Redact

核心功能：基于OCR识别，按CSV规则替换或遮蔽敏感信息，支持多格式，不同格式有对应遮蔽方式（图片/PDF用色块，文档/PPT用█）。典型场景：文件分享前隐私合规处理、匿名化内容、替换姓名/手机号等PII。输入为目标文件和规则CSV，输出处理后的文件，也可提取文本生成JSON结果。

darknoah

热度 46.70

最近更新 2026-05-11 17:34:51

分类：效率依赖：代码执行 / 读写文件应用场景：数据处理 / 法律文档

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add redact

Metadata

name: redact
description: Privacy redaction toolkit for images, PDFs, Word documents, and PowerPoint presentations. Use when the user needs to redact, mask, or replace sensitive/private information in files. Triggers: - Redacting or masking sensitive text in images, PDFs, documents, or presentations - Replacing names, phone numbers, IDs, or other PII in files - Processing privacy compliance for documents before sharing - Anonymizing content in visual files Supported formats: png/jpg images, PDF, docx/doc, pptx/ppt

Redact Skill

Privacy redaction toolkit using PPStructureV3 OCR for text detection and replacement.

Scripts

Script	Format	Command
`read.py`	Images / PDF / Word / PowerPoint	`read.py <input> [--info] [--mode json]`
`redact-image.py`	Images (png, jpg, etc.)	`redact-image.py <input> <rules.csv> <output>`
`redact-pdf.py`	PDF	`redact-pdf.py <input> <rules.csv> <output>`
`redact-document.py`	Word (docx, doc)	`redact-document.py <input> <rules.csv> <output>`
`redact-presentation.py`	PowerPoint (pptx, ppt)	`redact-presentation.py <input> <rules.csv> <output>`

CSV Rules Format

target_text,replacement_text
张三,李四
手机号,
身份证号,

Rule	Effect
`原文本,新文本`	Replace with new text
`原文本,`	Empty = mask with █ (documents) or solid color block (images/PDF)

Masking Behavior

Format	Empty Replacement
Images, PDF	Solid color block overlay
Word, PowerPoint	`█` characters (same length as target)

Read Features

read.py supports:

Reading text from images, PDF, Word, and PowerPoint files
OCR for image files and embedded images
Page-aware output for PDF / Word / PowerPoint
--info structured output:
- <page index="...">
- <image format='markdown' type='ocr'>...</image> for OCR text extracted from images

JSON Output

Document-like files (pdf, docx, doc, pptx) output:

{
  "type": "pptx",
  "pages": [
    {
      "page_index": 1,
      "content": [
        { "type": "text", "text": "..." },
        { "type": "image", "text": "ocr text..." }
      ]
    }
  ]
}

Image files output:

{
  "type": "image",
  "content": "..."
}

Features

Feature	Image	PDF	Document	Presentation
Read text	✅	✅	✅	✅
JSON output	✅	✅	✅	✅
Text replacement	✅	✅	✅	✅
Solid color mask	✅	✅	-	-
█ character mask	-	-	✅	✅
OCR detection	✅	✅	✅ (images)	✅ (images)
Tables	-	✅	✅	✅
Headers/Footers	-	✅	✅	-
Embedded images	-	✅	✅	✅

Environment Setup

使用 uv 安装依赖：

# 进入 skill 目录
cd skills/redact

# 同步依赖（自动创建虚拟环境并安装）
uv sync

Dependencies

Python 3.10+
PaddleOCR / PPStructureV3
python-docx, python-pptx, PyMuPDF, Pillow