{
  "id": "benchmark:cyber-bench",
  "_kind": "Benchmark",
  "_file": "benchmarks/benchmarks/benchmarks-security.yaml",
  "_cluster": "benchmarks",
  "attributes": {
    "displayName": "CyberBench",
    "homepageUrl": "https://cybench.github.io/",
    "kind": "security",
    "targetsKind": "AgentVersion",
    "description": "CyberBench (Princeton et al.) measures language-model and agent\ncapability on real-world cybersecurity capture-the-flag tasks\nspanning crypto, web, reverse engineering, pwn, and forensics.\n"
  },
  "outgoingEdges": [
    {
      "from": "benchmark:cyber-bench",
      "to": "domain:cybersecurity",
      "kind": "applies_to",
      "attributes": {}
    },
    {
      "from": "benchmark:cyber-bench",
      "to": "domain:security",
      "kind": "applies_to",
      "attributes": {}
    }
  ],
  "incomingEdges": [
    {
      "from": "scope-boundary:cyber-bench.scope",
      "to": "benchmark:cyber-bench",
      "kind": "bounds_subject"
    }
  ]
}