It is a standalone pure Node.js module maintained by uckkk, with no network dependencies. It mainly provides water-resistance rating assessment and can return maintenance recommendations for watches and jewelry based on parameters. Users can install it through the official command and invoke the registered `water_resist` tool during a session. Before using it, note that the plugin runs with DSH process permissions; reviewing the source code and MIT license is recommended.
Read Moredsh-sphere: Sphere Calculation Plugin
The DeepSeek Harness plugin dsh-sphere focuses on spherical geometry calculations, implemented in pure Node.js without requiring internet access. Maintained by istone, the plugin aims to provide lightweight, network-independent local computation for agent reasoning. Users can install it via the official command and invoke the corresponding tools directly within a DSH session. Before installation, ensure that the source code complies with the MIT license and that the environment meets the Node.js runtime requirements.
Read Moredsh-sleep-cycle: Sleep Cycle Bedtime Recommendations
The DeepSeek Harness plugin `dsh-sleep-cycle` implements sleep-cycle calculation in pure Node.js to provide sleep-onset suggestions for DSH sessions. With no network dependencies, it accurately computes and returns detailed results. Developers can install it using a command and directly invoke its registered tool in a session. Note that the results are estimates only and do not constitute medical advice; review the source code and license before installation.
Read Moredsh-next-prime
The DeepSeek Harness (DSH) plugin dsh-next-prime is a model inference tool implemented purely in Node.js, focused on calculating the next prime number. It requires no network dependencies, relies only on the DSH core, and runs with the current process permissions. Users can install it via terminal commands and invoke its functionality directly in a DSH session. This plugin embodies DSH's design philosophy of "everything is a plugin."
Read Moredsh-mode: Mode
dsh-mode is an open-source plugin for DeepSeek Harness (DSH), maintained by uckkk. It focuses on model inference and is released under the MIT License. Developed with Node.js, the plugin requires no network access and is suitable for local deployment. After installation, it registers the relevant tools, which developers can call directly within a session. Before use, ensure that the environment meets the required dependencies, and note that the plugin runs with the permissions of the current process.
Read Moredsh-min: DeepSeek Harness Minimal Plugin
`dsh-min` is a DeepSeek Harness plugin maintained by uckkk, designed to provide tools that do not rely on external network resources for local or offline inference scenarios. Implemented in pure Node.js, it can be invoked directly in DSH sessions after installation. It depends on specific versions of DeepSeek-related libraries, is open-sourced under the MIT License, and runs with the permissions of the current process. Users should carefully review the source code logic and license terms before use.
Read Moredsh-latency-pct: Latency Percentile Plugin
DeepSeek Harness (DSH) introduces the `dsh-latency-pct` plugin, designed to address the challenge of accurately assessing response quality in model inference development using average latency alone. Maintained by uckkk and implemented in pure Node.js, the plugin has no network dependencies and runs locally. Its core features include precise latency percentile calculation and detailed result output. Once installed via command and required dependencies are satisfied, users can invoke it in DSH sessions to quantitatively analyze the distribution of inference latency.
Read Moredsh-go-rules: Go Rules Plugin
The `dsh-go-rules` plugin for DeepSeek Harness (DSH) provides agents with lightweight Go rules processing capabilities and supports returning Go learning plans based on parameters. The plugin is implemented in pure Node, has no network dependencies, and can be invoked in a session after installation. Note that it runs with the permissions of the current process. It is recommended to review the source code and license before installation.
Read Moredsh-ferment-check: Fermentation Check Plugin
`dsh-ferment-check` is a DeepSeek Harness (DSH) model inference plugin maintained by uckkk, focused on dough fermentation assessment and preparation plans. Implemented entirely in Node.js, it requires no network dependencies and runs quickly. The installation command is `dsh plugin add github:uckkk/dsh-ferment-check`. It requires an environment that satisfies the version requirements for `@deepseek-ai/cordis` (^4.0.1) and `@deepseek-ai/dsh-tools` (>=0.1.0-rc.6). Developers can invoke the plugin locally to obtain dough preparation workflows.
Read Moredsh-big-purchase: High-Value Purchase Decision Plugin
The DeepSeek Harness (DSH) plugin `dsh-big-purchase` aims to provide agents with localized purchasing decision-making calculation capabilities, following the "everything is a plugin" design philosophy. Implemented in pure Node.js without network dependencies, it encapsulates high-value purchase evaluation logic as callable tools, supports precise calculations, and returns detailed results. Installation requires specific version dependencies, and the plugin runs with the permissions of the current process. It is suitable for scenarios that incorporate spending analysis and budget assessment into agent decision-making; developers should review the source code to ensure security.
