DeepSeek Harness Plugin Hub

Publish and manage complete Harness Profiles. Discover Plugins for your next setup.

Explore

PluginsPresetsDocsNews

Community

Publish a pluginContactReport an issue

Resources

Plugin Hub on GitHubDeepSeek HarnessSystem statusPrivacy notice
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

Independent and unofficial. Not affiliated with, authorized by, or endorsed by DeepSeek.

Vision Bridge — DSH Plugin for DeepSeek Harness
DeepSeek Harness Plugin Hub
ProfilesPluginsCategoriesNewsDocsSign inManage Profiles
ProfilesPluginsCategoriesNewsDocsSign in
← Plugins
V

@dsh-external/dsh-vision-bridge

Vision Bridge

DeepSeek Harness vision bridge: session screenshots route by model capability. Image-capable models receive attachments inline; text-only models see the standard placeholder, and the bridge reads the image through a vision-language model — on demand via the vision_bridge_read tool or automatically v

The plugin will be installed here. Keep web if you are unsure.

npx -y @deepseek-ai/dsh plugin --profile web add github:shenhuanageshei/dsh-vision-bridge#6ed206fdc11861a4f5fc5a47778c6ca17372b99b
READMECompatibilityVersions

Description

DeepSeek Harness vision bridge: session screenshots route by model capability. Image-capable models receive attachments inline; text-only models see the standard placeholder, and the bridge reads the image through a vision-language model — on demand via the vision_bridge_read tool or automatically via agent/pre-step context injection. 会话截图按模型能力自动分流:多模态模型原样直读;纯文本模型经 VLM 代读(tool/auto 双模式)。

Compatibility and provenance

Vision Bridge is published as @dsh-external/dsh-vision-bridge and currently resolves to version 0.1.1. The Hub verifies its manifest and preserves the exact installation source for reproducible installs.

DSH compatibility
*
Runtime surfaces
any
Release source
github
Registry updated
9/6/2026

Versions

0.1.1stable
9/6/2026
0.1.0stable
9/4/2026

Related plugins

Loading related plugins…

Latest
0.1.1
DSH
*
HMR
Process restart
Tree shaking
Safe tree shaking not declared
Unpacked size
Unavailable
Files
Unavailable
Surface
any
License
BSD-3-Clause
Source
github
GitHub
★ 0
Weekly downloads
0
Last push
9/16/2026
View source ↗
README badge

Click the badge to copy Markdown for your README.

Do you maintain this Plugin?Claim benefit · Priority security scan

Verify the GitHub repository declared in package.json to manage this listing. After you claim it, Hub will prioritize a security scan of the current version and publish the result when it passes.

Claim this Plugin →
Report an issue

Related plugins

More verified plugins in vision-media.

Tool Describe Image@linxin666/dsh-tool-describe-imageModel-facing describe_image tool for the dsh web GUI: gives a text-only model image understanding by asking a vision-language model at an OpenAI-compatible endpoint to describe one image (local path, http(s) URL, or attachment reference). Hot-pluggable — Modlens@liustack/modlensPlug-in vision for text-only LLMs, powered by the free Antigravity CLIDeepseek Ivideodeepseek-ivideoiPolloWork HyperFrames Video Studio and 27 editable video templates as a native DeepSeek Harness conversation view.Image Gendsh-image-genBring ChatGPT-like image generation to DeepSeek Harness — Gemini, OpenAI, Seedream, DashScope, local ComfyUI & more.

README

dsh-vision-bridge

DeepSeek Harness 会话截图按模型能力自动分流 —— 多模态模型原样直读,纯文本模型由 VLM 代读(tool / auto 双模式)。

它解决什么问题

在 DeepSeek Harness 里往会话发截图时:

  • 当前模型支持图像输入(模型声明含 image 模态)→ 截图原样内联进请求,模型原生直读;
  • 当前模型是纯文本 → 运行时把图投影成 [image omitted … attachment sha256:xxxxxxxx] 占位符。本插件接管这条路径:附件在请求前已持久化,插件回扫会话日志解析出完整引用,把截图交给视觉语言模型(VLM)代读,再把结构化描述交还模型。

