scripts/sample_actions_web.json

{
  "domain": "web",
  "scenario_id": "SearchGlitch",
  "mission_id": "searchglitch_web_sample",
  "reset_payload": {
    "seed": 42,
    "scenario_id": "SearchGlitch",
    "mission_id": "searchglitch_web_sample",
    "options": {
      "env_id": "omnibench_aegis_env:web.searchglitch",
      "max_steps": 5,
      "target_score": 1,
      "domain": "web"
    }
  },
  "action_examples": {
    "canonical": [
      {
        "name": "inspect_query",
        "args": {}
      },
      {
        "name": "fetch_results_page",
        "args": {
          "page": 1
        }
      },
      {
        "name": "deduplicate_results",
        "args": {}
      },
      {
        "name": "verify_source_consistency",
        "args": {}
      },
      {
        "name": "submit_result",
        "args": {
          "include_citations": true
        }
      }
    ],
    "shorthand": [
      {
        "action": "inspect_query"
      },
      {
        "action": "fetch_results_page",
        "page": 1
      },
      {
        "action": "deduplicate_results"
      },
      {
        "action": "verify_source_consistency"
      },
      {
        "action": "submit_result",
        "include_citations": true
      }
    ]
  },
  "expected_good_trajectory": [
    "inspect_query",
    "fetch_results_page",
    "deduplicate_results",
    "verify_source_consistency",
    "submit_result"
  ],
  "expected_bad_trajectory": [
    "fetch_results_page",
    "submit_result"
  ],
  "expected_flow": [
    "health",
    "reset",
    "step",
    "state"
  ],
  "notes": [
    "Aligned fixture for the SearchGlitch web domain.",
    "The good trajectory checks the query, fetches results, removes duplicates, verifies consistency, and submits a sourced answer.",
    "Submitting before deduplication or verification should remain a weak trajectory."
  ]
}