Read Moredsh-basic-training: Basic Training and Parameter Recommendations
The `dsh-basic-training` plugin for DeepSeek Harness (DSH) is a local tool implemented in pure Node.js and focused on providing basic training recommendations. It can run without an internet connection, making it suitable for network-restricted scenarios. Installation requires a specific environment version (e.g., `@deepseek-ai/cordis`), and it is important to inspect the source code and permissions. The plugin integrates into the DSH workflow as a lightweight solution and invokes basic training-related features through registered tool interfaces.
Read Moredsh-argon: A Pure Node.js Model Inference Plugin with No Network Dependencies
The DeepSeek Harness (DSH) plugin `dsh-argon` is designed to meet model inference requirements in offline environments. Built entirely on Node.js without any external network dependencies, it is lightweight. Installation requires `@deepseek-ai/cordis` ^4.0.1 and `@deepseek-ai/dsh-tools` >=0.1.0-rc.6. It is suitable for fully offline deployments, restricted intranet environments, or scenarios requiring source code auditing, and is released under the MIT license.
Read MoreDSH Plugin: satan9394/dsh-vector-search
This plugin is designed to provide vector retrieval capabilities for the DSH ecosystem and is classified as a “model inference” skill. It focuses on vector retrieval engineering, offering embedding and hybrid retrieval functions. The plugin runs with the permissions of the current process and is suitable for DSH developers who require context processing. Since fetching the README failed, it is recommended to inspect the source code directly to confirm installation instructions and licensing terms.
Read MoreDeepSeek Harness Plugin: dsh-caveman-speak - Output Token Compression Tool
`dsh-caveman-speak` is a model inference plugin in the DeepSeek Harness ecosystem, designed to significantly compress output tokens using a “caveman-speak” format. It can reduce token consumption by approximately 65% while maintaining byte-level accuracy for code, commands, and error messages. It also supports token sink-point auditing and progressive disclosure recall. It is suitable for cost-sensitive high-frequency calls, long conversation compression, and code debugging scenarios. It is recommended to inspect the source code before use to ensure security.
Read Moredsh-performance-slider: DeepSeek Harness's Codex-style 6-level Performance Slider
`dsh-performance-slider` is a performance-switching plugin for DeepSeek Harness that replicates the desktop experience. It uses a 1–6 level slider to synchronously switch models, reasoning levels, and 31 character background images, simplifying manual parameter tuning. The plugin supports automatic installation of the cordis patch and provides a standalone preview feature. User selections are automatically saved to local storage and can be restored across ports.
Read MoreDeepSeek Harness Plugin: workspace_digest Tool
`dsh-workspace-digest` is a bundle for DeepSeek Harness (DSH) that registers the `workspace_digest` tool to list filenames per directory in the Web UI. It is maintained by `cui-stack`, uses a pure JavaScript stack, and is licensed under the MIT license. Users can install and enable it via the official command. The configuration file is `cordis.patch.yml`, with a default `maxEntries` value of 20 and support for ID-based overrides in profiles.
Read MoreAngles for DeepSeek Harness
`anglesvideo-dsh-plugin` is an open-source plugin for DeepSeek Harness that leverages the Angles Developer API to transform local software repository context into executable video concepts and rendered results. The plugin provides a one-stop toolchain covering repository analysis, concept generation, template selection, rendering previews, and final video output. During installation, an API key must be configured, and after setup, an automated video production workflow is performed through a series of tool calls (such as `inspect_repository`, `create_launch_concepts`, and `render_launch_video`). This plugin is a developer preview and includes privacy protection mechanisms at runtime to ensure that sensitive data is not uploaded.
Read Moredsh-vision-plugin: Empowering DeepSeek Harness Text Models with Vision Capabilities
The `dsh-vision-plugin` is a plugin for DeepSeek Harness (DSH) that enables visual analysis capabilities for text-only models, such as DeepSeek v4. It uses an intermediate layer to send images to a vision model, such as Qwen3.7-plus, for transcription into text, which is then processed by the main model. The plugin supports pasting images directly into a conversation to ask questions and provides the `vision_analyze` tool for analyzing local files. Before use, image input support for the model must be declared in the configuration file. The plugin is compatible with DSH v0.1.0-rc.8 and later.
Read MoreIntroduction to the DeepSeek Harness CodeBuddy Plugin
The DeepSeek Harness (DSH) plugin “taikaikaikai-pixel/dsh-codebuddy-plugin” aims to integrate Tencent Cloud CodeBuddy capabilities into DSH, providing an unofficial but fully featured model inference and tool integration solution. It supports 18 mainstream models and includes image input, adjustable thinking intensity, and automatic routing. Key features include integration of CodeBuddy web search/scraping and image generation tools, with unified handling of OAuth login and multi-key management through a local streaming bridge. Users can configure credentials in the Web UI settings card and must run the streaming bridge locally to enable the primary chat function.
