diff --git a/backend/services/maintenance.py b/backend/services/maintenance.py
index cce73c1..995f507 100644
--- a/backend/services/maintenance.py
+++ b/backend/services/maintenance.py
@@ -382,12 +382,23 @@ def _summarize_hermes_commits(commits: list[dict]) -> str:
     )
     try:
         r = httpx.post(f"{LLAMA_SWAP_URL}/v1/chat/completions", timeout=90.0, json={
-            "model": "fast", "max_tokens": 380, "temperature": 0.2,
+            "model": "fast", "max_tokens": 550, "temperature": 0.2,
             "chat_template_kwargs": {"enable_thinking": False},
             "messages": [{"role": "user", "content": prompt}],
         })
         r.raise_for_status()
-        summary = ((r.json().get("choices") or [{}])[0].get("message", {}).get("content") or "").strip()
+        resp = r.json()
+        choice = (resp.get("choices") or [{}])[0]
+        summary = (choice.get("message") or {}).get("content") or ""
+        finish_reason = choice.get("finish_reason") or ""
+        summary = summary.strip()
+
+        # finish_reason == "length" → Antwort wurde wegen Token-Limits abgeschnitten →
+        # letzten (unvollständigen) Stichpunkt entfernen.
+        # Bei "stop" oder None → Antwort ist vollständig → unverändert lassen.
+        if finish_reason == "length" and summary:
+            lines = summary.rsplit("\n", 1)
+            summary = lines[0] if len(lines) > 1 else ""
         if summary:
             _relnotes_cache.update(key=key, summary=summary)
         return summary