skills/ag2/_scores.json

{
  "version": "1.0.0",
  "skillHash": "sha256:288e992d8abcf0a237c91d0d84330db7907aa50f65aebfa0680404a5ee283428",
  "scoredAt": "2026-05-13T12:21:09.800Z",
  "backend": "ollama",
  "model": "gpt-oss:20b",
  "quality": {
    "score": 87,
    "dimensions": {
      "clarity": "PASS",
      "completeness": "WEAK",
      "conciseness": "PASS",
      "actionability": "PASS",
      "crossPlatform": "WEAK",
      "examples": "PASS"
    },
    "issues": [
      {
        "severity": "MEDIUM",
        "category": "completeness",
        "detail": "The skill examples focus on happy path scenarios and do not address error handling or edge cases."
      },
      {
        "severity": "MEDIUM",
        "category": "crossPlatform",
        "detail": "The skill relies on Unix-specific tools like grep and Docker, limiting compatibility on non-Unix platforms."
      }
    ]
  },
  "security": {
    "verdict": "MALICIOUS",
    "issues": [
      {
        "severity": "HIGH",
        "category": "subprocess_injection",
        "detail": "search_codebase uses unsanitized user input in subprocess.run, enabling command injection."
      },
      {
        "severity": "HIGH",
        "category": "subprocess_injection",
        "detail": "run_tests uses unsanitized test_path in subprocess.run, enabling command injection."
      }
    ]
  },
  "impact": {
    "multiplier": 10,
    "baselineAvg": 5,
    "treatmentAvg": 100,
    "scenarios": [
      {
        "name": "two-agent-code-execution",
        "baseline": 0,
        "treatment": 100,
        "rationale": "Response B fully satisfies the rubric, while Response A does not include the required agent definitions or execution evidence."
      },
      {
        "name": "group-chat-notification-service",
        "baseline": 0,
        "treatment": 100,
        "rationale": "Response A fully satisfies the rubric, while Response B does not meet any of the required elements."
      },
      {
        "name": "tool-use-code-search-tests",
        "baseline": 15,
        "treatment": 100,
        "rationale": "Response A fully meets the rubric with proper function signatures, register_function calls, caller/executor, and tester usage; Response B lacks required registration, signatures, and workflow details."
      }
    ]
  }
}