dsh-tool-ocr：在 DSH 里从图片与 PDF 抽取文字
dsh-tool-ocr 是 yuehancn 维护的 DSH 工具插件，用于从图片与 PDF 抽取文本：电子版 PDF 直接读文字层，扫描件与图片走用户配置的 OCR 引擎（如 Tesseract、PaddleOCR）。提供 ocr_read 与 ocr_status 两个工具，支持页码范围、语言提示与结果 note。安装命令为 dsh plugin --profile desktop add github:yuehancn/dsh-tool-ocr，配置写在 cordis.patch.yml。MIT 许可，针对 dsh 0.2.0-rc.2 构建测试。
Read More多模态数据源转换引擎：把文档、图片与音频统一转成 Markdown
介绍 SkillHub 社区 Skill「多模态数据源转换引擎」（@user_317b55bd/source-to-markdown，v1.0.1）。支持 md/txt/docx/pdf/pptx/xlsx/csv/html/json/xml 等文档与结构化文件（默认 MarkItDown），图片与扫描 PDF 走腾讯 OCR，音频走腾讯 ASR，异常可降级 MiMo 与火山引擎；含 ZIP 材料包场景。说明 skillhub CLI 安装、各 Agent skills 目录、requires_api_key 配置及典型 Agent 触发用法。
Read MorePdf To Markdown Knowledge：把 PDF 笔记抽成结构化 Markdown
Pdf To Markdown Knowledge（@user_bbd16bf9/pdf-to-knowledge，v1.0.3）是 SkillHub 社区知识管理类 Skill，将 PDF 笔记抽取为结构化 Markdown。支持文本 PDF 与扫描件 OCR；自动检测 Agent 的 PDF 能力，优先原生阅读，备选 PyMuPDF、pdfplumber、Tesseract。无需 API Key。安装提示：按 skillhub.cn/install/skillhub.md 安装该坐标。触发词含 PDF知识抽取、整理PDF、扫描件OCR 等，输出可导入思维导图。
Read MoreOCR-Local-1.0：基于 Tesseract.js 的本地图片文字识别 Skill
OCR-Local-1.0 是 SkillHub 社区 Skill（office-efficiency），用 Tesseract.js 在本地从图片提取文字，requires_api_key=false，支持简体中文、繁体中文与英文。本文说明其定位、核心能力，以及按 skillhub.cn/install/skillhub.md 使用 skillhub install @user_8353d908/ocr-rlocal 安装到 Cursor、Claude Code、Codex 等 skills 目录的步骤，并给出典型对话式用法与本地 OCR 的适用注意。
Read More智能图片转文字：把 Word 中的图片批量解析为文字
介绍 SkillHub 社区 Skill「智能图片转文字」（@user_8a153b62/smart-ocr，v1.0.0）。官方能力为将一组图片放在 Word 文件中并解析成文字，分类 office-efficiency，标签 requires_api_key=true。说明 skillhub CLI 与对话安装方式、Cursor 等 skills 目录路径，以及适用场景与未在页面核实的功能边界。
Read MorePDF万能大师：在 Agent 里做全能 PDF 处理
PDF万能大师（@user_e867e415/pdf-master）是 SkillHub office-efficiency 类社区 Skill，基于 Docling/Marker/OpenParse，提供 33 项 PDF 能力，含高保真转换、OCR、版式编辑、拆分、PDF→CAD、多模态 RAG 问答、合同对比与企业 DRM 等，飞书原生、无需 API Key。可通过 SkillHub 安装文档或 skillhub install 装入 Cursor 等 skills 目录，在对话中上传 PDF 完成转换与审阅类任务。
Read MoreMacOCR：在 macOS 上用 Vision 做图片文字识别
MacOCR（@user_9a66f56b/macos-ocr，v1.0.0）是 SkillHub 社区 Skill，通过 macOS Vision 框架从图片、截图与扫描件中提取文字，支持简繁中文与英文，无需 API Key。文章说明其定位、核心能力、skillhub 安装命令与各客户端 skills 目录，并列出平台仅限 macOS 等注意事项。
Read More交易截图报销助手：把支付截图整理成报销 Excel
交易截图报销助手（@user_863b76c8/expense-receipt-ocr，v1.0.0）是 SkillHub 社区 office-efficiency 类 Skill，将微信、支付宝、银行转账等截图经 OCR 抽取金额、日期、商户等字段，并汇总为含日期、金额、交易单号等列的报销 Excel。本文基于官方页面与 SKILL.md 介绍安装命令（skillhub install）、Cursor/Codex 等 skills 目录、MiniMax 或 OpenAI API Key 配置，以及单张与批量用法与注意事项。
Read More银行流水扫描件整理：扫描件 OCR 与结构化 Excel 导出
SkillHub 社区 Skill「银行流水扫描件整理」（@user_d2f40577/bank-statement-ocr，v1.0.0）面向律师与财务场景，将图片/PDF 扫描流水通过视觉大模型 OCR 提取日期、金额、对手与摘要，标准化后按银行与开户人生成 Excel，支持批量与多 Worker 并行。安装需 SkillHub CLI 与 requires_api_key 的视觉 API。本文介绍定位、核心能力、安装路径与典型自然语言触发方式，并说明隐私与识别质量注意事项。
Read More翻译小助手：自动判向的图文翻译 Skill
