diff --git a/app/harness/runner.py b/app/harness/runner.py index 6e1752d..9d3b222 100644 --- a/app/harness/runner.py +++ b/app/harness/runner.py @@ -553,7 +553,9 @@ class Runner: from app.harness.log import HarnessLog from app.question_gen import load_benchmark - questions = load_benchmark(self._paths.questions_dir) + # CLI --questions 优先于 manifest 中的 questions(infer 不改 manifest) + questions_dir = Path(self._config.store_dir) / "questions" / self._config.questions + questions = load_benchmark(questions_dir) if task_types: allowed = set(task_types) questions = [q for q in questions if q.task_type in allowed]