skills/agent-sandbox/_scores.json

{
  "version": "1.0.0",
  "skillHash": "sha256:93e599803bba2306a06277d52a0632f080d1efc2c7c5c2e8a1a88e5d9b2358a5",
  "scoredAt": "2026-05-13T12:15:37.666Z",
  "backend": "ollama",
  "model": "gpt-oss:20b",
  "quality": {
    "score": 93,
    "dimensions": {
      "clarity": "PASS",
      "completeness": "PASS",
      "conciseness": "PASS",
      "actionability": "PASS",
      "crossPlatform": "WEAK",
      "examples": "PASS"
    },
    "issues": [
      {
        "severity": "MEDIUM",
        "category": "crossPlatform",
        "detail": "The skill requires Docker and Node.js, limiting cross-platform compatibility."
      }
    ]
  },
  "security": {
    "verdict": "SAFE",
    "issues": []
  },
  "impact": {
    "multiplier": 4.09,
    "baselineAvg": 22,
    "treatmentAvg": 90,
    "scenarios": [
      {
        "name": "sandbox-code-modification",
        "baseline": 10,
        "treatment": 100,
        "rationale": "Response B fully satisfies the rubric, while Response A lacks the required AgentSandbox instantiation and related actions."
      },
      {
        "name": "docker-isolated-execution",
        "baseline": 50,
        "treatment": 95,
        "rationale": "Response B fully meets the rubric by using DockerSandbox with the required options, starting, executing, and killing the container, while Response A omits DockerSandbox usage, readOnly, and explicit container cleanup details."
      },
      {
        "name": "kill-switch-audit",
        "baseline": 5,
        "treatment": 75,
        "rationale": "Response B meets most rubric points, while Response A does not use AgentSandbox or required actions."
      }
    ]
  }
}