Read Moredsh-vision-adapter: Adding drag-and-drop image recognition capability to DeepSeek Harness
dsh-vision-adapter is a DSH-exclusive plugin maintained by motongv (MIT License) that solves the issue where DSH text models cannot recognize images. The chat interface remains unchanged; images are first converted into text descriptions by the vision model, which is then submitted to DeepSeek for reasoning and response. The plugin acts as an in-process HTTP bridge within DSH to register vision.
Read Moredsh-auto-chess: Simplified Auto Chess Plugin for DSH Web
`dsh-auto-chess` is an MIT-licensed DSH Web plugin developed by `omdsh-dev`, designed to provide a simplified auto-battler interface for observing the decision logic of large language models under fixed rules. The plugin supports Human-vs-AI (where the user plays the Blue side) and Dual-AI battle modes, allowing users to configure independent models and edit prompts for the Blue and Red sides respectively. It also allows selecting thinking levels such as Off, High, and Max, automatically falling back to default settings if the model does not support them. The system defaults to a single decision timeout of 300,000ms and an output limit of 32,000 tokens. After each AI decision, users can view the full reasoning text on the panel; if an invalid operation or transient failure occurs, the system will indicate the cause and automatically retry within the budget. Building the plugin requires executing `pnpm run build` and running `dsh`...
Read Moredsh-think-any-lang: Select Model Thinking Language in DSH
The DSH plugin `dsh-think-any-lang` allows users to independently select the model inference language in Web settings, ensuring the final response remains in the user's language. The plugin provides a `Thinking Language` dropdown in `Settings → General`, supporting 12 languages (including Simplified Chinese and English) and an "Off" option, with selections being persisted. Its core mechanism injects thinking instructions written in the target language via `systemPrompt.section`, avoiding the latency caused by extra model calls and achieving zero extra call control. Installation requires the command `dsh plugin --profile <name> add github:lco117/dsh-think-any-lang`, and it is recommended to pin Com.
Read Moredsh-fate-spectrum: Providing Deterministic BaZi and ZiWei Charting for AI Agents
dsh-fate-spectrum is a DeepSeek Harness plugin maintained by EchoUser005, which performs Four Pillars of Destiny and Purple Star Astrology chart calculations locally offline, outputting structured and reproducible destiny chart facts for Agent invocation. It supports dual-chart viewing, candidate charts for unknown hours, and supplementary questioning when information is insufficient. Version v0.1.0 can be installed via `dsh plugin add`. It is suitable for DSH users who require verifiable chart calculations rather than model-inferred Heavenly Stems and Earthly Branches.
Read MoreDSH WorkBuddy Connect: Integrating the WorkBuddy Desktop App Model into DeepSeek Harness
DSH WorkBuddy Connect is a DeepSeek Harness plugin maintained by corrinehu, which reuses the login state of the WorkBuddy desktop app to seamlessly integrate built-in models such as GLM, DeepSeek, and Kimi into DSH with zero configuration. It supports Web, Desktop, and TUI interfaces. This article introduces its features, installation commands, typical usage, macOS verification, known limitations such as unofficial API dependencies, and reminds users to review the source code and MIT license before installation.
Read MoreToolshrink: Semantic Compression for Agent Tool Outputs
toolshrink is a DeepSeek Harness plugin maintained by unclecode, designed to address the issue of information loss caused by truncating tool outputs from the beginning and end. It features 13 built-in content-aware reducers (for tests, diffs, JSON, builds, logs, etc.), with a fallback to size-based truncation when no match is found. It also annotates the omitted volume and optionally saves the full original text via spill. Installation is straightforward with a single dsh plugin command, and it can also be used as a Node.js library or a custom cut extension. Licensed under MIT, the current version is 0.1.0.
Read MoreMiraculous Standard: DeepSeek V4 Pro/Flash Unified Anchored Presets
Miraculous Standard is an agent preset provided by rinDBeans for DeepSeek Harness, unifying coverage across four combinations of V4 Pro/Flash with the official API and opencode-go. The initial request only exposes a Minimal toolset for trajectory anchoring. After upgrade, it unlocks the Pro resident set or Flash full directory based on the model, coupled with context-gate and epoch management to address compaction and resume anchoring issues. This article compiles installation commands, model configurations, and key items from agent.cordis.yml, while explaining applicable scenarios and known limitations.
