Sec-Csp-Nonce-Validator

The model receives the prompt (and optional system message). The run uses scorer rubric_json_metrics with the JSON configuration below. Pass/fail and partial credit are determined entirely by that scorer against the model output; no human grading.

Return JSON only with keys verdict, risk, mitigation. Review CSP generation that reuses the same nonce for every request and allows unsafe-inline.

{
  "metrics": {
    "correct": {
      "checks": [
        {
          "contains": [
            "CSP"
          ]
        },
        {
          "contains": [
            "nonce"
          ]
        },
        {
          "contains": [
            "unsafe-inline"
          ]
        },
        {
          "contains": [
            "per request"
          ]
        },
        {
          "contains": [
            "script-src"
          ]
        }
      ]
    },
    "hidden": {
      "checks": [
        {
          "contains": [
            "random"
          ]
        },
        {
          "contains": [
            "base64"
          ]
        },
        {
          "contains": [
            "strict-dynamic"
          ]
        },
        {
          "contains": [
            "do not reuse"
          ]
        },
        {
          "contains": [
            "header"
          ]
        }
      ]
    }
  }
}