sunghunkwag/ast-grammar-induction-prototype ? reverse-engineered prompt

Reverse engineered prompt

Build me an experimental Python app that can improve its own code by mutating Python ASTs and learning which changes seem to work better over time.

I want it to scan the ARC_GYM task files, run a limited evolution loop, and try to solve more tasks by adjusting its mutation strategy from past results. It should have a safe mode with loop limits, rollback when a change breaks things, and a self check to make sure the sandbox is active before it runs.

Please include a way to run one round of evolution, a way to test and verify the full setup, and some simple debug and monitoring scripts so I can see what it is doing. If anything important is unclear, look up current docs online if you need to. Keep it experimental and cautious, since this kind of self modifying code can go wrong fast.

Are you gonna build this?

make sure you review the code using coderabbit

Try freeSponsored — opens CodeRabbit in a new tab