AgentSpan
Properties
Section titled “Properties”attributes
Section titled “attributes”attributes: object;| Name | Type | Description |
|---|---|---|
|
|
|
‐ |
|
|
|
‐ |
|
|
|
Agent active when the span opened; a turn span keeps its initiating agent across handoffs. |
|
|
|
‐ |
|
|
|
‐ |
|
|
|
‐ |
|
|
|
Prompt-cache read tokens for this turn (delta). |
|
|
|
Prompt-cache write / creation tokens for this turn (delta). |
|
|
|
‐ |
|
|
|
Context-window occupancy at this turn (last prompt tokens — a snapshot, not a delta). |
|
|
|
Out-of-node routing / decide / guard model call. |
|
|
|
‐ |
|
|
|
‐ |
|
|
|
Finish reason from the model call ( |
|
|
|
‐ |
|
|
|
‐ |
|
|
|
Imperative |
|
|
|
‐ |
|
|
|
Per-call input tokens on |
|
|
|
Model id on |
|
|
|
‐ |
|
|
|
‐ |
|
|
|
Per-call output tokens on |
|
|
|
‐ |
|
|
|
How a reply node’s user-facing text was produced. Token accounting reads this. |
|
|
|
‐ |
|
|
|
‐ |
|
|
|
‐ |
|
|
|
SHA-256 of the initiating agent’s validated skill snapshot. |
|
|
|
Skill snapshot hashes observed across all agents in this turn, including handoffs. |
|
|
|
How each collect slot was filled this node ( |
|
|
|
0-based model-call index within the enclosing node loop ( |
|
|
|
Immutable publication/runtime identity. Copied to every span in the turn. |
|
|
|
Input tokens consumed by this turn (delta — for cost attribution). |
|
|
|
Output tokens generated by this turn (delta — for cost attribution). |
|
|
|
‐ |
|
|
|
Milliseconds from Runtime.run() trace start to the first non-empty client text delta. |
endTime?
Section titled “endTime?”optional endTime?: number;events?
Section titled “events?”optional events?: object[];| Name | Type |
|---|---|
|
|
|
|
|
|
|
|
|
kind: SpanKind;name: string;parentSpanId?
Section titled “parentSpanId?”optional parentSpanId?: string;spanId
Section titled “spanId”spanId: string;startTime
Section titled “startTime”startTime: number;status
Section titled “status”status: "error" | "ok";traceId
Section titled “traceId”traceId: string;