llmove

llmove is a command-line tool that processes XML specification files and sends them to Claude (Anthropic's LLM) for AI-powered code generation.

What it does

llmove scans a directory of .xml specification files, extracts prompts and system instructions, and sends them to Claude to generate code or documentation. The tool:

• Recursively processes XML files with support for includes and plugins
• Tracks which files have been processed to avoid duplicates
• Supports dry-run mode to preview what will be sent to the API
• Can re-apply the last generated output
• Automatically handles file naming conflicts with timestamps

Installation

npm install -g llmove

Quick Start

1. Configure llmove

llmove configure

This will prompt you for:

• Your Anthropic API key
• The model to use (default: claude-opus-4-20250514)
• The specs folder path (default: specs)

Configuration is stored in ~/.llmove/config.json.

1.1 Initialize llmove in your project

llmove init

2. Write your first spec file

Create a file specs/my-feature.xml:

<prompt>
  Generate a Node.js function that validates email addresses using regex.
  The function should:
  - Return true for valid emails
  - Return false for invalid emails
  - Handle edge cases like multiple dots, special characters
</prompt>

3. Run llmove

# Preview what will be sent (dry run)
llmove --dryRun

# Generate code from your specs
llmove

Writing XML Spec Files

Basic Structure

Spec files use two main tags:

• <system> - Sets the system prompt/context for the LLM
• <prompt> - Contains the user prompt/request

Example Spec File

<!-- specs/api/user-service.xml -->
<system>
  You are an expert Node.js developer. Generate clean, well-documented code
  following REST API best practices.
</system>

<prompt>
  Create a user service with the following endpoints:
  - GET /users - List all users with pagination
  - GET /users/:id - Get user by ID
  - POST /users - Create new user
  - PUT /users/:id - Update user
  - DELETE /users/:id - Delete user
  
  Use Express.js and include input validation.
</prompt>

Advanced Features

Including Files

You can include content from other files:

<prompt>
  Update this function to handle async operations:
  
  <llmove:include path="./src/validator.js" lines="10:25" />
</prompt>

Source Code References

Use the source tag to include entire files:

<prompt>
  Refactor this code to use TypeScript:
  
  <llmove:source path="./legacy/user.js" />
</prompt>

Root Configuration

The specs/root.xml file can define global settings:

<system version="0.2.0">
  <!-- Global system context applied to all prompts -->
  You are a senior developer working on a Node.js microservices project.
  Follow clean code principles and include comprehensive error handling.
</system>

Chat

With

llmove chat

you can chat with your LLM having the XML root and conventions file as context.

Configuration

Environment Variables

You can override configuration using environment variables:

• LLMOVE_API_KEY - Your Anthropic API key
• LLMOVE_API_URL - Custom API URL (default: https://api.anthropic.com)
• LLMOVE_API_MODEL - Model to use

File Structure

project/
├── specs/              # Your XML specification files
│   ├── root.xml       # Global configuration
│   └── features/      # Organize specs in subdirectories
├── .llmove/           # Cache and metadata
│   ├── userPrompts.txt        # Tracks processed files
│   └── last-llmove-output.json # Last generation result

Tips

1. Organize your specs: Use subdirectories to group related specifications
2. Use conventions.xml: Place common patterns in conventions.xml files
3. Incremental processing: llmove tracks which files have been processed
4. Version control: Commit your specs to track changes over time
5. Dry run first: Always use --dryRun to preview before making API calls

How to Contribute

Architecture

llmove is designed to be minimal and extensible:

• Main CLI (llmove.cjs): Entry point and command handling
• Libraries (lib/): Core functionality split into focused modules
• Plugins (lib/plugins/): Transform and filter XML content

Plugin System

llmove supports two types of plugins:

Filter Plugins

Filter plugins process the list of files before parsing:

// lib/plugins/myFilter.js
module.exports = async function filterExample(config, files, metadata) {
  // Transform or filter the files array
  return files.filter(file => !file.path.includes('draft'));
};

Parse Plugins

Parse plugins transform individual file content:

// lib/plugins/myParser.js
module.exports = async function parseExample(config, file, metadata) {
  // Transform file content
  file.content = file.content.replace(/TODO/g, 'TASK');
  
  // Can return just the file or include new files to process
  return {
    file: file,
    newFiles: [] // Optional: additional files to process
  };
};

Adding Plugins

1. Create your plugin in lib/plugins/
2. Add it to the appropriate array in lib/plugins.js:

const filterPlugins = [
  require('./plugins/filterAddRootConventions'),
  require('./plugins/myFilter') // Add your filter plugin
];

const parsePlugins = [
  require('./plugins/source'),
  require('./plugins/includeFiles'),
  require('./plugins/myParser') // Add your parse plugin
];

Development Guidelines

1. Keep it tiny: No unnecessary dependencies
2. Small functions: Each function should do one thing well
3. Clear separation: Logic should be separated into appropriate modules
4. Error handling: Fail gracefully with helpful error messages
5. Cross-platform: Use Node.js built-ins for file system operations
6. Autocode: Use llmove itself to create plugins =)

Testing

Create test XML files in specs/test/ and run:

llmove --dryRun

Submitting Changes

1. Fork the repository
2. Create a feature branch
3. Write clear commit messages
4. Ensure your code follows the existing style
5. Submit a pull request with a description of your changes

License

CC0-1.0

Author

Luca Rainone