在 CI 中執行評分器

在 CI 管線中執行評分器，可提供量化指標來衡量 Agent 品質隨時間的變化。runEvals 函式會透過 Agent 或 Workflow 處理多個測試案例，並傳回彙總分數。

基本設定

你可以使用任何支援 ESM 模組的測試框架，例如 Vitest、Jest 或 Mocha。

建立測試案例

使用 runEvals 針對多個測試案例評估 Agent。此函式接受資料項目陣列，每個項目都包含 input，並可選擇提供 groundTruth 供評分器驗證。

src/mastra/agents/weather-agent.test.ts

import { describe, it, expect } from 'vitest'
import { createScorer, runEvals } from '@mastra/core/evals'
import { weatherAgent } from './weather-agent'
import { locationScorer } from '../scorers/location-scorer'

describe('Weather Agent Tests', () => {
  it('should correctly extract locations from queries', async () => {
    const result = await runEvals({
      data: [
        {
          input: 'weather in Berlin',
          groundTruth: { expectedLocation: 'Berlin', expectedCountry: 'DE' },
        },
        {
          input: 'weather in Berlin, Maryland',
          groundTruth: { expectedLocation: 'Berlin', expectedCountry: 'US' },
        },
        {
          input: 'weather in Berlin, Russia',
          groundTruth: { expectedLocation: 'Berlin', expectedCountry: 'RU' },
        },
      ],
      target: weatherAgent,
      scorers: [locationScorer],
    })

    // Assert aggregate score meets threshold
    expect(result.scores['location-accuracy']).toBe(1)
    expect(result.summary.totalItems).toBe(3)
  })
})

了解結果

runEvals 函式會傳回包含以下內容的物件：

• scores：每個評分器在所有測試案例中的平均分數
• summary.totalItems：處理的測試案例總數

{
  scores: {
    'location-accuracy': 1.0,  // Average score across all items
    'another-scorer': 0.85
  },
  summary: {
    totalItems: 3
  }
}

多種測試情境

為不同的評估情境建立個別測試案例：

src/mastra/agents/weather-agent.test.ts

describe('Weather Agent Tests', () => {
  const locationScorer = createScorer({/* ... */})

  it('should handle location disambiguation', async () => {
    const result = await runEvals({
      data: [
        {
          input: 'weather in Berlin',
          groundTruth: {/* ... */},
        },
        {
          input: 'weather in Berlin, Maryland',
          groundTruth: {/* ... */},
        },
      ],
      target: weatherAgent,
      scorers: [locationScorer],
    })

    expect(result.scores['location-accuracy']).toBe(1)
  })

  it('should handle typos and misspellings', async () => {
    const result = await runEvals({
      data: [
        {
          input: 'weather in Berln',
          groundTruth: { expectedLocation: 'Berlin', expectedCountry: 'DE' },
        },
        {
          input: 'weather in Parris',
          groundTruth: { expectedLocation: 'Paris', expectedCountry: 'FR' },
        },
      ],
      target: weatherAgent,
      scorers: [locationScorer],
    })

    expect(result.scores['location-accuracy']).toBe(1)
  })
})

後續步驟

• 了解如何建立自訂評分器
• 探索內建評分器
• 對啟用記憶體的 Agent 執行評分器
• 閱讀 runEvals API 參考