Read Moredsh-image2-draw: Integrating gpt-image-2 Image Generation in DSH Dialogues
dsh-image2-draw is a DeepSeek Harness image generation plugin maintained by JuneLearn, which calls gpt-image-2 through a third-party OpenAI Images-compatible interface. It requires only the configuration of a baseURL and API Key. The plugin offers image2-generate for text-to-image generation and image2-edit for image editing, supporting continuous generation of 1–8 images, various sizes, and PNG/JPEG/WebP reference images. Results are previewed within the session and saved to the outputs/image2/ directory. API Keys are stored only in DSH credentials, with timeouts and 524 errors not automatically retried to prevent duplicate billing. It is suitable for on-the-spot image generation within DSH Web agent workflows. Before installation, please review the source code and MIT license, and evaluate the relay service independently.
Read Moredsh-agy-link: Integrating the Google Antigravity Model into DSH
dsh-agy-link is a DSH community plugin maintained by amlyczz that integrates Google Antigravity models into DeepSeek Harness via the official local agy CLI. It supports multi-account pool rotation, official 5-hour and 7-day dual bucket quota monitoring, GUI-based OAuth login without pasting, streaming output, thinking, tool cards, and token usage mapping. The installation command is `dsh plugin --profile web add dsh-agy-link`, requiring Node.js ≥ 24 and an installed agy CLI.
Read Moredsh-drop-to-path: Delivering Images and Files to Text-Only Models via Workspace Path
dsh-drop-to-path is a DSH Web plugin maintained by loudMore, designed to address the issue where pure text models cannot receive image attachments and DSH rejects non-image files. It retains the native attachment UI and, during sending, uploads images along with PDF/Office/archive/video/audio files to the workspace `.drops/` directory and appends their paths, bypassing the `attachment-error` pre-check. When used in conjunction with dsh-vision-toolkit, it enables the capability to view images by pasting screenshots. It supports drag-and-drop and pasting, and can be installed using the command `dsh plugin --profile web add`.
Read Moredsh-think-zh: Enabling Simplified Chinese for DSH Reasoning Process
The default thinking process of DeepSeek Harness is often in English, which is not conducive to Chinese users' comprehension of the reasoning chain. dsh-think-zh injects concise language instructions into the system prompt for each request: the thinking process is always in Simplified Chinese, the response follows the language of the query, and code and identifiers remain in their original language. The plugin is injected via the cordis section, with a token cost of approximately 75 characters per request. It supports disabling or customizing the instructions. Installation command: dsh plugin --profile web add github:Len7183/DSH-Think-zh. Licensed under MIT, maintained by Len7183.
Read Moreoh-my-dsh：DeepSeek Harness Curated Distribution
oh-my-dsh, maintained by amplifthq, is a curated distribution of DeepSeek Harness, overlaying out-of-the-box coding profiles, lazy MCP reuse, LSP navigation, SSRF-hardened web scraping, and approval-gated capabilities without forking the upstream. It supports both portable and npm installation channels, offering an omd Web UI and headless task mode. Users should be aware that it runs with dsh process privileges, and plugins activate within the same process as the harness.
Read Moredsh-doc: DeepSeek Harness Full-Local Document Parsing and Offline OCR
dsh-doc is an MIT-licensed plugin for DeepSeek Harness released by Sqhao-O. It locally parses PDFs, Office documents, images, and scanned files, outputting Markdown, plain text, or JSON without requiring Docker, HTTP services, or API keys. On Windows x64, it can mount pre-built Python + Xberg runtimes and offline Tesseract (for Chinese and English); other platforms fall back to the Node engine. The installation command is `dsh plugin --profile web add dsh-doc`. After configuring the engine via `cordis.patch.yml`, it can be used through `dshdoc_health` and `dshdoc_extract`.
Read Moredsh-vision-opencode: Configuring Switchable Vision Models for Text-Primary Master Models
dsh-vision-opencode is a DSH Web plugin that converts chat images into text via configurable vision models, enabling pure-text main models like DeepSeek to continue reasoning without switching the primary model. It supports image recognition model selection in the input box, vision settings, the vision_read_image tool, and vision-image-analysis skill, including a 60-second timeout and failure retry fallback. Installation: dsh plugin --profile web add -w github:poiuyjie/dsh-vision-opencode. MIT license, maintained by poiuyjie.
Read Moredsh-search-boost: Enhancing Multi-Engine Search and In-Depth Research for DSH
dsh-search-boost is a bundle plugin developed by Mr-remon219 for DeepSeek Harness, enhancing the built-in web_search/web_fetch with multi-engine integrated search, X search, page fetching, and deep research capabilities. It supports both free and API dual search layers, enabling zero-configuration parallel retrieval via Bing/DDG/Yahoo/Exa-free, with optional integration for Tavily/Brave/Exa. Installation command: dsh plugin --profile web add dsh-search-boost.
