skills/jinko-vpop/evals/evals.json

{
  "skill_name": "jinko-vpop",
  "evals": [
    {
      "id": 1,
      "prompt": "I have a CSV with patientIndex and model descriptor columns. Help me create a Jinkō vpop safely with the SDK.",
      "expected_output": "Uses create_vpop_from_csv or create_vpop_from_dataframe, checks patientIndex and descriptor headers, and requires --apply before upload."
    },
    {
      "id": 2,
      "prompt": "Create a vpop design from marginal distributions for Dose and k_elim and generate 10 patients from it.",
      "expected_output": "Uses create_vpop_design_from_design with unique marginal IDs, checks distrib.json, prints design diagnostics, blocks on diagnostic errors, and only then uses design.generate_vpop()."
    },
    {
      "id": 3,
      "prompt": "Can I edit a generated vpop? I need to change the marginal distribution for k_elim.",
      "expected_output": "Explains that generated/uploaded Vpop items are not edited in place; edit the Vpop design and regenerate a new Vpop."
    },
    {
      "id": 4,
      "prompt": "Apply three marginal edits to vd-123. The second update fails after the first succeeds; tell me exactly what happened and do not claim the design is valid without checking it.",
      "expected_output": "Uses sequential descriptor mutators, reports the ID applied before the failure, and returns failure. On a fully applied sequence it always prints fresh post-edit diagnostics and only reports success when there are no errors."
    },
    {
      "id": 5,
      "prompt": "Create a Vpop design from a marginal list that contains Dose twice with different distributions.",
      "expected_output": "Rejects the duplicate Dose marginal ID before creating any project item rather than silently keeping the last distribution."
    }
  ]
}