Sec-Sql-Injection-Detector

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 a detector that only searches for OR 1=1 and misses stacked queries, comments, and parameterized-safe cases.

{
  "metrics": {
    "correct": {
      "checks": [
        {
          "contains": [
            "SQL injection"
          ]
        },
        {
          "contains": [
            "parameterized"
          ]
        },
        {
          "contains": [
            "stacked"
          ]
        },
        {
          "contains": [
            "comment"
          ]
        },
        {
          "contains": [
            "false positive"
          ]
        }
      ]
    },
    "hidden": {
      "checks": [
        {
          "contains": [
            "prepared statement"
          ]
        },
        {
          "contains": [
            "AST"
          ]
        },
        {
          "contains": [
            "escaping"
          ]
        },
        {
          "contains": [
            "taint"
          ]
        },
        {
          "contains": [
            "allowlist"
          ]
        }
      ]
    }
  }
}