name: Code Optimization
domain: code
default_metric: score
default_metric_direction: lower
suggested_cycle: pflk
description: >
  Optimize a single code file for a measurable metric.
  Inspired by Karpathy's autoresearch — modify code, run benchmark, keep or discard.
files:
  - experiment.py  # or .ts, .go, etc. — the file the agent modifies
  - evaluate.sh    # evaluation harness — runs benchmark, outputs metrics
  - program.md     # agent instructions