Function executeClaudeScenario

scripts/skill-bench.js:647–676

* Run a scenario against the real Claude CLI. * Returns { output: string, error: string|null, timedOut: boolean }

(scenario, claudeCmd, tmpDir)

 * Returns { output: string, error: string|null, timedOut: boolean }
 */
function executeClaudeScenario(scenario, claudeCmd, tmpDir) {
  const [bin, ...binArgs] = claudeCmd.split(' ');
  try {
    const output = execFileSync(
      bin,
      [...binArgs, '--plugin-dir', PLUGIN_ROOT, '--print', '--dangerously-skip-permissions', scenario.input],
      {
        cwd: tmpDir,
        timeout: scenario.timeout,
        encoding: 'utf8',
        env: {
          ...process.env,
          CLAUDE_PROJECT_DIR: tmpDir,
          CLAUDE_CODE_SUBPROCESS_ENV_SCRUB: '1',
        },
        maxBuffer: 10 * 1024 * 1024, // 10MB
      }
    );
    return { output, error: null, timedOut: false };
  } catch (err) {
    if (err.killed) {
      return { output: err.stdout || '', error: 'timed out', timedOut: true };
    }
    return {
      output: err.stdout || '',
      error:  err.message || 'unknown error',
      timedOut: false,
    };
  }
}

function executeCodexScenario(scenario, codexCmd, tmpDir) {
  const outputPath = path.join(tmpDir, '.planning', 'benchmark-results', `${scenario.skill}-${scenario.name}-codex.md`);