Multi Agent System Benchmarking

[djriffle]

This pull request introduces a comprehensive framework for managing and interacting with a multi-agent system in a benchmarking environment. Key changes include the implementation of core classes for defining agents and their interactions, tools for creating and configuring agent systems, and utilities for handling input/output operations with enhanced user interactivity.

Multi-Agent System Framework

• benchmarking/agents/AgentSystem.py: Added core classes Agent, Command, and AgentSystem to model agents, their commands, and their interactions. Includes methods for loading configurations from JSON, retrieving agents, and generating prompts for large language models (LLMs).

Agent System Configuration and Creation

• benchmarking/agents/create_agent_system.py: Introduced an interactive script for defining agents, connecting them, and saving configurations. Includes user-friendly prompts and error handling for creating agent systems.
• benchmarking/agents/system_blueprint.json: Added a sample JSON blueprint for a multi-agent system, including a master_agent and two specialist agents (coder_agent and research_agent).

Input/Output Enhancements

• benchmarking/core/io_helpers.py: Added utilities for rich-text terminal interactions, dataset selection, and resource collection. Includes a function to extract Python code from text and format execution responses with detailed output.

Miscellaneous

• benchmarking/.gitignore: Updated to ignore agent_systems/ directory, ensuring generated agent configurations are not tracked.