Read Moredsh-plan-graph: Visualizing Session Tool Calls as Interactive Flowcharts
dsh-plan-graph is an out-of-tree plugin for DeepSeek Harness maintained by HR2AY (MIT, 0.3.0). It adds a Plan Graph tab to the session view, rendering tool calls and messages as a pannable and zoomable flowchart. The plugin supports coloring by state, hiding tool calls, grouping by turn, favorites (with localStorage persistence), a right-click context menu for nodes, and locating within the conversation area. Installation can be done via `dsh plugin add` from a local directory, GitHub source (requires `allowBuilds`), tarball, or npm; before starting, you can use `--dump-config` to verify layer loading. It is designed for DSH developers who need to visualize agent call chains; before installation, please review the source code, as the plugin runs with the permissions of the dsh process.
Read Moredsh-advisor: Mounting Independent Review Sub-Models for DSH Sessions
dsh-advisor is a DSH plugin maintained by omdsh-dev that attaches an independent review model to each session, passively observes the main transcript, and injects suggestions categorized as nit, concern, or blocker after each stepped turn. The plugin is advisory-only and does not approve or reject actions on behalf of the main model. It supports a web Settings Advisor card and the `dsh-tui /advisor` command, and can be enabled by configuring the provider and model in `settings.yaml`.
Read Moredsh-codex-oauth: Accessing DSH's GPT, Image Generation, and Web Search via ChatGPT/Codex Subscription
@wnjxyk/dsh-codex-oauth is a DeepSeek Harness Web plugin maintained by WNJXYK. It binds ChatGPT/Codex subscriptions via OAuth, enabling the openai-codex model, generate_image for image creation, and web_search for internet searches, while displaying subscription quotas in the interface. It supports browser and device code login, requiring no configuration by default. Optional settings include a custom Cordis credential path and port. Verified on DSH 0.1.0-rc.6 under the MIT license. Installation: `dsh plugin --profile web add -w @wnjxyk/dsh-codex-oauth@latest`. Note: This relies on non-public Codex interfaces, so users should independently assess potential risks from protocol changes or account issues.
Read Moredsh-provider-model-configurator: Centralized Management of Provider Model Entries in DSH WebUI
dsh-provider-model-configurator is a DSH Web plugin maintained by LiangYin233. It provides centralized viewing, creation, editing, copying, and deletion of provider model entries on the "Model Pro" settings page, covering context windows, maximum output, input modality, reasoning tiers, and compat compatibility switches. It supports one-click form filling from llm-pi-ai presets or other configured providers, and explains the snapshot behavior of catalog providers. Installation requires the command `dsh plugin --profile web add`, with version v0.3.9 as an example, under the MIT license.
Read Moredsh-vision-proxy: DeepSeek Plain Text Models Can Also Recognize Images in GUIs
dsh-vision-proxy is a model inference plugin (v0.4.1, MIT) developed by Flyvhidbwo for DeepSeek Harness. It registers the deepseek-vision route, translating GUI-attached images into text via a VLM before passing them to DeepSeek for responses, enabling pure text models like V4-Pro to recognize images. By default, it can use the official deepseek-v4-flash-vision-exp, and it also supports Bailian, Zhipu, OpenRouter, and automatic detection of local Ollama. The installation command is dsh plugin --profile web add dsh-vision-proxy; after selection, choose DeepSeek + automatic image recognition, and you can paste images for conversation.
Read Moredsh-llm-fallbacks: Configuring Backup Model Chains for DSH Agents When LLM Fails
dsh-llm-fallbacks is a DSH model inference plugin maintained by omdsh-dev. When an agent's LLM request fails due to retry exhaustion, authentication, quota, or rate limiting, it automatically switches according to the current role's provider/model fallback chain and continues the current step on the target model. It supports 24-hour rootChain, time-based rotation, and sub-agent role rules, with a shared fallbacks configuration for both the web interface and dsh-tui. Installation: dsh plugin --profile web|dsh-tui add dsh-llm-fallbacks. It must be explicitly enabled with enabled: true, and the chain must end with the official V4 Default model.
Read Moredsh-media-skills: Enhancing DeepSeek Harness with Free Image Recognition and Generation
dsh-media-skills is a DSH plugin maintained by MJorgin (MIT, v0.3.0), supporting rc.7/rc.8/v0.1.1-rc.x. It offers automatic captioning for stickers, vision-review for image reading, and media-tools for image generation. The vision routing includes GLM-4V-Flash and DeepSeek-V4-Flash-Vision-Exp, while image generation utilizes SenseNova U1 Fast and Kolors. API keys are not stored in the repository, and the installation command is `dsh plugin --profile <name> add github:MJorgin/dsh-media-skills`.