翻译小助手（@user_674c7529/ttk5835-1）是 SkillHub education 类社区技能，自动识别语种并按固定规则翻译：中文译英文，英/日/韩/法/德/西班牙语译中文。支持文本直译与图片 OCR 后对照输出原文与译文，交互强调只给结果、少反问。requires_api_key=false。安装需 skillhub CLI，并用 --dir 指向 Cursor 等 Agent 的 skills 目录。适合文档对照与截图外文处理，专业术语仍需人工校对。
Read More翔云 OCR 表格识别 文字识别-Pro：在 Agent 里做文档与表格 OCR
介绍 SkillHub 社区 Skill「翔云 OCR 表格识别 文字识别-Pro」（@user_6b190ef3/xiangyun-table-ocr，v1.0.10）。说明其作为翔云通用文档/表格识别 Agent 的定位，涵盖文字与 PDF OCR、表格提取、转 Excel/Word/Markdown、批量识别、多语言及倾斜校正等官方列出的能力；标注 requires_api_key=true。给出 skillhub install 与 AI 安装提示两种启用方式，以及基于官方触发语义的对话示例与使用注意。
Read More基于深度学习图片与文字本地脱敏工具：在 Agent 工作流里做本地 PII 与截图脱敏
介绍 SkillHub 社区 Skill「基于深度学习图片与文字本地脱敏工具」（@user_cf24022c/image-text-desensitization，v1.0.0，it-ops-security，无需 API Key）。说明其通过 smart_desensitize_skill 在本地处理文本、JSON 与图片（OCR/模糊），SkillHub CLI 安装到 Cursor/Codex 等 skills 目录的方式，以及依赖、典型 Python 调用与自动写入 DESENSITIZE_RULES、MEMORY 的集成注意点。
Read MorePDF EDITOR：在 Agent 里完成 PDF 编辑、转换与提取
PDF EDITOR（@user_09701b7c/mainpdf）是 SkillHub 社区办公效率类 Skill，v1.0.1，无需 API Key。支持 PDF 合并拆分、插入图文、文字表格图片提取、扫描件 OCR，以及转 Word/图片/HTML 与文档总结。本文说明其定位、能力与 Trigger，并给出基于 skillhub.cn 安装文档的 CLI 安装步骤及 Cursor 等客户端 skills 目录路径，典型用法以对话触发为主。
Read More腾讯云护照识别（多国多地区）：为 Agent 封装 MLIDPassportOCR
SkillHub 上的 @tencent-adm/tencentcloud-ocr-mlidpassport（v1.0.3，腾讯云文字识别团队）将 MLIDPassportOCR 封装为 ai-agent Skill，支持 80+ 地区护照、MRZ/视读区字段、Base64 与 URL 输入、人像裁剪及国际站质量告警。文章说明安装命令、TENCENTCLOUD 密钥与 scripts/main.py 典型调用，便于在 Cursor 等 Agent 中稳定识别护照信息。
Read More邮箱发票自动整理：从邮箱拉附件到 A4 合并与费用清单
SkillHub 社区 Skill「邮箱发票自动整理」（@user_b8af0005/invoice-from-email，v1.1.2）面向报销归档：从邮箱搜索下载发票与行程单，经 PyMuPDF/Tesseract/PaddleOCR 自动降级提取文字，合并排版到 A4（火车票按真实尺寸上下排版），并生成单 Sheet 费用清单 Excel。官方称零 OCR 可处理九成以上电子发票。本文基于 SkillHub 页面与安装文档说明安装路径、CLI 与适用注意，需配置 API Key（requires_api_key=true）。
Read More本地 OCR 识别 PDF 转 Markdown：扫描件批量出可检索文本
SkillHub 社区 Skill「本地ocr识别pdf转md」（@user_543d2acf/pdf-ocr-zc，v1.0.0）面向扫描 PDF 批量 OCR：生成带文字层 PDF，并导出 Markdown 或纯文本，标签 requires_api_key=false。文章说明定位、核心能力、skillhub CLI 安装命令与各 Agent skills 目录，以及基于官方简介的可复现工作流与使用注意。
Read More文本识别 OCR：在 Agent 里用 Skill 做图像文字检测与识别
介绍 SkillHub 社区 Skill「文本识别 OCR」（@user_628f4cb3/ocr，v1.0.0）：输入含字图像并自动检测识别，适用于文档、广告牌与截图。说明 skillhub CLI 安装、各 Agent skills 目录、requires_api_key 配置要点，以及通过对话调用的典型流程；附官方技能页与 API 入口。
Read MoreDeepSeek Harness 插件：dsh-ocr-review
DeepSeek Harness (DSH) 的 `dsh-ocr-review` 插件将 `open-code-review` 外部 CLI 工具封装为 5 个模型工具和 1 个 Web 设置卡片，实现了参数验证、配置管理及结构化摘要返回。使用前需安装宿主环境及外部 CLI（版本 1.13.0），并注意 macOS/Linux 的兼容性及 Windows 的孤儿进程风险。
Read MoreOCR 文字识别：在 Agent 里做截图与图片 OCR
OCR 文字识别是 SkillHub 社区技能 @user_328225df/screenshot-ocr-pro（v1.0.1，office-efficiency），用于在 Agent 中提取截图与图片文字，支持中英文、批量处理与表格还原，且标注 requires_api_key=false。本文基于 SkillHub 公开页与安装文档说明定位、能力与 skillhub CLI 安装方式，并提示用法以安装包内 SKILL.md 为准。
Read MoreOCR中英文识别加强版：多引擎 OCR 与 LLM 纠错的三层架构
