{
  "artifact_profile": "l1_standard",
  "description": "Exercises agent-browser main: eval.throws through the agent-browser tool API. It passes when `agent-browser eval throw new Error('intentional-eval-error')` reports the expected command failure. Only these declared checks affect the verdict.",
  "driver": {
    "connect_env": "BROWSER_WS",
    "env": {
      "AB_CHECKS": [
        {
          "kind": "ab_fails",
          "step": 0
        }
      ],
      "AB_STEPS": [
        {
          "ab": [
            "eval",
            "throw new Error('intentional-eval-error')"
          ],
          "ignore_error": true
        }
      ]
    },
    "kind": "node_cdp_probe",
    "script": "runner/scripts/l1_ab_probe.js"
  },
  "features": [
    "tool.agent_browser.eval.exception"
  ],
  "grader": {
    "checks": [
      "script_reported"
    ],
    "kind": "inline_assertions"
  },
  "layer": "L1",
  "scene": {
    "kind": "self_hosted_fixture",
    "url": "/l1/core"
  },
  "subset_id": "l1.agent_browser_tool",
  "tags": [
    "version.v0_1"
  ],
  "task_id": "ab_main_eval_throws",
  "task_version": 1,
  "timeouts": {
    "hard_kill_ms": 120000,
    "task_ms": 90000
  },
  "title": "agent-browser main: eval.throws"
}