Read Moredsh-routing-suite: Adding Selectable Intelligent Routing Modes for DeepSeek Harness
dsh-routing-suite is a DSH community plugin maintained by dragonbaba (v0.1.2, MIT). It adds an "Intelligent Routing Mode" to the mode menu while preserving the Standard preset persona, context, and full tool directory. It supports three strategies: auto, inspect-first, and direct, which guide tasks by checking before modifying, or implementing directly and then verifying. It does not make additional model requests, nor does it read/write files or execute commands. The guide covers desktop and CLI installation, manual preset copying, configuration options, applicable scenarios, and security boundaries.
Read Moredsh-quant: AI-native Quantification Toolkit for DSH
dsh-quant is a DSH quant plugin maintained by pengpengyi92, offering 59 quant_* tools that cover six domains: data, factors, ML, risk control, execution, and ecosystem. It connects end-to-end research through the PDAT→PET pipeline. Primarily designed for use with agents, the plugin is implemented in TypeScript with zero runtime dependencies, installed via npm, and enabled in cordis.yml. This article introduces its module paradigm, core capabilities, and usage instructions, while also explaining applicable scenarios and security considerations.
Read Moredsh-user-experience: Conducting Frontend UX Audits from the Target User's Perspective
dsh-user-experience is a DeepSeek Harness plugin maintained by DietCokewithSugar, driven by target user personas to guide UX walkthroughs during development. It scans frontend source code such as React and Vue 3, optionally incorporating browser evidence, and uses natural language reports to pinpoint experience issues while generating verifiable optimization recommendations. It also supports automatic re-checks after code modifications. Installation command: dsh plugin --profile web add dsh-user-experience@0.4.2.
Read Moredsh-llm-codebuddy: Integrating WorkBuddy API and CodeBuddy China Region Model into DSH
dsh-llm-codebuddy is a DSH community plugin maintained by Axiaohungry, which adds the CodeBuddy China Region Provider to DeepSeek Harness. It supports two authentication methods: WorkBuddy API Key and CodeBuddy China site token. The plugin allows users to obtain and edit the model directory via the WebUI and configure thinking settings. It can be installed independently without modifying the DSH global directory and has been verified in DSH 0.1.0-rc.6. It is suitable for developers who only have WorkBuddy credits or a CodeBuddy account and need to call CodeBuddy models within DSH.
Read Moredsh-codex-subscription: Direct Access to ChatGPT/Codex Subscription via DSH
dsh-codex-subscription is the DeepSeek Harness model inference plugin maintained by WSL043. It integrates the Codex subscription model into DSH via ChatGPT OAuth, eliminating the need for an OpenAI API Key or Codex CLI. It offers genuine quota, search routing, Fast mode, and Beta image generation. The article introduces the plugin's positioning, core capabilities, official installation commands (dsh plugin / npx / Windows scripts), and activation steps, along with applicable scenarios and security considerations.
Read Moredsh-files: DSH Session File Upload and Document Reading Plugin
dsh-files is a DeepSeek Harness dual-sided plugin maintained by taxueseek. With a single line of cordis configuration, it provides paperclip/folder/drag-and-drop uploads, session-isolated storage, and colorful file cards for dsh web. It reads text/PDF/DOCX/XLSX files via the read_document tool based on content sniffing, accompanied by LRU caching and pagination. JPEG/PNG/WebP/GIF files are processed through the harness core attachment pipeline, supporting native image viewing in vision models. Install command: dsh plugin --profile web add dsh-files. Public network deployment requires trustedHosts configuration. Licensed under MIT, version v0.4.0.
Read MoreCustom-First Control Prompt: Injecting System Segments and Reference Dialogues Before Persona
@wm-coders/dsh-custom-first-control-prompt is a DSH deployment-side prompt prefix plugin maintained by WM-CODER. It registers ordered system prompt segments before the persona through sections and prepends user/assistant reference conversations as real alternating messages to each ordinary dialogue request via history, intercepting through llm/stream with zero session log writing. The static content facilitates prefix KV cache reuse. Installation is done via `dsh plugin --profile web add`, with configuration written to `cordis.patch.yml` for targeted overrides or saved through the panel. It is suitable for scenarios requiring unified deployment rules and reference conversations; note the token duplication overhead, that it takes effect after a restart, and consider the plugin process permissions.
Read Moredsh-lark: Directly Interact with DeepSeek Harness Agent in Feishu / Lark
@sugarforever/dsh-lark is a community-maintained DeepSeek Harness plugin for Feishu / Lark. It receives messages via WebSocket long connection, eliminating the need for public webhooks, and maps Feishu single chats, group chats, and topics to Harness Sessions, utilizing pre-configured models, tools, and Agent Presets. The article introduces the runtime requirements, Feishu app configuration, npm installation commands, Settings and patch configuration methods, as well as typical usage scenarios such as single chats, group chats, and whitelists, along with deployment considerations.