OCR中英文识别加强版（@user_6455574a/ocr-pro-v2，v2.1.1）是 SkillHub office-efficiency 类社区 Skill，采用多引擎集成、专项预处理与 Vision LLM 纠错三层架构，支持 PDF 与图片，面向中英文印刷体、手写体与数学公式。引擎含 RapidOCR、EasyOCR、TrOCR、Pix2Text、PP-OCRv5、腾讯 OCR（MCP）等。标签 requires_api_key=false。可通过 skillhub install 安装到 Cursor、Claude Code、Codex 等 skills 目录，或按官方提示让 AI 参照 skillhub.md 安装。
Read More智绘图片理解：在 Agent 里统一处理识图、OCR 与多图对比
智绘图片理解（@ixhlink/ixhlink-skills-image-explain）由星华智联科技发布，面向 design-media 场景，支持单张或多张图片的内容描述、视觉问答、文字识别与对比分析。用户说出识图、看图、描述图片、提取图片文字、对比图片等意图时可触发。SkillHub 标注为付费 Skill、requires_api_key=false，当前版本 1.0.3。安装方式为按 https://skillhub.cn/install/skillhub.md 安装上述坐标。文章说明能力边界、典型对话用法与使用注意，并给出官方 SkillHub 与 API 地址。
Read MorePDF and Image Text Extraction: Extracting Text and Tables from Images and PDFs in an Agent
This article introduces the SkillHub Skill “PDF and Image Text Extraction” (@user_5f9c21aa/pdf-image-text-extractor, v1.0.13). It supports text recognition from multiple image formats and PDFs, AI visual OCR for scanned documents, table extraction using PyMuPDF, and batch processing of directories; the v2.1 solution does not require tesseract or pdfplumber. It explains the `skillhub` installation command, the Cursor skills directory, and `requires_api_key` considerations, with capabilities provided by RedFoxHub.
Read Moredsh-umi-ocr-vision: Umi-OCR Vision Bridge Plugin for DeepSeek Harness
The DeepSeek Harness plugin `dsh-umi-ocr-vision` uses a plugin-based architecture to provide visual understanding to text-only main models (such as `deepseek-official`). It leverages the locally running Umi-OCR tool to automatically recognize image contents and transform them into “untrusted observation data” injected into the context. The plugin supports multiple attachments and both HTTP and CLI modes, ensuring that data remains local, and offers a rich set of `vision_*` tools. Installation is simple: Umi-OCR must be started first, and the relevant parameters configured in the configuration file.
Read MoreDSH Screen Agent: Windows Screen Recognition and Automation Plugin
DSH Harness (DSH) introduces the Lak321/dsh-screen-agent plugin, providing DeepSeek with desktop and web automation capabilities. The plugin is implemented using PowerShell and the Windows Host, supporting screenshots, OCR, mouse and keyboard control, and Paint operations without third-party dependencies. Installation must be performed by running the specified command through the Bundle layer stack. Commands can be issued via the HTTP interface or agent conversations. Note that it is limited to the Windows platform, where DPI scaling must be handled; when using the newer version of Paint, interaction issues such as color palette freezes and drag failures should be avoided.
