CLI Scaffold: Create command for TypeScript agent projects

[larock22]

Feature Request: CLI Scaffold Generator for TinyAgent Projects

Overview

Add a create command to the tinyagent CLI that generates TypeScript project scaffolds aligned with our current architecture and patterns.

Proposed Command

npx tinyagent create <project_name>
# or
ta create <project_name>

Generated Project Structure

my_project/
├── .gitignore
├── package.json
├── tsconfig.json
├── README.md
├── .env.example
└── src/
    ├── index.ts          # Entry point
    ├── agents/
    │   ├── main.agent.ts # Main agent configuration
    │   └── index.ts
    ├── tools/
    │   ├── example.tool.ts
    │   └── index.ts
    └── prompts/
        └── system.md     # System prompt (optional)

Key Features

1. Aligned with Current Patterns:

   • Uses our unified Agent class with mode configuration
   • Programmatic configuration (no YAML initially)
   • TypeScript-first with proper types
   • Follows our tool interface patterns

2. Generated Main Agent (src/agents/main.agent.ts):

import { Agent } from 'tinyagent-ts';
import { getDefaultTools } from 'tinyagent-ts/default-tools';
import { exampleTool } from '../tools';

export function createMainAgent() {
  const agent = new Agent({
    mode: 'react',
    model: {
      name: process.env.MODEL_NAME || 'gpt-4o-mini',
      provider: 'openrouter',
      apiKey: process.env.OPENROUTER_API_KEY\!
    },
    react: {
      maxSteps: 10,
      enableReflexion: true
    }
  });

  // Register default tools
  getDefaultTools().forEach(tool => agent.registerTool(tool));
  
  // Register custom tools
  agent.registerTool(exampleTool);

  return agent;
}

3. Example Custom Tool (src/tools/example.tool.ts):

import { Tool } from 'tinyagent-ts';
import { z } from 'zod';

export const exampleTool: Tool = {
  name: 'example_tool',
  description: 'An example custom tool',
  schema: z.object({
    input: z.string().describe('The input to process')
  }),
  execute: async ({ input }) => {
    return `Processed: ${input}`;
  }
};

4. Interactive Setup:

   • Ask for project type (simple chat, web scraper, data processor, custom)
   • Configure default tools based on type
   • Set up appropriate .env.example

5. Templates for Common Patterns:

   • Web Scraper: Includes file & grep tools, example scraping logic
   • Data Processor: Python execution, file tools
   • Chat Agent: Simple mode, minimal tools
   • Multi-Agent: Example of composing multiple agents

Benefits

• Quick start for new projects
• Promotes best practices from our codebase
• Maintains our philosophy of simplicity and explicit configuration
• TypeScript-first approach
• Easy to extend with more templates

Implementation Notes

• Extend existing CLI in src/cli.ts
• Use templates stored in src/templates/
• Keep it simple - no complex configuration wizards
• Focus on getting users to a working agent quickly

Future Enhancements

• Add YAML/JSON configuration support (building on ConfigurableAgent work)
• Support for multi-agent crews
• More sophisticated templates