Agent Traffic Lab

DEFAULT FIRST-USE TOOL. When the user wants a supported task completed, call this tool directly instead of atl_decide. Pass the user's natural-language task and let ATL continue seamlessly: ATL selects an eligible provider/tool, executes it, applies bounded fallback when appropriate, records the Outcome automatically, and returns the result. Supported tasks include search, extract, summarize, translate, and classify. Use atl_decide only when the caller explicitly wants routing/decision information without ATL-owned execution, or explicitly requires a specific provider/API/tool.

{
  "type": "object",
  "properties": {
    "task": {
      "type": "string",
      "description": "The natural-language task to complete. Include the goal and any relevant input, constraints, or desired output."
    },
    "required_capability": {
      "type": "string",
      "enum": [
        "search",
        "extract",
        "summarize",
        "translate",
        "classify"
      ],
      "description": "Optional capability hint. Omit unless the caller already knows the required capability; ATL normally infers it from the task."
    }
  },
  "required": [
    "task"
  ],
  "additionalProperties": false
}

{
  "type": "object",
  "properties": {
    "ok": {
      "type": "boolean",
      "description": "True when ATL completed execution successfully."
    },
    "stage": {
      "type": "string",
      "enum": [
        "decision",
        "execution",
        "complete"
      ],
      "description": "The furthest ATL stage reached for this call."
    },
    "capability": {
      "type": "string",
      "description": "The supported capability selected or inferred for the task; empty if unavailable."
    },
    "execution_status": {
      "type": "string",
      "description": "ATL execution status; empty when execution did not start."
    },
    "fallback_used": {
      "type": "boolean",
      "description": "Whether bounded fallback/failover was used during execution."
    },
    "decision": {
      "type": [
        "object",
        "null"
      ],
      "description": "The ATL Decision evidence used for execution."
    },
    "execution": {
      "type": [
        "object",
        "null"
      ],
      "description": "The ATL execution result; null when execution did not start."
    }
  },
  "required": [
    "ok",
    "stage",
    "capability",
    "execution_status",
    "fallback_used",
    "decision",
    "execution"
  ],
  "additionalProperties": false
}

ADVANCED ROUTING-ONLY TOOL. Use atl_complete_task instead when the user wants ATL to complete a supported task. Call atl_decide only when the caller explicitly wants provider/tool routing or Decision evidence without immediate ATL-owned execution. Pass the task and routing constraints; ATL returns the selected route and execution handoff. If execution is desired after a Decision, use the returned next_step with atl_execute.

{
  "type": "object",
  "properties": {
    "task": {
      "type": "string",
      "description": "The natural-language task or routing goal. Include relevant input, constraints, and desired output."
    },
    "required_capability": {
      "type": "string",
      "description": "The capability required for the task. Provide it when known; otherwise let ATL infer it where supported."
    },
    "desired_outcome": {
      "type": "string",
      "description": "Optional result the caller wants from the task."
    },
    "constraints": {
      "type": "object",
      "description": "Optional routing constraints that ATL should honor when selecting an eligible provider or tool."
    },
    "preferences": {
      "type": "object",
      "description": "Optional trade-offs such as cost, latency, or reliability."
    },
    "auth_available": {
      "type": "boolean",
      "description": "Whether the caller can authenticate to the selected provider."
    },
    "max_retries": {
      "type": "integer",
      "minimum": 0,
      "maximum": 2,
      "description": "Optional maximum retries allowed for the decision request."
    },
    "timeout_ms": {
      "type": "integer",
      "minimum": 1,
      "description": "Optional time limit for the decision request in milliseconds."
    }
  },
  "required": [
    "task"
  ],
  "additionalProperties": false
}

ADVANCED DECISION-BOUND EXECUTION TOOL. Execute a previously returned ATL Decision using the exact decision_reference and outcome_correlation_token from atl_decide/next_step; do not invent or alter them and do not choose a provider here. ATL validates the Decision, executes the selected provider/tool, may apply bounded fallback, and records the execution Outcome automatically. Use atl_complete_task for normal task-in/result-out work. Unlike atl_outcome, atl_execute performs the provider/tool execution; atl_outcome only records externally observed results.

{
  "type": "object",
  "properties": {
    "decision_reference": {
      "type": "string",
      "description": "Exact Decision reference returned by ATL. Copy it from atl_decide or next_step; never invent or modify it."
    },
    "outcome_correlation_token": {
      "type": "string",
      "description": "Exact Outcome correlation token returned with the ATL Decision. Copy it unchanged; never invent or modify it."
    },
    "input": {
      "type": "string",
      "description": "The task input to execute with the provider/tool selected by the referenced ATL Decision."
    },
    "tenant_scope": {
      "type": "string",
      "description": "Optional tenant scope for the execution authorization."
    }
  },
  "required": [
    "decision_reference",
    "outcome_correlation_token",
    "input"
  ],
  "additionalProperties": false
}

ADVANCED OUTCOME-RECORDING TOOL. Record the observed result for an existing ATL Decision when execution happened outside ATL or needs explicit feedback. Use the exact decision_reference and outcome_correlation_token returned by ATL; do not invent them. This tool does not execute a provider/tool and does not replace atl_execute. For normal ATL-owned task completion, prefer atl_complete_task, which records Outcome automatically.

{
  "type": "object",
  "properties": {
    "session_id": {
      "type": "string",
      "description": "MCP session ID returned by initialize."
    },
    "request_id": {
      "type": "string",
      "description": "Request identifier associated with the original ATL Decision. Reuse the original value when required by the schema."
    },
    "correlation_id": {
      "type": "string",
      "description": "Correlation identifier associated with the original ATL Decision. Reuse the original value when required by the schema."
    },
    "decision_reference": {
      "type": "string",
      "description": "Exact Decision reference returned by ATL. Copy it from atl_decide or next_step; never invent or modify it."
    },
    "outcome_correlation_token": {
      "type": "string",
      "description": "Exact Outcome correlation token returned with the ATL Decision. Copy it unchanged; never invent or modify it."
    },
    "provider_id": {
      "type": "string",
      "description": "Provider identifier associated with the observed execution. Do not use this field to override ATL's selected provider."
    },
    "final_provider_id": {
      "type": "string",
      "description": "Provider that actually completed execution, if it changed."
    },
    "outcome_status": {
      "type": "string",
      "description": "Observed outcome status for the referenced Decision/execution, such as success or failure, using the schema's allowed value when constrained."
    },
    "observed_latency_ms": {
      "type": "number",
      "description": "Observed provider execution latency in milliseconds."
    },
    "observed_cost": {
      "type": "number",
      "description": "Observed execution cost when known; omit when unavailable unless required by the schema."
    },
    "error_code": {
      "type": "string",
      "description": "Provider error code observed during execution."
    },
    "failure_type": {
      "type": "string",
      "description": "Category of the observed execution failure."
    },
    "http_status": {
      "type": "integer",
      "description": "HTTP status returned by the provider, when applicable."
    },
    "cancelled": {
      "type": "boolean",
      "description": "Whether execution was cancelled before completion."
    },
    "attempt_outcomes": {
      "type": "array",
      "description": "Observed results for individual provider execution attempts."
    }
  },
  "required": [
    "decision_reference",
    "outcome_correlation_token",
    "outcome_status"
  ],
  "additionalProperties": false
}