name: Prompt Engineering
domain: prompts
default_metric: accuracy
default_metric_direction: higher
suggested_cycle: gree
description: >
  Optimize prompts against an evaluation dataset.
  Gather examples, refine prompt structure, experiment with variations, evolve best approach.
files:
  - prompt.md      # the prompt template to optimize
  - evaluate.ts    # evaluation script
  - dataset.json   # evaluation dataset
  - program.md     # agent instructions