dsh-design-qa
Design-fidelity QA for text-only models: a `deepseek_vision` tool borrows an eye from any OpenAI-compatible vision route, so the model can judge whether an implementation matches its mock — shipped with the benchmark behind that judgement (four fixtures, 23 injected defects, raw transcripts) and the questioning discipline it depends on.
给纯文本模型的设计稿保真判定：`deepseek_vision` 工具从任意 OpenAI 兼容视觉路由借来一只眼，让模型判断实现与设计稿是否一致——并附上支撑该判定的基准（4 组夹具、23 处注入缺陷、逐格原始输出）与其依赖的提问纪律。
How to install
dsh plugin add dsh-design-qa About
dsh-design-qa 简体中文 | English **让 DeepSeek Harness 里的纯文本模型能看图。** 做这个是为了在 DSH 上实现**产品设计**这项能力：从设计稿写出实现、再自己判断实现得像不像、 不像就修 —— 而 DeepSeek 写得了代码却看不见图，判定那一环做不了。本插件通过一个工具把 多模态能力借给它，让它具备执行产品设计功能所需的那只眼睛。 **任何纯文本模型都能读图。** 不只是 DeepSeek —— 你在 DSH 里自己接的那些 OpenAI 兼容端点 同样适用（实测过 OpenRouter 上的 z-ai/glm-5.2）。**前提是该模型支持 tool calling**： 它得能自己调 deepseek_vision。官方多模态上线那天，本插件自动让位、可原样留着。 **把识图做成一个 tool。** 图片不进主模型上下文；它看到一行 [图片 …] 提示，需要时自己调 deepseek_vision。不看就不产生任何成本，问什么由模型自己决定。 **附 eval 与原始输出。** 4 组夹具、23 处注入缺陷、四条通过线 —— 回答的是「借来的这只眼 够不够格当判定闭环里的裁判」，而不是「模型跑没跑通」。**能看见 ≠ 可用于判定**：看得见但不 主动看、会编、不稳、说不清，四种失效各对应一条通过线。真值四组齐全、有像素级注入…
Recommendation signals
Meta
- License
- MIT
- Language
- JavaScript
- GitHub stars
- 1
- mo. downloads
- –
- Last push
- 2026-08-20
- Created
- 2026-08-17
Basic safety check
- Findings
- None
- Sources
- curated:awesome-dsh-plugin.com, curated:awesome-dsh-plugin/awesome-dsh-plugin
- Topics
- benchmark, deepseek-harness, design-qa, design-review, dsh, dsh-plugin, llm-eval, multimodal, openai-compatible, qwen-vl, vision, visual-regression
Related plugins
dsh-vision-router
ysr666/dsh-vision-router
Free vision for text-only agents: built-in keyless vision chain plus pixel tools (Q&A, grounding, crop, pixel diff, colors, OCR, SVG trace, cutout, screenshots); paste an image to use it.
modlens
liustack/modlens
Vision bridge for text-only models: paste an image, get structured JSON evidence (OCR, layout, semantics).
dsh-vision-toolkit
Anionex/dsh-vision-toolkit
Vision for text-only models: paste an image and the model switches to a Vision Toolkit variant for image Q&A, multi-image comparison, long-screenshot OCR, screenshot-to-UI reproduction, element grounding, and pixel diff. No API key by default — images are processed by the author-hosted free service, 100 per machine per day; configurable to your own provider.
picturereader
jing-hy/picturereader
Image "reading" for text-only models: downscale + reduce color depth + structure/color fingerprints into text grids fed back to the conversation, letting the model zoom, sample and OCR autonomously like a multimodal model; fully local with zero external model dependency, ships an image-reading methodology skill and optional PaddleOCR.