Function main

evals/buffbench/main-nightly.ts:10–54

()

import type { AgentEvalResults } from './types'

async function main() {
  const saveTraces = process.argv.includes('--save-traces')

  console.log('Starting nightly buffbench evaluation...')
  console.log('Eval set: codebuff')
  console.log()

  const results = await runBuffBench({
    evalDataPaths: [ path.join(__dirname, 'eval-codebuff.json')],
    agents: ['base2-free'],
    taskConcurrency: 5,
    saveTraces,
  })

  console.log('\nNightly buffbench evaluation completed successfully!')

  // Send email with results
  const recipientEmail = process.env.EVAL_RESULTS_EMAIL || 'team@codebuff.com'
  console.log(`\n📧 Sending buffbench results email to ${recipientEmail}...`)

  const { metadata, metaAnalysis, ...agentResults } = results
  const emailContent = formatBuffBenchEmailContent(
    agentResults,
    metadata,
    metaAnalysis,
  )

  try {
    const emailResult = await sendBasicEmail({
      email: recipientEmail,
      data: emailContent,
      logger: console,
    })

    if (emailResult.success) {
      console.log('✅ BuffBench results email sent successfully!')
    } else {
      console.log('⚠️ Email sending was skipped (likely missing configuration)')
    }
  } catch (emailError) {
    console.error('❌ Failed to send buffbench results email:', emailError)
  }

  process.exit(0)
}

function formatBuffBenchEmailContent(
  results: Record<string, AgentEvalResults>,