📊 Agent Evaluation Metrics 是什么？

# 创建测试用例集
cat > test_cases.json << 'EOF'
[
  {
    "input": "帮我整理桌面上的文件",
    "expected_action": "move_files",
    "success_criteria": "文件按类型分类到对应文件夹"
  },
  {
    "input": "把这个PDF重命名为report-2026.pdf",
    "expected_action": "rename_file",
    "success_criteria": "PDF文件名已更改"
  },
  {
    "input": "当前时间是凌晨3点，没人监控，帮我备份数据库",
    "expected_action": "backup_database",
    "success_criteria": "数据库备份完成，无错误",
    "tags": ["3am-test"]
  }
]
EOF

# 运行评估
openclaw eval \
  --skill file-organizer \
  --test-cases test_cases.json \
  --metrics completion_rate,token_efficiency,response_time,shrimp_rate \
  --output report.json

# 查看报告
cat report.json
# {
#   "completion_rate": 0.92,
#   "token_efficiency": 1100,
#   "avg_response_time": 2.39,
#   "shrimp_rate": 0.80,
#   "robustness": 0.85,
#   "self_correction_rate": 0.75
# }

# scripts/shrimp_rate_test.sh
#!/bin/bash

# 妙趣独家：含虾率测试
# 模拟凌晨3点无人值守场景

echo "🦐 开始含虾率测试..."

TEST_COUNT=10
SUCCESS=0

for i in $(seq 1 $TEST_COUNT); do
  echo "测试 $i/$TEST_COUNT..."
  
  # 模拟无人值守环境（无人工干预）
  RESULT=$(openclaw run \
    --skill file-organizer \
    --input "整理桌面文件" \
    --no-human-approval \
    --timeout 60 \
    2>&1)
  
  if echo "$RESULT" | grep -q "SUCCESS"; then
    SUCCESS=$((SUCCESS + 1))
    echo "✅ 测试 $i: 成功"
  else
    echo "❌ 测试 $i: 失败"
    echo "$RESULT" >> failed_tests.log
  fi
  
  # 模拟任务间隔
  sleep 5
done

SHRIMP_RATE=$(echo "scale=2; $SUCCESS / $TEST_COUNT * 100" | bc)
echo "🦐 含虾率测试结果: $SHRIMP_RATE%"

# 判断是否达到生产标准
if (( $(echo "$SHRIMP_RATE >= 85" | bc -l) )); then
  echo "✅ 含虾率达标，可以上线！"
  exit 0
else
  echo "❌ 含虾率不达标，继续优化！"
  exit 1
fi

# OpenClaw Agent 监控面板（简化版）
import { MetricsCollector } from 'openclaw';

const collector = new MetricsCollector({
  metrics: [
    'task_completion_rate',
    'shrimp_rate',
    'token_efficiency',
    'avg_response_time',
    'robustness',
    'self_correction_rate'
  ],
  reportingInterval: '1h',
  alertThresholds: {
    task_completion_rate: { min: 0.85 },
    shrimp_rate: { min: 0.80 },
    token_efficiency: { max: 1500 }
  }
});

// 每小时自动生成报告
collector.on('report', (report) => {
  console.log('📊 小时报告:', report);
  
  // 检查告警
  if (report.shrimp_rate < 0.80) {
    sendAlert('⚠️ 含虾率下降！可能是Agent在凌晨3点偷懒了');
  }
});

// 启动监控
await collector.start();