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.

Sensenova Vision Aid — DSH Plugin for DeepSeek Harness
← Plugins
S

dsh-sensenova-vision-aid

Sensenova Vision Aid

Unofficial DeepSeek Harness vision plugin: when the main model receives an image-recognition request, it spawns a child agent switched to the SenseNova vision model (reusing the dsh-sensenova-freeapi credential), with a dedicated settings panel and a direct fallback path.

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

npx -y @deepseek-ai/dsh plugin --profile web add github:modesthub/dsh-sensenova-vision-aid#218b76310f2ac68326c9fdfca1d5c54dce0c1347
READMECompatibilityVersions

Compatibility and provenance

Sensenova Vision Aid is published as dsh-sensenova-vision-aid and currently resolves to version 0.1.0-alpha.1. The Hub verifies its manifest and preserves the exact installation source for reproducible installs.

DSH compatibility
*
Runtime surfaces
web
Release source
github
Registry updated
10/2/2026

Versions

0.1.0-alpha.1prerelease
10/2/2026

Related plugins

Loading related plugins…

Latest
0.1.0-alpha.1
DSH
*
HMR
Process restart
Tree shaking
Safe tree shaking not declared
Unpacked size
Unavailable
Files
Unavailable
Surface
web
License
MIT
Source
github
GitHub
★ 1
Weekly downloads
0
Last push
10/2/2026
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
DeepSeek Harness Plugin Hub
ProfilesPluginsCategoriesNewsDocsSign inManage Profiles
ProfilesPluginsCategoriesNewsDocsSign in

Related plugins

More verified plugins in vision-media.

Modlens@liustack/modlensPlug-in vision for text-only LLMs, powered by the free Antigravity CLITool 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 — Vision Routerdsh-vision-routerEyes for text-only DeepSeek Harness agents: built-in free vision chain + pixel-level tools. Remote vision providers, including the default free fallback, receive image content unless strict local-only use is configured.Vision Toolkit@anionex/dsh-vision-toolkitDeepSeek Harness-native integration for agent-vision-toolkit: image Q&A, OCR, grounding, UI restoration, pixel diff, Artifacts, and Web UI.

README

dsh-sensenova-vision-aid

简体中文 | English

dsh-sensenova-vision-aid 是一个非官方的 DeepSeek Harness(DSH)视觉辅助插件。主模型收到图片识别请求时,自动派生子 agent 并切换到 SenseNova 视觉模型(参考 @nanmicoder/dsh-agent-teams 子 agent 机制),提供专门的配置面板(主要配 api key;可勾选「复用 dsh-sensenova-freeapi 凭据」从而免配第二把 key)。未装 freeapi 或子 agent 不可用时自动回落直连兜底路径,识别永远有结果。

🔭 搭配推荐:与 dsh-sensenova-freeapi 搭配使用效果更好 —— freeapi 提供 sensenova LLM 路由(子 agent 主路径启用)+ 多账户 key 轮换,本插件自动复用 freeapi 同一把 key(默认 reuseFreeapiCredentials=true,无需再配第二把 key)。最合适的组合:主模型用 deepseek-v4-flash(freeapi,快且稳、成本低)处理文本/代码/Agent,图片识别交给 vision-aid 辅助视觉,又快又好。

架构设计与决策详见 docs/design.md。


功能一览

🖼️ 3 个全局工具(与 MCP 版逐字兼容)

  • describe_image(image, prompt?, model?, base_url?, image_mode?) —— 描述/理解一张图片
  • describe_images(images, prompt?, model?, base_url?, image_mode?) —— 同一请求内理解多张图片
  • chat(text, model?, base_url?) —— 纯文本对话(验证 key/端点连通性,视觉失败时的兜底)

返回 MCP 兼容 JSON 信封:{ok, task_type, tool_used, confidence, result, metadata.attempts}(attempts 记录每轮尝试)。

👶 子 agent 识别主路径

识别时派生子 agent(spawn provider)并切换到视觉模型(agentOptions:{provider:'sensenova', model},与 agent-teams spawnMember 同构);图片经附件服务投递为 ImageBlock,纯文本主模型全程看不到像素。预检失败(无 subagents / 无 spawn provider / sensenova 路由未注册 / 子 agent 失败)自动回落直连。

🔁 直连兜底路径

子 agent 不可用时直接调用 https://token.sensenova.cn/v1/chat/completions,按 sensenova-6.8-flash-lite → deepseek-flash → kimi-k3 故障转移(首个成功即返回),本地文件内联为 data: base64,redirect:'error' 防凭据跟随重定向。

🔑 凭据复用(免配第二把 key)

默认 reuseFreeapiCredentials=true,与 dsh-sensenova-freeapi 共用同一把 key(~/.dsh/.credentials.yaml 的 refs: 块或环境变量),解析顺序:credentials 服务 → launchEnvironmentOf → YAML 兜底。关闭复用后可在设置页输入自己的 key(写入 DSH 凭据服务,页面永不回显明文)。

⚙️ 专门配置面板

设置页「SenseNova 视觉辅助」段(order 14)+ Models 页只读卡片(provider-card slot key vision-sensenova),中文为主英文兜底。


安装

# 本地路径装入(推荐,同 freeapi README)
dsh plugin --profile desktop add D:\TRAE_Space\research-deepseek-harness\scratch-plugin-to-github\dsh-sensenova-vision-aid

# 发布后(GitHub / npm)
dsh plugin --profile <name> add github:modesthub/dsh-sensenova-vision-aid
dsh plugin --profile <name> add dsh-sensenova-vision-aid

cordis.patch.yml 随安装追加进 dsh.profile.bundles(条目 id vision-sensenova,与 freeapi 的 llm-sensenova 互不冲突);profile 重启(或 patchReload live)后生效。

前置依赖:

  • 推荐同时装入 dsh-sensenova-freeapi(提供 sensenova LLM 路由,子 agent 主路径启用);未装时插件仍可用(直连兜底),面板显示路由不可用提示。
  • 需要 profile 已挂 subagent-spawn-in-process(spawn provider)与 dsh-session-persistence(one-shot 子 agent 后端,同 tool-subagent 的要求)。

凭据准备(任一即可):

  • ~/.dsh/.credentials.yaml 的 refs: 块:SENSENOVA_API_KEY: sk-xxx(与 freeapi 相同)
  • 环境变量 SENSENOVA_API_KEY
  • 关闭「复用 freeapi 凭据」后在设置页直接输入(写凭据服务)

配置

设置命名空间 vision-sensenova(全部 .volatile()):

字段类型默认含义
reuseFreeapiCredentialsbooltrue复用 dsh-sensenova-freeapi 凭据(credential-ref 共享槽位,免配第二把 key)
keyRefcredential-refSENSENOVA_API_KEY凭据引用名(可切换 _2.._10 免费额度池)
apiBasestringhttps://token.sensenova.cn/v1chat-completions base URL
modelChainstringsensenova-6.8-flash-lite,deepseek-flash,kimi-k3直连故障转移模型链(逗号分隔)
imageModeenumimage_url图片传输模式(image_url / 遗留 image_base64)
timeoutMsint180000直连单模型尝试超时
useSubagentbooltrue识别走子 agent 路径开关
subagentProviderstringspawn子 agent provider 名
subagentModelstringsensenova-6.8-flash-lite子 agent 固定视觉模型(= 链首)
useContinuableboolfalse续接会话变体(预留,首版默认关)

API Key 一律经 DSH 凭据服务存储(credential-ref),永不写入日志、永不回显明文;诊断路由只回引用名与配置态(configured 布尔)。


目录结构

dsh-sensenova-vision-aid/
├── package.json              # dsh.bundle.patch / dsh.client.inject / dsh.client.platform / exports(./client)
├── tsconfig.json             # 照抄 freeapi(strict, moduleResolution:bundler, jsx react-jsx, noEmit)
├── tsdown.config.ts          # 双 config:lib(esm/node22/dts) + client(cjs/browser, ModuleLoader banner)
├── cordis.patch.yml          # 服务行声明(insert vision-sensenova 条目)
├── README.md / README.en.md  # 本文档(中英双语)
├── LICENSE                   # MIT
├── docs/
│   └── design.md             # 架构设计文档
├── src/
│   ├── index.ts              # host 入口:装配(工具注册 / 设置命名空间 / 诊断路由)
│   ├── config.ts             # VisionAidConfig + ConfigSchema(全部 .volatile())+ plainConfig() + resolveAdapterOptions()
│   ├── credentials.ts        # resolveApiKey:credentials 服务 → launchEnvironment → YAML 兜底(server.py 移植)
│   ├── image-input.ts        # 图片输入归一化:path/http/data:/base64 → 字节/mime;附件保存 / data: URL
│   ├── direct.ts             # 直连兜底:fetch 调 chat/completions,复刻 server.py 故障转移链(redirect:'error')
│   ├── subagent-vision.ts    # 子 agent 主路径:spawn one-shot(agentOptions 指定 provider/model;续接变体预留)
│   ├── orchestrator.ts       # 调度:describe_* 先试子 agent,失败回落直连;chat 恒直连;组装信封 + attempts
│   ├── tools.ts              # defineTool 定义 3 个工具(参数与 MCP 版逐字一致)
│   ├── status-api.ts         # GET /api/sensenova-vision-aid/status 与 POST /api/sensenova-vision-aid/test
│   └── client/
│       ├── index.ts          # client 入口:locale 注册、credentials 桥、configForms.get、slots 注入(双槽)
│       ├── section.tsx       # 设置页 React 组件
│       ├── card.tsx          # Models 页只读卡片
│       ├── settings.ts       # 领域控制器(无 JSX):快照 store / CredentialsFace / 草稿保存 / 测试连接
│       └── locales.ts        # zh/en 文案(settings.vision-sensenova 命名空间)
└── tests/
    ├── credentials.test.ts   # resolveApiKey 三态回退、ref 语法过滤
    ├── image-input.test.ts   # path/http/data/base64 归一化、mime 表
    ├── direct.test.ts        # 链式故障转移(注入 fetchImpl)、信封形状、redirect:'error'
    └── orchestrator.test.ts  # 子 agent 可用→走子 agent;不可用/失败→直连

开发

pnpm install
pnpm run typecheck   # tsc --noEmit
pnpm run build       # tsdown → lib/index.js + lib/index.d.ts + lib/client.js
pnpm test            # node --import tsx --test tests/**/*.test.ts(39 用例)

端到端验证步骤(详见 docs/design.md §7.3):

  1. chat('Reply with exactly: pong') → 信封 result: "pong"(直连路径、key 解析链路通)。
  2. describe_image(<本地测试图>) → 子 agent 路径成功,tool_used: "sensenova:sensenova-6.8-flash-lite";metadata.attempts 含子 agent 单条。
  3. 停用 freeapi 插件后重复 2 → 自动回落直连,attempts 逐模型记录(6.8 优先命中;强制 model="kimi-k3" 验证读图)。
  4. 设置页:勾选/取消「复用 freeapi 凭据」,保存后 host 用 GET /api/sensenova-vision-aid/status 核对 refs 配置态(只回引用名,不回 key)。

许可证

MIT