laurentvv/graph-orchestrator-smolagents ? reverse-engineered prompt

Build me a Python project that acts like a small team of AI agents instead of one chatty agent doing everything in order.

I want a central coordinator that breaks a task into smaller pieces, sends them out to multiple worker agents in parallel, checks their results, then has a judge agent compare the answers and decide what to keep. The workers should be able to write code, run tests, and fix mistakes, while the coordinator keeps the flow deterministic in Python rather than letting the model make up routing on its own.

Please include safe file and shell actions, basic validation of generated outputs, and a simple way to store shared progress so later runs can reuse what was learned. If it helps, use current docs online for any external libraries. I’d also like a small demo or server entry point so I can try the whole workflow end to end and see the agents collaborate on a coding task.