Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add DDDFXYqiming/dsh-vision-skill --skill dsh-vision-skillgit clone --depth 1 https://github.com/DDDFXYqiming/dsh-vision-skillWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/dddfxyqiming/dsh-vision-skill/dsh-vision-skill)<a href="https://agentmods.dev/skills/dddfxyqiming/dsh-vision-skill/dsh-vision-skill"><img src="https://agentmods.dev/badge/skills/dddfxyqiming/dsh-vision-skill/dsh-vision-skill/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/dddfxyqiming/dsh-vision-skill/dsh-vision-skill"><img src="https://agentmods.dev/badge/skills/dddfxyqiming/dsh-vision-skill/dsh-vision-skill.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00050 | $0.02869 |
| Opus 5.5 | $0.00020 | $0.01148 |
| Sonnet 5.5 | $0.00010 | $0.00574 |
| Haiku 4.5 | $0.00005 | $0.00287 |
Grade A, and why
vision scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
This is a copy
100% identical to vision — 0 lines differ, which has more behind it and is treated as the original. This page carries a canonical link to it rather than competing with it.
How it starts
The opening of the file, as written. The whole thing — 120 lines — stays where its author put it; the contents beside it link to each section on GitHub.
识图技能（DSH 标准插件版）
当主模型不支持直接读取图片时，图片不会进入对话上下文，但本插件现在会：
- 直接贴图：client 插件在粘贴进入 DSH 附件管线前截获图片，输入框显示
📎 image1.png / image2.png ...编号 chip（从 1 开始）；发送时自动上传到工作区.dsh-vision/pasted/并把路径文本交给模型（无需任何框架补丁）； - 或用户消息中本来就带图片本地路径（形如
C:/Users/.../xxx.png）。
图片如何进入工具（三种方式）
本插件的所有工具只接收图片的本地路径（文本），不接收图片本体。路径来源有三种（前两种无需 pi-ai 补丁）：
- 路径直发（最常见）：用户消息中的图片会以路径文本出现——可能是附件占位符（
[图片附件 sha256:...，本地路径 C:\Users\...png，模型不支持直接读图，请用 vision skill 读取]），也可能是用户直接给出路径。直接调用vision_analyze等工具即可。 - 剪贴板：用户说"看图"且图片在剪贴板（如 Win+Shift+S 截屏自动复制）→ 调用
vision_clipboard，它会自动把剪贴板图片保存到工作区.dsh-vision/再识别。 - 直接贴图：本插件自带 paste-to-path client——在输入框粘贴图片时，图片先上传到工作区
.dsh-vision/pasted/，再以路径文本进入消息；消息里没有 image 块，因此 DSH 不会报MODEL_DOES_NOT_SUPPORT_IMAGES。旧 pi-ai 补丁仅作为兼容保留，不再必需。
DeepSeek Harness（DSH）插件模式
本技能已打包为标准 DSH 插件 @dsh-external/dsh-vision-skill（工具 + 运行时 skill，无需任何框架补丁）：
vision_analyze：识别指定路径的本地图片（image_path必填；可选mode/prompt/crop/budget）。mode=evidence返回结构化证据 JSON（summary / ocr_full_text / layout 阅读顺序 / semantics 实体关系 / uncertainty）；多 provider 自动 failover，429 自动退避vision_ocr：独立 OCR 工具——提取图片中全部可见文字，保持原始排版（image_path必填；可选prompt/crop/budget）vision_ground：定位工具——在图片中查找指定目标（如「所有按钮」「微信图标」），返回每个目标的像素坐标框（bbox_pixel）与归一化坐标（bbox_normalized，0-1000），可选output保存带标注框的预览图vision_detect：枚举工具——清点图片中某一类元素（默认所有 UI 元素），逐个编号 + 像素坐标框；与vision_ground互补（ground 找一个，detect 数一类）vision_dominant_colors：主色分析——提取图片（或区域）主要颜色与占比（本地像素算法，无需视觉 API），用于取主题色/配色分析vision_long_screenshot_ocr：超长截图分块 OCR——聊天记录/整个网页等超高图自动切块（带重叠）→ 逐块识别 → 合并全文，带块边界信息；每块先跑本地 tesseract（chi_sim+eng），失败自动回退 VLMvision_clipboard：读取剪贴板中的图片，保存到会话工作区.dsh-vision/后识别——用户在输入框粘贴图片被"当前模型不支持图片"拦截时，只需把图片复制到剪贴板（如 Win+Shift+S 截屏自动复制）后说"看图"即可- 渐进式工具暴露：加载本 skill 后自动为当前 Agent 激活上述 7 个识图工具；若工具未出现，调用一次
vision_activate兜底 - 模型配置走插件 config：
apiUrl/model/apiKey（任意 OpenAI 兼容的多模态模型均可接入——如 Qwen-VL、MiniMax-M3、Gemini、GPT-4o 等；默认 MiniMax-M3）。新增visionProviders数组可配置 fallback 链路（顺序=优先级，自动 failover；429 按 Retry-After 退避重试一次）。密钥支持 DSH Credential 引用（credential: VISION_API_KEY），推荐后者避免明文 - 分辨率预算：
budget支持small(≈512²) /normal(≈1024²) /large(≈1448²) /mega(≈4096²，约 16M 像素超高清，对应 Qwen 官方高分辨率模式) - 识别流程：脚本输出描述后原样转述，重要文字、报错信息逐字复述，不概括、不脑补
What ships with it
5 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago Changed f2b0bb1b6e57
- 2mo ago First seen · 120 lines · 50 tokens per session scan A 6ab7edded864
vision is a skill published in the GitHub repository DDDFXYqiming/dsh-vision-skill (2 stars, last pushed 19d ago), licensed MIT. It adds 50 tokens to every session and 2,869 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it A with 0 findings. It is 100% identical to vision, differing in 0 lines, and is treated as a copy.
Other skills, from other repositories
dsh-paddleocr-skills-setup
Install and configure the native PaddleOCR plugin for DeepSeek Harness (DSH) from the Settings → PaddleOCR GUI. Use for OCR and image-to-text from screenshots, scans, and PDFs; Chinese/CJK text; PDF-to-Markdown; structured document parsing with tables, formulas, layout, and reading order; or DSH endpoint, credential…
paddleocr-doc-parsing
Use the native DeepSeek Harness PaddleOCR tool to parse document layout and Markdown.
paddleocr-text-recognition
Use the native DeepSeek Harness PaddleOCR tool to extract text from an image or PDF.
manage-taskboard
Manage work in the native DeepSeek Harness Taskboard with exact task ids and optimistic versions. Use when an Agent must inspect project work, claim an eligible todo, record progress or blockers, verify an implementation, submit it for human review, or release its own claim; also use when a human asks how to accept…
company-research-method
A documented process for researching a company using supplied files and selected public sources. It produces a company overview with business details, sourced financial figures, and stated risks.
dsh-plugin-guide
Use when developing, reviewing, packaging, debugging, or answering questions about DeepSeek Harness (DSH) plugins — the plugin-based agent harness on vendored Cordis. Applies the official plugin-development constraints (plugin contract, cordis.yml layers, services/events/effects, tool DSL, bundles/profiles) backed by…