Read Moredeepseek-harness-workbench-plugin: Integrating Dialogues, Editors, and Terminals into a Unified Workbench within the DSH Web UI
The dsh-workbench-plugin maintained by loadingvx provides a three-column workspace for the DeepSeek Harness Web UI: conversation on the left, an editor and intelligent terminal in the center, and files/Git/usage/Ultra Slash on the right. It supports Agent Control Plane execution traces, Canvas React previews, slash commands without interrupting the conversation round, AI terminal assistance, and streaming AI message submissions. Installation requires specifying the version: dsh plugin --profile web add dsh-workbench-plugin@0.1.31. After restarting dsh web, a new session can be started to use it. MIT license, npm latest version 0.1.31.
Read Morepicturereader: Adding Local Image Recognition Capability to the Pure Text DSH Model
picturereader is a DSH model inference plugin (v3.2.0) maintained by jing-hy. It enables pure-text models to achieve native thumbnails through a visual twin adapter, converting images into structured textual evidence using local tools such as image_scan and image_ocr. It supports three routing modes—privacy, intelligence, and rigor—and an optional external VLM. Additionally, it includes document_to_image for converting documents to images and image_edit for local image editing. It runs entirely locally by default and is licensed under MIT.
Read Moredsh-deepread: Evidence-First Deep Reading for AI Agents
DeepRead (xiehuan123/dsh-deepread) is an evidence-first deep reading plugin within the DSH ecosystem. It extracts traceable claims, evidence, confidence scores, and knowledge maps from sources such as articles, PDFs, and books. It offers five modes: quick, deep, map, Feynman, and book, and supports two forms: Portable Agent Skill and Host plugin. The Host version includes PDF extraction, batch comparison, cost preview, background job progress, and HTML/XMind export. Installation command: npx skills@latest add xiehuan123/dsh-deepread or dsh plugin --profile web add dsh-deepread. Licensed under MIT, requires Node ^22.19 or >=24.
Read Moredsh-claude-provider: Configuring and Discovering Models for DeepSeek Harness with Claude Provider
@mofeng2223/dsh-claude-provider offers a standalone Claude Provider type and Anthropic Messages format for DeepSeek Harness. DSH 0.1.1-rc.1 natively supports Claude reasoning-effort, so this plugin no longer intercepts requests. Instead, it writes settings such as reasoning level into the settings.yaml file during front-end saving. It also complements the GET /v1/models endpoint for model discovery, sets default context and output values for known Claude models, and provides three reasoning modes: five-level, four-level, and toggle. Installation: npx @deepseek-ai/dsh plugin --profile web|headless add @mofeng2223/dsh-claude-provider. Licensed under MIT, maintained by MoFeng2223.
Read MoreBillion-Context-DSH: Empowering Models to Decide When to Compress Context
Billion-context-dsh provides an Active Context Pruning backend for DeepSeek Harness, where the model determines when to compress and restore context through the compress, decompress, search_context, and acp_status functions. Ported from billion-context-pi and reusing acp-kernel, it can be installed via the bundle command: dsh plugin --profile web add billion-context-dsh. Version v0.2.13 (beta) is suitable for long conversation and long task scenarios.
Read Moredsh-AuthInOne: Complete Provider Login, Model Switching, and Usage Statistics within DSH
dsh-AuthInOne is a DeepSeek Harness community plugin (MIT, v0.2.0-alpha.4) maintained by Stormycry-cryp. It integrates Provider/account login, API Key and custom OpenAI-compatible configurations, future session default models, visual fallback for pure-text models, and cross-session token and cost statistics into the native Models and Usage settings. OpenAI Codex serves as the stable pathway, while Kimi Code, xAI, Anthropic, and others are experimentally compatible. A single command via `pnpm dlx` enables web installation on the same port without modifying the DSH core. It is suitable for developers who need to manage authentication, models, and usage uniformly within DSH. Before installation, please review the source code, as the plugin runs with the DSH process permissions.
Read Moredsh-vision: Supplementing DeepSeek Harness with Near-Native Image Understanding
dsh-vision is a DeepSeek Harness plugin (MIT, v0.1.2) maintained by oil-oil, supporting rc.6/rc.7. When the main model is vision-capable, it transmits images directly; for text-only models, a separate visual backend observes the images and injects untrusted context, with fallback to macOS Vision or Tesseract when the cloud backend is unavailable. Backends such as ZenMux and Model Studio can be configured via the Vision Recognition card or settings.yaml, and it is compatible with see-skill. Installation: npx @deepseek-ai/dsh plugin --profile web add github:oil-oil/dsh-vision.
