| name | evolve |
| description | Start an autonomous evolutionary code optimization run |
Launch an evolution run that optimizes code using Claude Code models as mutation operators.
Usage
/evolve [config.json]
/evolve --init initial.py --eval evaluate.py --generations 50
Requirements
initial.py: Seed program with # EVOLVE-BLOCK-START / # EVOLVE-BLOCK-END markers
evaluate.py: Evaluator script that returns metrics and correctness
What Happens
- Loads the initial program and evaluator
- Runs N generations of evolutionary optimization:
- Selects parent programs from island populations
- Queries Claude models (via UCB1 bandit) to generate mutations
- Applies diff/full/crossover patches
- Evaluates fitness via the user's evaluator script
- Updates the archive of best solutions
- Returns the best-performing solution
The run is autonomous -- it proceeds without interaction until complete or stopped.