ModelCallTrace

CORE · interface · 类型合同

Accumulated facts of one model call, reported on success and on failure.

导入

TypeScript

import type { ModelCallTrace } from '@jangts/agent-core/model';

JavaScript

/** @typedef {import('@jangts/agent-core/model').ModelCallTrace} ModelCallTrace */

其他公开入口：@jangts/agent-core、@jangts/agent-core/model、@jangts/agent-core/types、@jangts/agent-core/structs。

定义

interface ModelCallTrace

继承或实现： InvocationMetadata。

字段结构

字段	类型	必需	说明
callSequence	number	必需	One-based position of the call within its runner's lifetime.
stageId	string	必需	见类型定义及调用场景。
modelId	string | undefined	可省略	Model binding used for the call, when one was configured.
inputChars	number	必需	Counted input characters (messages, instruction and schemas).
inputTokens	number | undefined	可省略	Measured input tokens; present only with an explicit token counter.
inputTokenLimit	number | undefined	可省略	Token ceiling the measurement was compared against.
outputChars	number	必需	Characters received from the provider.
firstOutputMs	number | null | undefined	可省略	Time to first streamed part in milliseconds; null until a part arrived.
outputDurationMs	number | undefined	可省略	Milliseconds from the first part to the end of the stream.
status	InvocationStatus	必需	见类型定义及调用场景。
codes	string[]	必需	Error or protocol codes observed during the call, in occurrence order.
usage	ModelUsage | undefined	可省略	见类型定义及调用场景。
functionCall	{ id: string; name: string; argumentChars: number; } | undefined	可省略	Selected function call, when the response produced one.
calls	readonly { id: string; name: string; argumentChars: number; toolId?: string; }[] | undefined	可省略	All calls produced by a session response; omitted for single-selection transports.
context	{ fragmentUsage: ContextFragmentUsage[]; stablePrefixChars: number; } | undefined	可省略	Context accounting: per-fragment usage and characters shared with the previous call.
runId	string | undefined	可省略	只读。Root run this invocation belongs to, when the host runs inside a diagnostics scope.
spanId	string | undefined	可省略	只读。Diagnostics span of this invocation.
parentSpanId	string | undefined	可省略	只读。Span that caused this invocation; absent for a root invocation.
parentInvocationId	string | undefined	可省略	只读。Enclosing invocation when one call triggers another.
invocationId	string	必需	只读。Opaque per-invocation id shared by receipt, events and diagnostic records.
startedAtMs	number	必需	只读。Start instant in epoch milliseconds.
durationMs	number	必需	只读。Elapsed milliseconds from start to completion.

使用示例

• 过程监听与节点报告：App 订阅 Core 执行事实；节点报告自身 usedModel，并用实例式子日志先记意图、再补结果，Loop 负责身份和日志。
• ModelRunner：输入可以省写，输出合同保持一致：用真实 Provider 端口执行一次模型交互，返回 value 和本次 invocation。
• 实际请求测量与同次准备快照：需要提前处理上下文时才 prepare；多次测量与最终发送复用同一次传输准备。
• 子执行：隔离记录，共享祖先额度：子 Runner 不继承对话和输出接收器，但不能绕过父级预算、取消和期限。

监听任务经过的节点与模型事件

入门 · Node.js

TypeScript

import { AgentCoreEngine, defineWorkflow, route } from '@jangts/agent-core/engine';

type State = { name: string; answer?: string };
const workflow = defineWorkflow<State, string>({
  id: 'greeting',
  nodes: [{
    id: 'greet', description: '根据本次姓名生成问候', startsRound: true,
    run: state => ({ answer: '你好，' + state.name }),
    routes: [route.deliver('answer')],
  }],
});
const agent = new AgentCoreEngine({ workflow, execution: {
  observe: { event(event) { console.log(event.type, event.eventId); } },
} });
console.log(await agent.run({ name: 'Ivan' }));

JavaScript

import { AgentCoreEngine, defineWorkflow, route } from '@jangts/agent-core/engine';
const workflow = defineWorkflow({
    id: 'greeting',
    nodes: [{
            id: 'greet', description: '根据本次姓名生成问候', startsRound: true,
            run: state => ({ answer: '你好，' + state.name }),
            routes: [route.deliver('answer')],
        }],
});
const agent = new AgentCoreEngine({ workflow, execution: {
        observe: { event(event) { console.log(event.type, event.eventId); } },
    } });
console.log(await agent.run({ name: 'Ivan' }));

节点执行时报告业务阶段

配置 · Node.js

节点只报告它知道的事实；Core 负责节点身份和执行关联。

TypeScript

import { AgentCoreEngine, defineWorkflow, route } from '@jangts/agent-core/engine';
type State = { name: string; answer?: string };
const workflow = defineWorkflow<State, string>({ id: 'greeting', nodes: [{
  id: 'greet', description: '生成问候', startsRound: true,
  run(state, context) {
    context.report({ payload: { phase: 'formatting' } });
    return { answer: '你好，' + state.name };
  }, routes: [route.deliver('answer')],
}] });
const agent = new AgentCoreEngine({ workflow, execution: {
  observe: { event(event) { console.log(event.type); } },
} });
console.log(await agent.run({ name: 'Ivan' }));

JavaScript

import { AgentCoreEngine, defineWorkflow, route } from '@jangts/agent-core/engine';
const workflow = defineWorkflow({ id: 'greeting', nodes: [{
            id: 'greet', description: '生成问候', startsRound: true,
            run(state, context) {
                context.report({ payload: { phase: 'formatting' } });
                return { answer: '你好，' + state.name };
            }, routes: [route.deliver('answer')],
        }] });
const agent = new AgentCoreEngine({ workflow, execution: {
        observe: { event(event) { console.log(event.type); } },
    } });
console.log(await agent.run({ name: 'Ivan' }));

相关策略源码

• 共享执行事实源和消费隔离
• 子步骤状态及推进关系
• Core 的基础预算缺省值
• 同次准备与真实输入计量

相关定义

• ContextFragmentUsage
• InvocationMetadata
• InvocationStatus
• ModelUsage