feat(agent): harden API-first service

tests/test_retrieve.py

@@ -1,5 +1,7 @@
 """Tests: Hybrid-Retrieval (BM25-only offline): Fusion, Entry-Dedup,
 cross_ref-Erweiterung, leeres Retrieval."""
+from concurrent.futures import ThreadPoolExecutor
+
 import pytest
 
 from agent.retrieve import Retriever
@@ -62,6 +64,14 @@ def test_recency_boost_prefers_newer_stand(retriever):
         assert main[0].entry_id in {"lb-min-01", "lb-min-02"}
 
 
+def test_shared_retriever_is_safe_across_worker_threads(retriever):
+    """FastAPI darf dieselbe Retriever-Instanz in mehreren Threads nutzen."""
+    questions = ["Altersteilzeit Lohnausgleich", "Urlaubsanspruch Werktage"] * 4
+    with ThreadPoolExecutor(max_workers=4) as pool:
+        result_sets = list(pool.map(retriever.search, questions))
+    assert all(results for results in result_sets)
+
+
 def test_stats_report(mini_index):
     r = Retriever(mini_index)
     try: