Smart Repomix

An AI-powered wrapper for repomix that intelligently filters unnecessary files from your codebase using Google's Gemini AI.

What it does

Smart Repomix enhances the standard repomix tool by:

1. Running repomix in-memory to scan your entire codebase
2. Using AI to analyze which files are unnecessary based on your instruction
3. Re-running repomix with intelligent ignore patterns
4. Producing a cleaner, more focused output

Installation

Option 1: Use directly with npx (recommended)

1npx smart-repomix -I "ignore test files and build artifacts"

Option 2: Install globally

1npm install -g smart-repomix
2smart-repomix -I "ignore test files and build artifacts"

Option 3: Install as a dev dependency

1npm install --save-dev smart-repomix

Prerequisites

• Node.js >= 14.0.0
• A Google Gemini API key (get one here)
• npx (comes with npm 5.2+)

Quick Start

For local development:

1# Clone/download the package files
2git clone <your-repo-url>
3cd smart-repomix
4npm install
5
6# Set API key
7export GEMINI_API_KEY="your-api-key"
8
9# Test locally
10node cli.js -I "ignore test files" --dry-run

See QUICKSTART.md for detailed setup instructions.

Configuration

Smart Repomix can be configured through your existing repomix.config.json file:

1{
2  // Your regular repomix config...
3  "output": {
4    "filePath": "my-output.xml"
5  },
6  
7  // Smart Repomix specific config
8  "smartRepomix": {
9    "apiKey": "your-gemini-api-key",
10    "promptTemplate": "Custom prompt with {{instruction}} and {{codebase}} placeholders"
11  }
12}

Configuration Options

• smartRepomix.apiKey: Your Gemini API key (alternative to env var or CLI option)
• smartRepomix.promptTemplate: Custom prompt template with placeholders:
  • {{instruction}}: Will be replaced with your -I instruction
  • {{codebase}}: Will be replaced with the XML codebase content

Custom Prompt Template Example

1{
2  "smartRepomix": {
3    "promptTemplate": "Analyze this {{codebase}} and identify files to ignore based on: {{instruction}}. Return only comma-separated glob patterns."
4  }
5}

Usage

Basic usage

1# Uses your repomix.config.json for output settings
2npx smart-repomix -I "ignore test files, documentation, and build outputs"
3
4# Or specify output explicitly
5npx smart-repomix -I "ignore test files" -o my-output.xml

Advanced options

1# Use a different Gemini model
2npx smart-repomix -I "ignore test files" -m gemini-2.5-pro
3
4# Verbose output
5npx smart-repomix -I "ignore test files" -v
6
7# Dry run (see what would be ignored without creating output)
8npx smart-repomix -I "ignore test files" --dry-run
9
10# Pass API key directly
11npx smart-repomix -I "ignore test files" -k "your-api-key"

Command Line Options

• -I, --instruction <instruction> - Required. Instruction for AI to determine which files to ignore
• -o, --output <path> - Output file path (defaults to repomix config or repomix-output.xml)
• -k, --api-key <key> - Gemini API key (or use GEMINI_API_KEY env var or config)
• -m, --model <model> - Gemini model to use (default: gemini-2.5-flash)
• -v, --verbose - Show verbose output
• --dry-run - Show what files would be ignored without creating final output

API Key Priority

The API key is resolved in this order:

1. -k command line option
2. GEMINI_API_KEY environment variable
3. smartRepomix.apiKey in repomix.config.json

Example Instructions

Here are some effective instruction examples:

1# For sharing code for review
2npx smart-repomix -I "ignore test files, build artifacts, and documentation"
3
4# For production code analysis
5npx smart-repomix -I "ignore development files, tests, examples, and build outputs"
6
7# For debugging
8npx smart-repomix -I "keep only source code and configuration files"
9
10# For documentation
11npx smart-repomix -I "ignore node_modules, build directories, and temporary files"

How it Works

1. Initial Scan: Runs repomix --stdout to capture your entire codebase in memory
2. AI Analysis: Sends the codebase to Gemini with your instruction
3. Pattern Generation: AI returns glob patterns for files to ignore
4. Filtered Output: Runs repomix again with the ignore patterns, respecting your config

Programmatic Usage

You can also use smart-repomix in your Node.js code:

1import smartRepomix from 'smart-repomix';
2
3const ignorePatterns = await smartRepomix({
4  instruction: 'ignore test files and build artifacts',
5  outputPath: 'output.xml', // optional, uses repomix config if not specified
6  apiKey: 'your-api-key',
7  model: 'gemini-2.5-flash',
8  verbose: true
9});
10
11console.log('Ignored patterns:', ignorePatterns);

Tips

• Be specific in your instructions for better results
• Use --dry-run first to preview what will be ignored
• The gemini-2.5-flash model is fast and works well for most cases
• For very large codebases, consider using gemini-2.5-pro for better analysis
• Smart Repomix respects your existing repomix.config.json settings

License

MIT