Function main

scripts/eval_gemini.ts:179–272

()

}

async function main() {
  const apiKey = process.env.GEMINI_API_KEY;
  if (!apiKey) {
    throw new Error('GEMINI_API_KEY environment variable is required.');
  }

  const {values, positionals} = parseArgs({
    options: {
      model: {
        type: 'string',
        default: 'gemini-3-flash-preview',
      },
      debug: {
        type: 'boolean',
        default: false,
      },
      repeat: {
        type: 'boolean',
        default: false,
      },
      'include-skill': {
        type: 'boolean',
        default: false,
      },
      'server-args': {
        type: 'string',
      },
    },
    allowPositionals: true,
  });

  const modelId = values.model;
  const debug = values.debug;
  const repeat = values.repeat;
  const includeSkill = values['include-skill'];
  const extraServerArgs = values['server-args']
    ? values['server-args'].split(/\s+/)
    : [];

  const scenarioFiles =
    positionals.length > 0
      ? positionals.map(p => path.resolve(p))
      : fs
          .readdirSync(SCENARIOS_DIR)
          .filter(file => file.endsWith('.ts') || file.endsWith('.js'))
          .map(file => path.join(SCENARIOS_DIR, file));

  const server = new TestServer(TestServer.randomPort());
  await server.start();

  let successCount = 0;
  let failureCount = 0;

  try {
    for (const scenarioPath of scenarioFiles) {
      for (let i = 1; i <= (repeat ? 3 : 1); i++) {
        try {
          if (debug) {