Read Moredsh-commandcode-provider: Integrating Command Code Models in DeepSeek Harness
dsh-commandcode-provider is a community LLM provider plugin maintained by Mars-Sea, which registers Command Code as the commandcode route for DSH. The plugin provides a real-time model directory, reasoning-effort support, authorization and usage cards on the Settings page, automatic multi-account rotation, and a Command Code card on the Models page. The installation command is `dsh plugin --profile web add @mars-sea/dsh-commandcode-provider@latest`, and a Command Code API key is required to be prepared by the user. It is licensed under the MIT License.
Read Moredsh-reasoning-effort: Adding a Codex-style Model and Reasoning Intensity Slider to DSH
dsh-reasoning-effort is a DSH Web plugin maintained by HanaAyane, providing a Codex-style model and reasoning intensity slider below the input box. The slider levels automatically adapt based on the `reasoning.efforts` setting in the model directory and synchronize with the `/model` command. It supports light and dark themes, custom provider level guidelines, and a dismissible "big fat fish" slider button. The installation command is `dsh plugin --profile web add github:HanaAyane/dsh-reasoning-effort#main`, requiring a Web Host restart to take effect. It targets DSH 0.1.0-rc.6 and is open-sourced under the MIT license.
Read MoreModLens: Augmenting Pure Text Models with Visual Capabilities
ModLens is a DSH vision plugin maintained by liustack (MIT, v3.25.0). After pasting an image, an independent vision engine generates structured JSON evidence for OCR, layout, and semantics, which can be used by pure text models like DeepSeek and GLM for answering. It supports both direct pasting and the (modlens vision) model paths, with six built-in providers and the ability to reuse local CLI login states. DSH installation command: npx -y @deepseek-ai/dsh plugin --profile web add @liustack/modlens@3.25.0.
Read MoreTokenLedger: Tracking DeepSeek Harness Token Usage Per Station and Project
TokenLedger is a DSH web plugin maintained by zh667, which automatically attributes token usage to actual relay stations and working directories without any configuration. It supports multiple relay balances/subscription quotas, usage heatmaps, cost estimation, and CSV export. This article introduces its positioning, core capabilities, installation commands, and typical usage, while also explaining the attributes of the community directory and security considerations.
Read MoreDirect Image Generation in DeepSeek Harness: Getting Started with the dsh-image-gen Plugin
dsh-image-gen is an open-source DeepSeek Harness image generation plugin by shanliuling, which supports direct image generation via the generate_image tool within web dialogs. It is compatible with multiple providers including Gemini, OpenAI, Volcano Ark Seedream, and Tongyi Wanxiang, and comes with built-in gallery management and workspace persistence. This article covers installation commands, API Key configuration, typical usage, as well as plugin permissions and security considerations.
Read Moredsh-plugin-subscriptions: Integrate ChatGPT, Claude, and Grok into DSH via Subscription Accounts, No API Key Required
dsh-plugin-subscriptions is a DeepSeek Harness model inference plugin maintained by V1ki. It bridges ChatGPT (Codex), Claude, Grok, and GitHub Copilot subscriptions into DSH's LLM Provider via a Web-based OAuth interface, eliminating the need for API keys. The plugin supports a real-time model catalog, subscription usage display, inference level and Codex speed switching, and includes tools such as x_search, image_generate, and video_generate. This article introduces its positioning, installation commands, and typical usage, while also reminding users about plugin permissions and compliance considerations.
Read MorePaste Images and View: Getting Started with the DeepSeek Harness Visual Plugin dsh-vision-toolkit
dsh-vision-toolkit is a vision toolkit developed by Anionex for DeepSeek Harness, built upon agent-vision-toolkit, enabling pure text models to support image-based Q&A, long screenshot OCR, UI reconstruction, and pixel comparison. This article introduces its core tools, installation methods via npm/GitHub, usage of vision-skills, and configuration considerations, with data cross-verified from SkillHub and GitHub.
Read MoreGive Pure Text DeepSeek Harness Agent Eyes with dsh-vision-router
dsh-vision-router is a DeepSeek Harness vision routing plugin maintained by ysr666: the vision model serves only as the eyes, while DeepSeek continues reasoning. It features a built-in key-free OVH vision link and includes 14 pixel-level tools (Q&A, localization, cropping, pixel comparison, OCR, etc.). With a one-command installation, you can click "👁 Vision" before sending an image to perform multi-step image recognition just like using regular tools. This article verifies the installation command, tool list, and usage notes based on the SkillHub directory page and GitHub README.
Read More