primer/agent-eval

Agent eval navigation

000-nextjs-template

Prompt

Example prompt

Checks

  • node-tests

    vitest.config.scenario.ts
    import {defineConfig} from 'vitest/config'
    
    export default defineConfig({
      test: {
        include: ['scenario.test.ts'],
        reporters: [
          [
            'json',
            {
              outputFile: 'vitest-scenario-report.json',
              includeTaskLocation: true,
            },
          ],
        ],
      },
    })
    
    scenario.test.ts
    import fs from 'node:fs/promises'
    import path from 'node:path'
    import {expect, test} from 'vitest'
    
    const pagePath = path.resolve(import.meta.dirname, 'src', 'app', 'page.tsx')
    const page = await fs.readFile(pagePath, 'utf8')
    
    test('src/app/page.tsx example test', () => {
      expect(page).toMatch(/function IndexPage/)
    })
    

Scenario configuration

import {defineConfig} from '@primer/agent-eval/scenario'
import type {JsonTestResults} from 'vitest/node'

export default defineConfig({
  prompt: `Example prompt`,
  tags: [],
  checks: [
    {
      name: 'node-tests',
      files: ['vitest.config.scenario.ts', 'scenario.test.ts'],
      async run({sandbox}) {
        const commandResult = await sandbox.runCommand(
          'npx',
          ['vitest', 'run', '--config', 'vitest.config.scenario.ts'],
          {allowNonZeroExitCode: true},
        )
        if (commandResult.exitCode !== 0 && commandResult.exitCode !== 1) {
          throw new Error(`Vitest failed with exit code ${commandResult.exitCode}: ${commandResult.stderr}`)
        }
        const contents = await sandbox.readFile('vitest-scenario-report.json')
        const json: JsonTestResults = JSON.parse(contents)
        if (commandResult.exitCode !== 0 && json.numFailedTests === 0) {
          throw new Error(`Vitest failed without reporting failed tests: ${commandResult.stderr}`)
        }
        return {
          outcomes: json.testResults.flatMap(({assertionResults}) => {
            return assertionResults.map(assertion => {
              return {
                type: 'outcome',
                id: assertion.fullName,
                status: assertion.status === 'passed' ? 'passed' : assertion.status === 'failed' ? 'failed' : 'skipped',
              }
            })
          }),
        }
      },
    },
  ],
})

Experiments

This scenario is not included in any experiments.