Document OCR Skill (docr)

Uses Gemini 2.5 Flash, PaddleOCR, or RapidOCR (local) to recognize text from scanned PDFs and images. Compiled as a single Go binary.

Prerequisites

• API Key configured in ~/.ocr/config (not needed for Paddle/Rapid)
• For RapidOCR engine: pip install rapidocr_onnxruntime
• For PaddleOCR engine: pip install paddleocr paddlepaddle

API Key Configuration

Create the config file:

mkdir -p ~/.ocr
cat > ~/.ocr/config \x3C\x3C EOF
# Google Gemini API Key
gemini_api_key=your_gemini_key
EOF

Quick Start

Path Variable: All commands below use $DOCR. Before running any command, set this variable:

SKILL_DIR="$(cd "$(dirname "\x3Cpath-to-this-SKILL.md>")" && pwd)"
DOCR="$SKILL_DIR/scripts/docr/docr"

# OCR a single document using RapidOCR (default)
$DOCR document.pdf
$DOCR image.jpg

# Use Gemini engine
$DOCR -engine gemini document.pdf

# Use PaddleOCR local engine
$DOCR -engine paddle document.pdf

# Specify output file
$DOCR document.pdf -o result.txt

# Batch process all supported files in a directory
$DOCR -batch ./docs/ -o ./outputs/

Engines

CLI Reference

docr [options] \x3Cfile or directory>

Options:
  -engine string   OCR engine: rapid (default) / gemini / paddle
  -e string        Engine (short flag)
  -o string        Output file path or directory (batch mode)
  -output string   Output path (long flag)
  -batch           Batch mode: process all files in directory
  -prompt string   Custom recognition prompt (gemini)

Installation

We provide pre-compiled binaries to get you started quickly.

cd doc-ocr-skills/scripts
./install.sh

This script will detect your OS (darwin/linux) and architecture (amd64/arm64) and download the appropriate version of docr.

Building from Source (Optional)

If you prefer to build from source, ensure you have Go 1.21+ installed:

cd doc-ocr-skills/scripts/docr
go build -o docr .

Error Handling