feat(code-template): Add Code Template System and Tool Calling Infrastructure

[thecodacus]

Add Code Template System and Tool Calling Infrastructure

Overview

This PR introduces two major architectural enhancements:

1. A code template system for project scaffolding
2. A foundational infrastructure for AI tool calling with an agent-based architecture

These changes enable more sophisticated project initialization and lay the groundwork for AI-driven tooling capabilities.

Key Changes

1. Template System Implementation

• Added project template import functionality
• Implemented GitHub-based template fetching
• Added template configuration and management
• Introduced template ignore patterns for selective imports
• Added support for template-specific prompts

2. Tool Calling Infrastructure

• Implemented base agent architecture for tool execution
• Added tool calling parser and execution pipeline
• Integrated tool configuration management
• Added support for agent-based routing and coordination
• Implemented template variables system for dynamic prompting

3. Agent System

• Added AgentPromptGenerator for structured prompt generation
• Implemented AgentOutputParser for processing tool calls
• Added coordinator agent for query routing
• Implemented project scaffolding agent
• Added support for agent registration and management

4. UI Enhancements

• Added tool configuration toggle in chat interface
• Implemented tool execution status display
• Added template selection interface
• Enhanced message parsing for tool responses
• Added support for tool calling visualization

Technical Implementation

Agent Architecture

• Base agent class with configurable tools and parameters
• XML-based tool call protocol for structured communication
• Support for template variables and dynamic prompt generation
• Coordinator agent for intelligent query routing

Template System

• GitHub API integration for template fetching
• Support for .bolt configuration directory
• Template ignore patterns for selective imports
• Custom prompt support for templates

Tool Execution Pipeline

1. Tool call parsing and validation
2. Agent selection and routing
3. Parameter validation
4. Tool execution
5. Response formatting and streaming

Testing

• Tested template import functionality
• Verified tool calling pipeline
• Tested agent routing system
• Verified template ignore patterns
• Tested tool configuration persistence

Migration Impact

• No breaking changes for existing chat functionality
• Tool calling requires explicit opt-in via toggle
• Template system is additive to existing functionality
• Existing message parsing remains unchanged

Future Enhancements

• Add support for more complex tool chains
• Enhance template management system
• Add more specialized agents
• Improve tool execution visualization
• Add support for concurrent tool execution

Preview

https://github.com/user-attachments/assets/ba3c180c-625b-4d5d-87d1-2580ed6ff5a0

[thecodacus]

@wonderwhy-er , sorry but I tried to make the changes as little as possible 😅

[thecodacus]

Also this approach is text based, to it can be used by any llm, as long as it follows the structured output and does not get confused. that require prompt tuning which can be done in separate task

we can now add agents for github project import or commit, using this framework

[wonderwhy-er]

Ok, will need to take a deeper look in to how it works

[thecodacus]

this is an example of git import using the tooling framework

https://github.com/user-attachments/assets/6ab045c6-9614-4c41-9410-7263ff10affe

[wonderwhy-er]

Ahh its still big, will take time to fully review...

[champ2050]

@thecodacus Awesome work! Thank you for your contribution. Have you had a chance to check out Microsoft's Magnetic-One? It includes an orchestrator that passes commands to specific agents, each designed to handle particular tasks. If you could take a look at it, that might make a fantastic addition to this project. Magnetic-One was recently released, so not many people have seen it yet, but it works really well. Thanks again for your awesome work, I really appreciate it!

[thecodacus]

no I didn't get the chance to see that yet. will definitely check that