/**
 * Benchmark Generator — 从进化目标自动生成 Benchmark 探针
 *
 * 所有生成由 LLM Agent 完成，无规则降级。
 */
const evolutionQueries = require('../queries').evolutionQueries;
const { callEvaluator } = require('../evaluator');

/**
 * 用 LLM 为指定 Skill 自动生成 Benchmark 探针
 */
async function generateBenchmark(skillName, goal, options = {}) {
  const { difficulty = 2, probeCount = 2 } = options;

  return callEvaluator(
    `你是 Agent Skill 评估专家。为一个 Skill 设计 Benchmark 测试探针。

每个探针测试一个认知维度：
1. trap — 陷阱规避：容易犯的错误场景
2. positive — 正向能力：处理场景的能力
3. boundary — 边界识别：知道自己的能力边界

6 个认知维度：mentalModels, patternLibrary, judgment, toolFluency, taste, scarTissue
难度等级: ${difficulty}/5`,

    `Skill: ${skillName}
目标: ${goal}
每个维度生成 ${probeCount} 个探针

返回 JSON:
{
  "description": "对 ${skillName} 的评估基准",
  "probes": [
    { "type": "trap", "dimension": "scarTissue", "scenario": "...", "expected": "...", "weight": 1.0 },
    { "type": "positive", "dimension": "patternLibrary", "scenario": "...", "expected": "...", "weight": 1.0 }
  ]
}`,
  );
}

/**
 * 生成并持久化 Benchmark
 */
async function generateAndSaveBenchmark(skillName, goal, options = {}) {
  const { difficulty = 2, probeCount = 2 } = options;

  console.log(`🎯 为 ${skillName} 生成 Benchmark...`);
  const result = await generateBenchmark(skillName, goal, { difficulty, probeCount });

  const benchmarkId = `bench-${skillName}-auto-${Date.now()}`;
  const benchmark = await evolutionQueries.upsertBenchmark({
    id: benchmarkId,
    skillName,
    description: result.description || `${skillName} 自动生成评估基准`,
    difficulty,
    probes: result.probes || [],
    passThreshold: 0.6,
  });

  console.log(`  ✅ 生成 ${(result.probes || []).length} 个探针 (难度 ${difficulty})`);
  return benchmark;
}

/**
 * 为 run 中的 Skill 生成 Benchmark
 */
async function generateRunBenchmarks(runId, goal, options = {}) {
  const genomes = await evolutionQueries.getGenomesByRun(runId, { aliveOnly: true });
  const skillNames = [...new Set(genomes.map(g => g.skill_name))];
  console.log(`🎯 为 ${skillNames.length} 个 Skill 生成 Benchmark\n`);

  const results = [];
  for (const skillName of skillNames) {
    try {
      const bench = await generateAndSaveBenchmark(skillName, goal, options);
      results.push(bench);
    } catch (err) {
      console.error(`  ❌ ${skillName}: ${err.message}`);
    }
  }

  console.log(`\n✅ Benchmark 生成完成: ${results.length}/${skillNames.length}`);
  return results;
}

module.exports = {
  generateBenchmark,
  generateAndSaveBenchmark,
  generateRunBenchmarks,
};