AI Agent Trace Failure Analysis

{
  "type": "object",
  "required": [
    "trace"
  ],
  "properties": {
    "trace": {
      "type": "string",
      "description": "Agent log/trace output"
    }
  },
  "additionalProperties": false
}

{
  "type": "object",
  "properties": {
    "_meta": {
      "type": "object",
      "properties": {
        "provenance": {
          "type": "object",
          "properties": {
            "fetched_at": {
              "type": "string",
              "description": "ISO 8601 timestamp of when the analysis was generated."
            }
          },
          "additionalProperties": true
        }
      },
      "description": "Analysis metadata. Only the timestamp below is kept; internal capability/latency/payment/model-provenance fields observed at this same level are not surfaced (see this tool's verification.notes).",
      "additionalProperties": true
    },
    "severity": {
      "type": "string",
      "description": "Observed as \"warning\" (a failing run that still produced a final answer) and \"info\" (a clean run) across the two real fixtures -- other levels (e.g. a hard \"error\"/\"critical\") are plausible but unconfirmed."
    },
    "root_cause": {
      "type": "string",
      "description": "Prose explanation of why the failure happened, or \"No failures detected in execution trace\" on a clean run."
    },
    "step_count": {
      "type": "number",
      "description": "Number of entries in steps_summary."
    },
    "failure_type": {
      "type": "string",
      "description": "A short failure category, e.g. \"tool_error\". The literal string \"none\" on a clean run."
    },
    "trace_length": {
      "type": "number",
      "description": "Character length of the trace text that was analyzed."
    },
    "failure_point": {
      "type": "string",
      "description": "Where in the trace the failure occurred, in prose. The literal string \"none\" on a clean run (crewai-clean-run.json)."
    },
    "steps_summary": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "step": {
            "type": "number",
            "description": "1-indexed step number."
          },
          "action": {
            "type": "string",
            "description": "What happened at this step, in prose."
          },
          "status": {
            "type": "string",
            "description": "Observed values: \"success\", \"failure\", \"warning\" across the two real fixtures."
          }
        },
        "additionalProperties": true
      },
      "description": "One entry per step the analyzer identified in the trace, in order."
    },
    "suggested_fix": {
      "type": "string",
      "description": "Prose suggestion for fixing the failure, or \"N/A - execution completed successfully\" on a clean run."
    },
    "patterns_detected": {
      "type": "array",
      "items": {
        "type": "string"
      },
      "description": "Named failure patterns the analyzer recognized (e.g. \"loop_detected\", \"excessive_retries\", \"no_backoff_strategy\"). Empty on a clean run."
    },
    "token_usage_breakdown": {
      "type": "object",
      "properties": {
        "total": {
          "type": "number"
        },
        "estimated_input_tokens": {
          "type": "number"
        },
        "estimated_output_tokens": {
          "type": "number"
        }
      },
      "description": "Estimated token usage for the analyzed run.",
      "additionalProperties": true
    }
  },
  "description": "The trace-analysis result, as returned at routing.responseResultPath.",
  "additionalProperties": true
}