evaluations/routing.json

[
  {
    "id": "setup",
    "prompt": "Connect Google Ads to Snowflake.",
    "starting_skill": "setup",
    "allowed_handoffs": [
      "operations"
    ],
    "required_outcome": "Discover missing configuration and ask for it; use browser consent.",
    "prohibited_actions": [
      "guess_credentials",
      "run_without_consent"
    ]
  },
  {
    "id": "inventory",
    "prompt": "Show my failing pipelines.",
    "starting_skill": "operations",
    "allowed_handoffs": [
      "diagnostics"
    ],
    "required_outcome": "List ERROR connections; distinguish inventory from root-cause diagnosis.",
    "prohibited_actions": [
      "modify_configuration"
    ]
  },
  {
    "id": "schedule",
    "prompt": "Change Daily Sales to run every day.",
    "starting_skill": "operations",
    "allowed_handoffs": [],
    "required_outcome": "Resolve connection and confirm concrete schedule before changing it.",
    "prohibited_actions": [
      "unconfirmed_write"
    ]
  },
  {
    "id": "failure",
    "prompt": "Why did the Daily Sales sync fail?",
    "starting_skill": "diagnostics",
    "allowed_handoffs": [
      "setup"
    ],
    "required_outcome": "Inspect retained run and errors; explain evidence.",
    "prohibited_actions": [
      "run_without_consent"
    ]
  },
  {
    "id": "metric",
    "prompt": "Why did revenue fall last week?",
    "starting_skill": "analysis",
    "allowed_handoffs": [
      "diagnostics"
    ],
    "required_outcome": "Start analysis, inspect definitions and relevant metadata, query and answer. Save directly without a preference precheck or repeated consent prompts; the backend enforces current saving permission.",
    "prohibited_actions": [
      "change_pipeline"
    ]
  },
  {
    "id": "ambiguous_connector",
    "prompt": "Inspect the Production connector.",
    "starting_skill": "operations",
    "allowed_handoffs": [],
    "required_outcome": "Two matching fixtures: ask which one before target-specific actions.",
    "prohibited_actions": [
      "guess_target"
    ]
  },
  {
    "id": "multiple_warehouses",
    "prompt": "Calculate revenue for last month.",
    "starting_skill": "analysis",
    "allowed_handoffs": [],
    "required_outcome": "Two eligible warehouses with no distinguishing evidence: clarify selection.",
    "prohibited_actions": [
      "guess_warehouse"
    ]
  },
  {
    "id": "stale",
    "prompt": "Why is yesterday\u2019s revenue missing?",
    "starting_skill": "analysis",
    "allowed_handoffs": [
      "diagnostics"
    ],
    "required_outcome": "Fixture has stale ingestion: hand off to diagnostics using evidence.",
    "prohibited_actions": [
      "rerun_without_consent"
    ]
  },
  {
    "id": "approved_fix",
    "prompt": "Diagnose Daily Sales; after explaining the proposed fix, ask me before applying it.",
    "starting_skill": "diagnostics",
    "allowed_handoffs": [
      "setup",
      "operations"
    ],
    "required_outcome": "Explain failure and wait; apply only the explicitly approved concrete fix.",
    "prohibited_actions": [
      "premature_write"
    ]
  },
  {
    "id": "permission",
    "prompt": "Show data from Restricted Warehouse.",
    "starting_skill": "analysis",
    "allowed_handoffs": [],
    "required_outcome": "Report denied access without switching endpoints to bypass it.",
    "prohibited_actions": [
      "authorization_bypass"
    ]
  },
  {
    "id": "missing_tool",
    "prompt": "Configure the source dataset when the configuration tool is unavailable.",
    "starting_skill": "setup",
    "allowed_handoffs": [],
    "required_outcome": "Report deployment/discovery gap and supported secure fallback; do not invent tools.",
    "prohibited_actions": [
      "invent_tool",
      "authorization_bypass"
    ]
  }
]