ysr666/dsh-vision-router
Eyes for text-only DeepSeek Harness agents: built-in free vision chain (no key) + pixel-level vision tools (Q&A, grounding, crop, pixel diff, colors, OCR, SVG trace, cutout, screenshots). One-command install, no Python, image turns work like ordinary tool-calling turns.
View Origin LinkProduct Positioning & Context
Related Ecosystem & Alternatives
Discover adjacent products, open-source repositories, and developer tools sharing similar technical architecture.
Deep-Dive FAQs
What is ysr666/dsh-vision-router?
Where did ysr666/dsh-vision-router originate?
When was ysr666/dsh-vision-router publicly launched?
How popular is ysr666/dsh-vision-router?
Which technical categories define ysr666/dsh-vision-router?
Are there active development issues for ysr666/dsh-vision-router?
What are some commercial alternatives to ysr666/dsh-vision-router?
How does the creator describe ysr666/dsh-vision-router?
Active Developer Issues (GitHub)
Community Voice & Feedback
我的仓库也包含这个功能,欢迎共建
- `packages/client/ui-conversation/src/client/service.ts` 的 `createDraftImages()` 仍直接用浏览器声明的 `file.type` 调 `imageMediaType()`,只接受 PNG/JPEG/WebP/GIF;
- `packages/client/ui-conversation/src/client/skeleton/InputBar.tsx` 的 `intakeImages()` 仍按 `file.type` 做整批预检,批内出现 `image/bmp` 就整批拒绝。
因此 Firefox/Windows 剪贴板把 DIB 暴露成 `image/bmp` 时,图片仍然会在进入附件栏之前被官方 composer 拒绝。这个阶段 Vision Router 还拿不到图片/附件,插件侧没有稳定、受支持的 API 可以做 MIME 归一化;强行做 DOM capture/paste monkey-patch 会和官方输入框生命周期、IME、拖拽/粘贴逻辑强耦合,回归风险明显高于收益。
结论:#138 继续保持打开并跟踪上游;暂不加入脆弱的 DOM 级 hack。最合理的修复仍应在官方 composer intake:按字节嗅探/把 BMP 转 PNG,并对同一剪贴板里的 DIB + FileDrop 做去重。
这条已单独拆到 #155,并在 PR #156 修复合入 `main`:隐藏的 onboarding / 引导状态持久化现在会按原始 user settings 做同值去重,同一写入即使 Host 拒绝或 snapshot 反复刷新,也不会再被 subscribe 链路无限重发;真正的状态变化仍正常写入。完整测试已通过。
#138 原本的 QQ/微信剪贴板格式问题仍继续作为 deepseek-harness 上游问题跟踪。
结论与你在「归属说明」中的判断一致:报错文案与校验逻辑都在 deepseek-harness 上游——
- 客户端 `packages/client/ui-conversation`:`imageMediaType()` 只精确匹配四个 MIME,`intakeImages()` 按整批拒绝;
- 服务端 `packages/attachment/attachment-local`:sharp 嗅探字节,声明与实际不符抛 `IMAGE_TYPE_MISMATCH`。
粘贴的图片在进入附件栏之前就被上游 UI 拦截,vision-router 插件层拿不到任何事件或数据,无法在插件侧兜底或绕过,因此本仓库无法单独修复,只能跟踪。
处理方式:
1. 你已在官方仓库开了 Discussion(https://github.com/deepseek-ai/deepseek-harness/discussions/2684 ),修复应在上游进行;你提出的按魔数嗅探字节、`image/bmp` 经 `createImageBitmap` → `canvas.toBlob("image/png")` 转码、粘贴批次去重等建议都很有价值,可一并提供给上游。
2. 本 issue 保持开放,作为该问题对 vision-router paste-and-go 流程影响的跟踪条目;上游修复发布后我们会回来回归验证并关闭。
给遇到此问题的用户的临时规避:
- Windows + Firefox 下不要直接 Ctrl+V 粘贴截图,先把截图保存为 PNG 文件(QQ/微信/系统截图都支持另存),再用附件按钮或拖拽上传;
- Chrome/Edge 下粘贴可用,但同图可能重复进入附件栏,提交前删掉重复项即可。
再次感谢对上游源码逐行定位的努力 👍
Discovery Source
GitHub Open Source Aggregated via automated community intelligence tracking.
Tech Stack Dependencies
No direct open-source NPM package mentions detected in the product documentation.
Media Tractions & Mentions
No mainstream media stories specifically mentioning this product name have been intercepted yet.
Deep Research & Science
No direct peer-reviewed scientific literature matched with this product's architecture.
SaaS Metrics