feat: expand to 36 tests (5 new domains) + cost tracker

新增: 安全(API密钥/供应链)、架构(微服务/技术选型)、故障响应(宕机/post-mortem)、协作(跨团队/代码质量)、成本优化(云成本/预算)

成本追踪: cost-tracker.py 自动记录每次 eval 的 token 消耗和费用
基线: baseline-20260709-v2.json (36/36通过, $0.0090, 32K tokens)

.woodpecker/main.yml

@@ -9,24 +9,13 @@ steps:
       DEEPSEEK_API_KEY:
         from_secret: DEEPSEEK_API_KEY
     commands:
-      - apk add --no-cache git
+      - apk add --no-cache git python3 py3-pip
       - npm install -g promptfoo
       - cd promptfoo-eval
-      - promptfoo eval --max-concurrency 2 --no-cache
-      - promptfoo results --format json > /tmp/eval-results.json
+      - promptfoo eval --max-concurrency 2 --no-cache -o results.json
+      - python3 cost-tracker.py
       - |
-        FAILURES=$(cat /tmp/eval-results.json | grep -o '"failures":[0-9]*' | grep -o '[0-9]*')
-        if [ "$FAILURES" != "0" ] && [ -n "$FAILURES" ]; then
-          echo "❌ Eval FAILED: $FAILURES test failures"
-          exit 1
-        fi
+        FAILURES=$(python3 -c "import json; d=json.load(open('results.json')); s=d['results']['stats']; print(s['failures']+s['errors'])")
+        echo "Tests: $(python3 -c "import json; d=json.load(open('results.json')); s=d['results']['stats']; print(s['successes'])") passed, $FAILURES failed"
+        if [ "$FAILURES" != "0" ]; then exit 1; fi
       - echo "✅ All eval tests passed!"
-
-  report:
-    image: alpine:latest
-    commands:
-      - echo "Hermes CI Eval Baseline - $(date -u '+%Y-%m-%d %H:%M UTC')"
-      - echo "Status: ✅ PASS"
-      - echo "Next steps: Review results and update eval dataset"
-    when:
-      status: success