Read Moredsh-quicksight: DeepSeek Harness Dual-Layer Image Recognition Plugin
The dsh-quicksight plugin in DeepSeek Harness (DSH) supplements image recognition for text-only models using a two-layer "local OCR + vision model" strategy. Maintained by Isanti2016 and released under the MIT license, the plugin follows a "fast first, accurate later" approach: it first attempts rapid local text extraction with RapidOCR, which takes about 2–3 seconds, runs fully offline, and incurs no cost. If the result is insufficient—such as too few extracted characters—it falls back to a vision model for deeper analysis. Installation requires configuring a Python environment to run Tier-1, and optionally configuring modlens to run Tier-2. The plugin does not listen on any ports, and Tier-1 processes data purely locally to protect privacy. It supports common image formats but does not support PDFs.
Read Moredsh-tesseract-ocr: Connecting text models to local image OCR
`dsh-tesseract-ocr` is a DeepSeek Harness (DSH) plugin designed to address the limitation where text models cannot process image inputs. By invoking the local Tesseract OCR engine to recognize text in attached images, the plugin sends the recognized plain text to the model API. By default, it retains the original image bytes locally without uploading them to the model provider, thereby avoiding changes to the model's input modality or the transmission of sensitive data. Core features include support for text models to process image attachments, local OCR recognition, sending only text by default, an optional visual model passthrough mode (`passthrough: true`), and a fail-closed security mechanism that rejects image attachments when the plugin is not loaded. It is suitable for scenarios where DSH text models are used and require image-text Q&A.
Read Morepicturereader: Adding Local Image Recognition Capability to the Pure Text DSH Model
picturereader is a DSH model inference plugin (v3.2.0) maintained by jing-hy. It enables pure-text models to achieve native thumbnails through a visual twin adapter, converting images into structured textual evidence using local tools such as image_scan and image_ocr. It supports three routing modes—privacy, intelligence, and rigor—and an optional external VLM. Additionally, it includes document_to_image for converting documents to images and image_edit for local image editing. It runs entirely locally by default and is licensed under MIT.
Read MoreConnect Image and Audio Capabilities to DeepSeek Harness with dsh-vision-complete
`dsh-vision-complete` is a MIT-licensed community plugin for DeepSeek Harness maintained by Yts1919, which equips plain-text models with capabilities including image understanding, OCR, object detection, video processing, speech recognition and PDF handling. The plugin itself does not include any vision models, and by default calls Qwen via MCP. The alternative `vision.py` can be connected to OpenAI-compatible interfaces. The installation command via the catalog page is `dsh plugin add github:Yts1919/dsh-vision-complete`. For full activation, you also need the Windows installation script, a valid DASHSCOPE_API_KEY, and a restart of Harness.
Read MoreEquip Plain-text DeepSeek Harness Agent with "Eyes" Using dsh-vision-router
### 版本1（正式学术/技术文档译法，适配海外开发者语境）： DSH-Vision-Router is a DSH tool and capability plugin (MIT License, v1.4.4) maintained by ysr666. It treats vision models as its "eyes" and DeepSeek as its "brain", and provides free-key-free anonymous fallback service from OVH by default, along with 11 pixel-level tools including question answering, positioning, cropping, pixel comparison and OCR. This article verifies the installation commands, usage of automatic image recognition model groups and permission precautions in accordance with the community directory and GitHub README. --- ### 版本2（更贴合国内开源项目对外宣传的自然译法）： Maintained by ysr666, DSH-Vision-Router is a DSH utility and capability plugin under MIT License (v1.4.4). It takes vision models as its "eyes" and DeepSeek as its "brain", offers free anonymous fallback service powered by OVH by default, and has built-in 11 pixel-level tools such as Q&A, positioning, cropping, pixel comparison and OCR. This article checks the installation commands, usage of the automatic image recognition model suite and permission notes against the community repository and GitHub README.
Read MoreAdding Visual Capabilities to DeepSeek Harness Plain Text Model Using dsh-vision-toolkit
DSH Vision Toolkit is a DeepSeek Harness interface enhancement plugin maintained by Anionex, which connects agent-vision-toolkit to Web/Headless Profile. After pasting images with a plain-text model, users can switch to the Vision Toolkit variant to complete intent-based image question answering, element positioning, long-screenshot OCR and UI restoration with pixel comparison. The directory installation command is `dsh plugin add github:Anionex/dsh-vision-toolkit`, it is licensed under MIT, and the current npm version is 0.1.20.
Read MoreEnd-to-End Recognition of Captchas Based on PaddlePaddle 2.0
Your code has covered most aspects of the CAPTCHA recognition project, including data processing, model training, and inference. Below are some suggestions for improvements and enhancements to your provided code: ### 1. Data Preprocessing Ensure the image dimensions are consistent (27x72), as this is the input size used during training. ### 2. Model Definition Your `Model` class has already encapsulated the network structure well. You can further optimize it and add more comments to facilitate understanding. ### 3. Training Process During the training process, ensure that when using multi-GPU training,
Read More