"""Write synthetic data: data/<name>/{secrets,injection,toolcall}.jsonl.

  python scripts/build_eval.py                                  # eval set (eval templates)
  python scripts/build_eval.py --split train --n 4000 --seed 1 --out data/train
  python scripts/build_eval.py --split train --n 400 --seed 2 --out data/dev
"""
import argparse
import json
import logging
import random
from pathlib import Path

from vela_guard import gen_injection, gen_secrets, gen_toolcall

GENERATORS = {"secrets": gen_secrets, "injection": gen_injection, "toolcall": gen_toolcall}
EVAL_SEED = 20261012

log = logging.getLogger("build_eval")


def main() -> None:
    parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
    parser.add_argument("--out", type=Path, default=Path("data/eval"))
    parser.add_argument("--n", type=int, default=300, help="examples per task")
    parser.add_argument("--seed", type=int, default=EVAL_SEED)
    parser.add_argument("--split", choices=("eval", "train"), default="eval")
    args = parser.parse_args()

    args.out.mkdir(parents=True, exist_ok=True)
    for offset, (task, module) in enumerate(GENERATORS.items()):
        rows = module.generate(random.Random(args.seed + offset), args.n, split=args.split)
        path = args.out / f"{task}.jsonl"
        path.write_text("".join(json.dumps(r, ensure_ascii=False) + "\n" for r in rows))
        log.info("%s: %d rows (%d hard) -> %s", task, len(rows), sum(r["hard"] for r in rows), path)


if __name__ == "__main__":
    logging.basicConfig(level=logging.INFO, format="%(message)s")
    main()