Multi-Agent Collaboration Prototype

System uses agents. Agents collaborate. Shared canvas holds work. Adapting for code improvement.

Features

• Multi-agent system.
• Shared canvas.
• Phase workflow.
• Structured logging (JSONL).
• Extensible agents.

Code Improvement Plan

Based on "Task-Agnostic Code Improvement Plan". Plan guides phases.

Project Structure

• main.py: Original web design demo.
• main_orchestrator.py: Code improvement workflow. Runs phases.
• agents/agent.py: Agent base class. Defines agent.
• agents/canvas.py: Merges agent work. Updates canvas.
• agents/extractors.py: Gets data from agent output.
• agents/logging_utils.py: Sets up JSONL logs.
• files/f-Code Improvement Task Agnostic Plan.md: The plan document.
• sample_code.py: Example code for agents.

Setup

Need Python. Need Ollama. Install deps: pip install colorama ollama

Usage

Run python main_orchestrator.py. Agents process sample_code.py. Output shows progress. Logs are JSONL.

Extending

Add agents in get_agents_for_phase. Add phases to list.

Notes

Prototype state. Canvas is text. No file writes yet.