---

name: redaction-tool description: | Development skill for CaseMark's Smart Redaction Tool - an intelligent document redaction application with two-pass PII detection combining regex patterns and AI-powered semantic analysis. Built with Next.js 14, pdf.js/pdf-lib, and Case.dev LLMs. Use this skill when: (1) Working on the redaction-tool codebase, (2) Adding or modifying regex patterns, (3) Implementing AI detection features, (4) Building PDF export functionality, or (5) Adding new PII types.

Redaction Tool Development Guide

An intelligent document redaction application with two-pass PII detection—regex patterns for standard formats plus AI semantic analysis for contextual data.

Live site: https://redaction-tool.casedev.app/

Architecture

Copysrc/
├── app/
│   ├── page.tsx                    # Main application UI
│   └── api/
│       ├── detect-pii/             # Two-pass PII detection
│       ├── export-pdf/             # PDF generation
│       ├── detect/                 # Database-backed detection
│       ├── export/                 # Database-backed export
│       ├── jobs/                   # Job management
│       └── upload/                 # File upload
├── components/
│   ├── redaction/
│   │   ├── PatternSelector.tsx     # Redaction type selection
│   │   ├── EntityList.tsx          # Detected entities
│   │   └── DocumentPreview.tsx     # Preview with highlights
│   ├── upload/
│   │   └── DropZone.tsx            # File upload
│   └── ui/                         # shadcn components
└── lib/
    ├── redaction/
    │   ├── detector.ts             # Two-pass detection logic
    │   └── patterns.ts             # Regex patterns & presets
    ├── case-dev/
    │   └── client.ts               # Case.dev API client
    ├── db.ts                       # Database connection
    └── utils.ts                    # Utilities

Core Workflow

CopyUpload Doc → Extract Text → Pass 1: Regex → Pass 2: AI → Review → Export PDF
     ↓            ↓             ↓              ↓           ↓          ↓
  PDF/TXT      pdf.js       SSN, CC,       Names,      Toggle      Redacted
  images       extraction   phone, email   addresses   entities    document
                            patterns       context

Tech Stack

Key Features

Two-Pass Detection

See references/pii-detection.md for patterns and AI prompts.

Pass 1: Regex (Fast, High-Precision)

• SSN: XXX-XX-XXXX with validation
• Credit cards: Luhn-valid patterns
• Phone: US formats
• Email: Standard format
• Dates: Common formats

Pass 2: AI/LLM (Semantic, Aggressive)

• Non-standard formats ("SSN: one two three...")
• Contextual references ("my social is...")
• Names and addresses
• OCR errors and typos
• Obfuscated data

Redaction Presets

Case.dev Integration

See references/casedev-redaction-api.md for API patterns.

LLM Detection

Copyconst aiEntities = await detectWithLLM(text, piiTypes);

OCR for Images

Copyconst text = await extractTextFromImage(imageUrl);

Development

Setup

Copynpm install
cp .env.example .env
# Add CASEDEV_API_KEY
npm run dev

Environment

CopyCASEDEV_API_KEY=sk_case_...               # Required
DATABASE_URL=postgresql://...             # Optional for job persistence

Database (Optional)

Copynpx prisma migrate dev

API Endpoints

Common Tasks

Adding a New PII Type

1. Add regex pattern to lib/redaction/patterns.ts
2. Add to AI prompt in lib/redaction/detector.ts
3. Add UI toggle in PatternSelector.tsx
4. Add masking function

Improving AI Detection

Modify the LLM prompt to be more/less aggressive or handle specific formats.

Adding Export Format

1. Add generation function in lib/export/
2. Add endpoint in api/export-[format]/
3. Add UI option

Troubleshooting