对用户来说效果是:截图随便发,任何模型都"看得见"。

能力面

工具/行为说明
vision_bridge_read(tool 模式)模型自主调用:按占位符里的 8-hex 前缀(或 latest)读取本会话截图,返回结构化描述(概览/可见文字/布局/对提问的回答)。输出恒带 UNTRUSTED 声明。只收 { ref?, question? },无 path/url 面。
auto 预描述注入新到 user/message 图片后台异步分析,经 agent/pre-step 同轮追加 <vision-context>(按进入消息的 attachment id 相关性匹配;超时/失败/无匹配一律放行),迟到结果经 PromptContext 兜底、消费后剪除。

多命中(8-hex 前缀碰撞)返回候选列表而非猜测;compaction/fork/resume 后旧截图仍可解析(原始日志 append-only)。

设置(命名空间 vision-bridge,设置页自动渲染,live 生效)

字段说明
modetool / auto / both(默认 both)
provider.baseURL / model留空则首次运行从 dsh-vision-toolkit 拷贝并本地固化
credentialDSH CredentialRef(默认 VISION_API_KEY,与 vision-toolkit 同名共享凭证条目)
其余timeoutMs / concurrency / cache / maxImageBytes / maxImagePixels / visionCapabilities / language / autoMode.maxPerTurn —— 见 docs/design.md §5.7

安装(web profile)

  1. profile/profiles/web/package.json 依赖加一行: "@dsh-external/dsh-vision-bridge": "file:../../../plugins/dsh-vision-bridge" (本仓库放到 profile/plugins/dsh-vision-bridge 时用此相对路径;其它位置自行调整。)
  2. dsh.profile.bundles 数组追加 "@dsh-external/dsh-vision-bridge"。
  3. cd profile && pnpm install(file: 依赖是复制不是链接;以后改插件源码也要重跑或手动同步安装副本)。
  4. 重启 DSH web。

⚠️ 挂载由插件自带的 cordis.patch.yml(dsh.bundle.patch)自动完成。不要在 profile 的 cordis.patch.yml 里再手动加同名 insert 行——两处同 id 会以 duplicate loader entry id 崩掉启动。配置覆盖走 settings.yaml 的 vision-bridge: 键。

已知限制

  • 回合取消只覆盖附件读取:取消时 readImageRequest(附件字节读取)即时中止;经引擎 analyze() 公开 API 的 VLM HTTP 调用与重试退避不可取消(AnalyzeParams 不收 signal,包装层无法把信号传入引擎内部的 adapter.call,受不改库源码约束)。上游库为 analyze 增加 signal 透传后可根治。
  • 8-hex 前缀为 32bit 空间,理论碰撞率低但非零;多命中时返回候选列表。

文档

  • docs/design.md — 设计文档(架构/运行时事实/验收标准,本仓内快照)
  • docs/VERIFY.md — 人工验收清单(七条)
  • vendor/vision-bridge/VENDOR.md — 引擎库来源与指纹

开发

npm run build:vendor   # tsc 出 vendor dist(改 vendor src 后)
npm test               # node --test(零依赖;递归含 vendor 测试)

开发环境注意:插件的 peer 依赖(@deepseek-ai/*)不在公共 npm,测试解析依赖把本目录 node_modules/@deepseek-ai/* 以 junction 指向 web profile 的 node_modules(已 gitignore)。新 clone 上跑测试前需自建这些 junction,或在 DSH profile 内开发。 另:settings-lifecycle 测试会临时改 process.env.DSH_HOME 以隔离持久化目录;node --test 每个测试文件独立子进程,无跨文件污染。

引擎库来源

vendor 自 shenhuanageshei/vision-bridge(快照 2026-09-04;逐文件 sha256 指纹见 vendor/vision-bridge/VENDOR.md)。源码逻辑零改动,仅补 package.json/tsconfig.json 并 tsc 出 dist;其 8 个 vitest 测试已移植为 node:test。

License

BSD-3-Clause(见 LICENSE)