Autonomous Web Agent with Dynamic Prompt Management

This system allows you to create autonomous web agents using Puppeteer and large language models. It features a dynamic prompt management system that can generate detailed web automation instructions from simple natural language commands.

Features

• Dynamic Prompt Generation: Turn simple commands into detailed step-by-step instructions
• Task Classification: Automatically categorize user requests and extract parameters
• Template Management: Organize and manage prompt templates for different tasks
• Prompt Repository: Store and retrieve generated prompts
• Multiple Task Types: Support for shopping, search, media, and more
• Browserbase Integration: Option to use Browserbase for cloud-based browser automation

Installation

1. Clone the repository
2. Install dependencies:

   npm install
   

3. Copy the .env.example to .env and fill in your API keys:

   cp .env.example .env
   

4. Build the TypeScript code:

   npm run build
   

Usage

Local Browser Automation

Start the agent with a local Chrome instance:

npx ts-node client.ts

Browserbase Cloud Automation

Start the agent with Browserbase cloud browser:

npm run browserbase

or

npx ts-node client.ts --browserbase

The interactive CLI will guide you through:

1. Generating detailed instructions from simple commands
2. Viewing available templates
3. Managing previously generated prompts

Browserbase Integration

This project supports Browserbase for cloud-based browser automation. Browserbase provides:

• Cloud-based Chromium instances
• Anti-detection fingerprinting
• Proxy support
• Captcha solving

To use Browserbase:

1. Set the following environment variables in your .env file:

   BROWSERBASE_API_KEY=your_api_key
   BROWSERBASE_PROJECT_ID=your_project_id
   USE_BROWSERBASE=true
   

2. Or run with the --browserbase flag:

   npx ts-node client.ts --browserbase
   

You can view your Browserbase sessions at: https://browserbase.com/sessions

Directory Structure

• /src - TypeScript source code
  • /prompts - Prompt generation and templating
  • /storage - File storage and retrieval
  • /utils - Utility functions including LLM client
  • /agent - Agent execution logic (future)
• /templates - Prompt templates
  • /system - System prompts
  • /tasks - Task-specific templates
  • /generated - Generated detailed instructions
• /config - Configuration files

Creating Custom Templates

Create your own templates in the /templates directory. Use the {{variable}} syntax for dynamic content. For example:

I need to search for information about {{query}} and collect the results.

Please provide detailed step-by-step instructions for...

Then define the task in /config/tasks.json.

Environment Variables

• ANTHROPIC_API_KEY - API key for the Claude AI model
• BROWSERBASE_API_KEY - API key for Browserbase
• BROWSERBASE_PROJECT_ID - Project ID for Browserbase
• USE_BROWSERBASE - Set to "true" to use Browserbase instead of local Chrome
• AGENT_MODE - Set to "interactive" or "execute"
• AGENT_INSTRUCTIONS_PATH - Path to instructions file for agent mode

License

MIT