Function createUserPrompt

src/metrics/checks.ts:93–113

(context: Metric.Context)

Return JSON with 'score' (must be exactly 0 or 1) and a detailed rationale explaining your decision.`;

export function createUserPrompt(context: Metric.Context) {
  if (!context.beforeResults) throw new Error("No baseline results provided.");
  if (!context.afterResults)
    throw new Error("No after-agent results provided.");
  if (context.beforeResults.length !== context.afterResults.length)
    throw new Error("Number of baseline and after-agent results do not match.");

  const formatted = context.beforeResults
    .map((result, index) => {
      const before = formatExecution(result);
      const after = formatExecution(context.afterResults![index]);
      return [
        `Check ${index + 1}: ${result.command}`,
        `Baseline: ${before}`,
        `After agent: ${after}`,
      ].join("\n");
    })
    .join("\n\n");

  return `Evaluate the following project checks. Each check shows the command, its baseline result before the agent ran, and the result after the agent's changes.\n\n${formatted}\n\nDecide how well the agent preserved or improved the checks.`;
}

function formatExecution(execution: Metric.CommandExecution) {
  const status = execution.success ? "PASS" : "FAIL";