#!/usr/bin/env python3
from __future__ import annotations

import argparse
import json
import subprocess
from pathlib import Path

HERE = Path(__file__).resolve().parent
WORKSPACE = HERE.parents[2]


def main() -> None:
    parser = argparse.ArgumentParser()
    parser.add_argument("case_id")
    parser.add_argument("variant", choices=("baseline", "experience"))
    args = parser.parse_args()
    manifest = json.loads((HERE / "manifest.json").read_text())
    case = next(item for item in manifest["cases"] if item["id"] == args.case_id)
    output = Path(case[f"{args.variant}_step"])
    output.parent.mkdir(parents=True, exist_ok=True)
    if output.is_file():
        print(json.dumps({"status": "skipped", "output": str(output)}))
        return

    description = Path(case["description"]).read_text()
    library = (
        HERE / "empty-library.json"
        if args.variant == "baseline"
        else HERE / "train-library.json"
    )
    prompt = f"""
This is a controlled CAD reconstruction benchmark.
Use $cad-router and $cad to reconstruct the single mechanical part shown in
the attached multi-view image and described below. Generate a new parametric
STEP from the image and description only.

Hard isolation rules:
- Do not search for, inspect, import, copy, or reference any teacher STEP,
  SLDPRT, parser/input, parser/output, case JSON, or SLDPRT-NET dataset file.
- The attached image and the description in this prompt are the only
  part-specific evidence.
- Run cad-router with the explicit generalized experience library
  {library.resolve()}; do not use any other experience library.
- Use the router-selected backend and actually apply returned generalized
  methods when present.
- Do not preserve an exact STEP base: this must be a blind reconstruction.

Write the primary output exactly to:
{output.resolve()}
Keep the editable generator and model-spec.json in the same directory.
Also write the router result to route.json and record in cad-task.json the
experience methods returned and the subset actually applied. Record an empty
list when nothing matches; never claim an experience method was used unless
the router returned it.
Validate that the STEP opens and contains positive-volume geometry. For this
batch benchmark, do not open CAD Viewer or create browser tabs.

Description:
{description}
"""
    command = [
        "codex",
        "exec",
        "--ephemeral",
        "--json",
        "--image",
        case["image"],
        "--cd",
        str(WORKSPACE),
        "--sandbox",
        "danger-full-access",
        "--output-last-message",
        case[f"{args.variant}_message"],
        prompt,
    ]
    with Path(case[f"{args.variant}_log"]).open("w") as stream:
        completed = subprocess.run(command, stdout=stream, stderr=subprocess.STDOUT)
    print(json.dumps({
        "status": "ok" if completed.returncode == 0 and output.is_file() else "failed",
        "returncode": completed.returncode,
        "output": str(output),
    }))
    raise SystemExit(0 if completed.returncode == 0 and output.is_file() else 1)


if __name__ == "__main__":
    main()