Function runL3Build

scripts/context-execution-benchmark.mjs:1363–1412

({ claudeCommand, model, cwd, mode, contextFile, contextText, tier, maxRetries })

}

async function runL3Build({ claudeCommand, model, cwd, mode, contextFile, contextText, tier, maxRetries }) {
  const basePrompt = buildL3BuildPrompt(mode, contextFile, contextText, tier);
  let totalUsage = { inputTokens: 0, outputTokens: 0, totalTokens: 0 };
  let totalDurationMs = 0;
  let totalCostUsd = 0;
  let lastResult = null;
  let testResult = null;

  for (let attempt = 1; attempt <= maxRetries; attempt++) {
    const prompt = attempt === 1 ? basePrompt : buildRetryPrompt(basePrompt, testResult);
    const args = buildClaudeArgs({ cwd, model });
    const { stdout, stderr, exitCode } = await spawnCapture(claudeCommand, args, {
      cwd,
      input: prompt,
      timeoutMs: CLAUDE_TIMEOUT_MS,
      allowFailure: true,
    });
    const parsed = parseClaudeJson(stdout);
    lastResult = { ...parsed, stdout, stderr, exitCode };

    totalUsage.inputTokens += parsed.usage.inputTokens;
    totalUsage.outputTokens += parsed.usage.outputTokens;
    totalUsage.totalTokens += parsed.usage.totalTokens;
    totalDurationMs += parsed.durationMs;
    totalCostUsd += parsed.costUsd;

    testResult = await runTests(cwd);
    if (testResult.testsTotal > 0 && testResult.testsPassed === testResult.testsTotal) {
      return {
        usage: totalUsage,
        attempts: attempt,
        testResult,
        completed: true,
        durationMs: totalDurationMs,
        costUsd: totalCostUsd,
        lastResult,
      };
    }
  }

  return {
    usage: totalUsage,
    attempts: maxRetries,
    testResult,
    completed: false,
    durationMs: totalDurationMs,
    costUsd: totalCostUsd,
    lastResult,
  };
}

/**
 * L3 Full Workflow: Design → Handoff → Build.