feat: 摘要中间件增加压缩日志,提炼模板编辑框定高

- summarization_middleware: 真正压缩与复用缓存均记录 info 日志,便于无 stream writer 场景(定时任务/渠道/测试)排查
- KnowledgeManage: 提炼模板 system prompt 编辑框固定为 50vh 高度,内部滚动

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

frontend-web/src/components/wiki/KnowledgeManage.tsx

@@ -380,7 +380,7 @@ export function ExtractTemplateManageDialog({
               placeholder="提示词正文(system prompt)。这里直接控制提炼粒度——想保留更多信息,就在此放宽要求、要求结构化更详细的输出。"
               value={draft.system_prompt}
               onChange={(e) => setDraft((d) => ({ ...d, system_prompt: e.target.value }))}
-              className="font-mono text-xs"
+              className="h-[50vh] max-h-[50vh] resize-none overflow-y-auto font-mono text-xs"
             />
             <div className="flex flex-wrap items-center gap-4 text-sm">
               <label className="flex items-center gap-1.5">

offline-backend-20260512/backend/packages/harness/deerflow/agents/middlewares/summarization_middleware.py

@@ -269,6 +269,12 @@ class DeerFlowSummarizationMiddleware(SummarizationMiddleware):
         no-op when there is no active stream writer (e.g. unit tests, non-stream
         runs), so callers never branch.
         """
+        # Always leave a backend trace of the genuine (blocking) compaction, even
+        # when there is no stream writer to render the frontend toast (scheduled
+        # runs, channels, tests). Grep ``上下文压缩`` / ``context compaction`` in
+        # the Gateway log to confirm a real summarization LLM call fired.
+        logger.info("summarization: 上下文压缩(真正摘要,触发一次 LLM 调用) context compaction firing")
+
         try:
             from langgraph.config import get_stream_writer
 
@@ -327,6 +333,14 @@ class DeerFlowSummarizationMiddleware(SummarizationMiddleware):
             if not self._exceeds_trigger_by_count(effective):
                 # Refresh LRU recency and reuse without an LLM call.
                 self._touch_cache_entry(thread_id)
+                logger.info(
+                    "summarization: 上下文压缩(复用缓存摘要,无 LLM 调用) context compaction reusing cached summary "
+                    "thread=%s full_tokens=%d compressed_tokens=%d preserved_msgs=%d",
+                    thread_id,
+                    self.token_counter(messages),
+                    self.token_counter(effective),
+                    len(preserved),
+                )
                 return ("reuse", effective)
             # New messages have pushed the compressed context back over the
             # threshold — fall through and re-summarize a larger prefix.