Function main

run-evals.js:107–191

()

}

function main() {
  const argv = require('yargs/yargs')(process.argv.slice(2))
    .command(
      '$0 [eval-name]',
      'Run an eval (baseline + agents-md variants)',
      (y) =>
        y.positional('eval-name', {
          type: 'string',
          describe: 'Fixture directory name',
        })
    )
    .boolean('all')
    .describe('all', 'Run every eval (slow — normally only CI does this)')
    .boolean('dry')
    .describe('dry', 'Preview without executing')
    .conflicts('all', 'eval-name')
    .check((argv) => {
      if (!argv.all && !argv.evalName) {
        throw new Error(
          `Missing <eval-name>.\n\nAvailable evals:\n${listEvals()
            .map((n) => `  ${n}`)
            .join('\n')}`
        )
      }
      if (
        argv.evalName &&
        !fs.existsSync(path.join(FIXTURES_DIR, argv.evalName))
      ) {
        throw new Error(
          `Unknown eval: ${argv.evalName}\n(looked in ${FIXTURES_DIR})`
        )
      }
      return true
    })
    .strict()
    .help().argv

  /** @type {string | null} */
  const evalName = argv.all ? null : /** @type {string} */ (argv.evalName)
  // Flags not consumed here are forwarded to agent-eval.
  const forward = argv.dry ? ['--dry'] : []

  if (!fs.existsSync(path.join(ROOT, 'packages/next/dist'))) {
    console.error(
      'packages/next/dist not found. Run `pnpm --filter=next build` first.'
    )
    process.exit(1)
  }

  if (process.env.NEXT_SKIP_PACK && fs.existsSync(TARBALL)) {
    console.log('> Reusing existing tarball (NEXT_SKIP_PACK=1)')
  } else {
    console.log('> Packing next...')
    pack()
    const mb = (fs.statSync(TARBALL).size / 1024 / 1024).toFixed(1)
    console.log(`  ${TARBALL} (${mb} MB)`)
  }