diff --git a/s08_context_compact/README.en.md b/s08_context_compact/README.en.md
index 6b2f23593..e10fabbdf 100644
--- a/s08_context_compact/README.en.md
+++ b/s08_context_compact/README.en.md
@@ -52,8 +52,21 @@ def snip_compact(messages, max_messages=50):
     if _is_tool_result_message(messages[tail_start]) and _message_has_tool_use(messages[tail_start - 1]):
         tail_start -= 1
     snipped = tail_start - head_end
-    placeholder = {"role": "user", "content": f"[snipped {snipped} messages from conversation middle]"}
-    return messages[:head_end] + [placeholder] + messages[tail_start:]
+
+    snip_text = f"\n\n[System Note: Snipped {snipped} history messages here for context compression]\n\n"
+    
+    head_part = messages[:head_end]
+    tail_part = [msg.copy() for msg in messages[tail_start:]]
+    
+    first_tail_msg = tail_part[0]
+    
+    if isinstance(first_tail_msg["content"], str):
+        first_tail_msg["content"] = snip_text + first_tail_msg["content"]
+        
+    elif isinstance(first_tail_msg["content"], list):
+        first_tail_msg["content"].insert(0, {"type": "text", "text": snip_text})
+
+    return head_part + tail_part
 ```
 
 Messages are still trimmed directly; this just adds one boundary guard. `tool_result` content within remaining messages still keeps accumulating — message #34 may still hold 30KB of old file contents. → L2.
diff --git a/s08_context_compact/README.ja.md b/s08_context_compact/README.ja.md
index 84bfb381a..359e76834 100644
--- a/s08_context_compact/README.ja.md
+++ b/s08_context_compact/README.ja.md
@@ -52,8 +52,21 @@ def snip_compact(messages, max_messages=50):
     if _is_tool_result_message(messages[tail_start]) and _message_has_tool_use(messages[tail_start - 1]):
         tail_start -= 1
     snipped = tail_start - head_end
-    placeholder = {"role": "user", "content": f"[snipped {snipped} messages from conversation middle]"}
-    return messages[:head_end] + [placeholder] + messages[tail_start:]
+
+    snip_text = f"\n\n[System Note: Snipped {snipped} history messages here for context compression]\n\n"
+
+    head_part = messages[:head_end]
+    tail_part = [msg.copy() for msg in messages[tail_start:]]
+
+    first_tail_msg = tail_part[0]
+    
+    if isinstance(first_tail_msg["content"], str):
+        first_tail_msg["content"] = snip_text + first_tail_msg["content"]
+        
+    elif isinstance(first_tail_msg["content"], list):
+        first_tail_msg["content"].insert(0, {"type": "text", "text": snip_text})
+
+    return head_part + tail_part
 ```
 
 切り捨て自体は単純なままで、境界だけを保護する。残ったメッセージ内の `tool_result` 内容はまだ蓄積され続けている。34 番目のメッセージに 30KB の古いファイル内容が残っているかもしれない。→ L2。
diff --git a/s08_context_compact/README.md b/s08_context_compact/README.md
index 22d967156..5f8d40d14 100644
--- a/s08_context_compact/README.md
+++ b/s08_context_compact/README.md
@@ -52,8 +52,27 @@ def snip_compact(messages, max_messages=50):
     if _is_tool_result_message(messages[tail_start]) and _message_has_tool_use(messages[tail_start - 1]):
         tail_start -= 1
     snipped = tail_start - head_end
-    placeholder = {"role": "user", "content": f"[snipped {snipped} messages from conversation middle]"}
-    return messages[:head_end] + [placeholder] + messages[tail_start:]
+
+    # 构造清晰的系统断层通知，告知 Claude 上下文有缺失，避免其产生幻觉
+    snip_text = f"\n\n[System Note: Snipped {snipped} history messages here for context compression]\n\n"
+    
+    # 对后半段进行深拷贝（Deep Copy）或浅拷贝，防止直接修改和污染外部传入的原始 messages 数组
+    head_part = messages[:head_end]
+    tail_part = [msg.copy() for msg in messages[tail_start:]]
+    
+    # 将提示信息“无损”合并到后半段第一条消息中，天然继承原有的角色交替和工具流结构
+    first_tail_msg = tail_part[0]
+    
+    if isinstance(first_tail_msg["content"], str):
+        # 场景 A：普通的纯文本消息
+        first_tail_msg["content"] = snip_text + first_tail_msg["content"]
+        
+    elif isinstance(first_tail_msg["content"], list):
+        # 场景 B：Claude 3 常见的多模态（含图片）或包含 tool_use/tool_result 的 Content Block 列表
+        # 在列表最前端插入一个标准的 text block
+        first_tail_msg["content"].insert(0, {"type": "text", "text": snip_text})
+
+    return head_part + tail_part
 ```
 
 裁掉的是消息本身，只是在切口处多做一步保护；剩下的消息里 `tool_result` 内容仍在累积——第 34 条消息里可能躺着 30KB 的旧文件内容。→ L2。
diff --git a/s08_context_compact/code.py b/s08_context_compact/code.py
index b9d78d425..c04ba397c 100644
--- a/s08_context_compact/code.py
+++ b/s08_context_compact/code.py
@@ -306,8 +306,27 @@ def snip_compact(messages, max_messages=50):
     if head_end >= tail_start:
         return messages
     snipped = tail_start - head_end
-    return messages[:head_end] + [{"role": "user", "content": f"[snipped {snipped} messages]"}] + messages[tail_start:]
 
+    # 构造清晰的系统断层通知，告知 Claude 上下文有缺失，避免其产生幻觉
+    snip_text = f"\n\n[System Note: Snipped {snipped} history messages here for context compression]\n\n"
+    
+    # 对后半段进行深拷贝（Deep Copy）或浅拷贝，防止直接修改和污染外部传入的原始 messages 数组
+    head_part = messages[:head_end]
+    tail_part = [msg.copy() for msg in messages[tail_start:]]
+    
+    # 将提示信息“无损”合并到后半段第一条消息中，天然继承原有的角色交替和工具流结构
+    first_tail_msg = tail_part[0]
+    
+    if isinstance(first_tail_msg["content"], str):
+        # 场景 A：普通的纯文本消息
+        first_tail_msg["content"] = snip_text + first_tail_msg["content"]
+        
+    elif isinstance(first_tail_msg["content"], list):
+        # 场景 B：Claude 3 常见的多模态（含图片）或包含 tool_use/tool_result 的 Content Block 列表
+        # 在列表最前端插入一个标准的 text block
+        first_tail_msg["content"].insert(0, {"type": "text", "text": snip_text})
+
+    return head_part + tail_part
 
 # L2: microCompact — old result placeholders
 def collect_tool_results(messages):