[
  {
    "id": "finetune-vae-from-datalist",
    "question": "Fine-tune the NV-Generate VAE using CT and MRI images listed in /data/ctmr/vae_datalist.json.",
    "expected_skill": "nv-generate-vae-finetune",
    "ground_truth": "The agent uses scripts/run_vae_finetune.py with --data-base-dir, --output-dir, and the datalist path.",
    "expected_behavior": [
      "the command uses skills/nv-generate-vae-finetune/scripts/run_vae_finetune.py",
      "the command includes --data-base-dir",
      "the command includes an explicit --output-dir",
      "the agent states the output checkpoint is experimental and not clinically validated"
    ]
  },
  {
    "id": "preflight-before-gpu-run",
    "question": "Check whether my VAE finetune datalist is runnable before launching the GPU job.",
    "expected_skill": "nv-generate-vae-finetune",
    "ground_truth": "The agent runs the wrapper with --preflight and does not start training.",
    "expected_behavior": [
      "the command includes --preflight",
      "the command does not execute train_vae_tutorial.ipynb",
      "the agent explains that real training still requires NIfTI images, CUDA, and model weights"
    ]
  }
]