INNER CODE UNIT · Python

load_evals

wgpsec/AboutSecurity · scripts/bench-skill.py:78

def load_evals(skill_path: Path) -> dict:
    """Load evals/evals.json from skill directory."""
    evals_file = skill_path / "evals" / "evals.json"
    if not evals_file.exists():
        raise FileNotFoundError(f"No evals found: {evals_file}")
    return json.loads(evals_file.read_text(encoding="utf-8"))


def run_with_claude(prompt: str, timeout: int = 120, model: str = None) -> dict:
    """Run a prompt via `claude -p` and capture output + metrics."""
    cmd = ["claude", "-p", prompt, "--output-format", "json"]
    if model:
        cmd.extend(["--model", model])

    env = {k: v for k, v in os.environ.items() if k != "CLAUDECODE"}

    start = time.time()
    try:

View source record →