fix(responses): don't overwrite existing content-block cache_control; add list-content test

Per Greptile review on the PR:
- The message-level cache_control was unconditionally overwriting the
  last content block's cache_control when that block already carried
  its own (preserved by _transform_responses_api_content_to_chat_completion_content).
  Now only propagates onto the last block when it doesn't already have one.
- Added the missing test coverage for the content-block propagation path
  with list content, which no existing test exercised (all used plain
  string content, so the isinstance(content, list) branch was never hit).

litellm/responses/litellm_completion_transformation/transformation.py

@@ -1430,10 +1430,11 @@ class LiteLLMCompletionResponsesConfig:
             if input_item.get("cache_control"):
                 msg["cache_control"] = input_item.get("cache_control")
                 # Also propagate to the last content element for providers
-                # that read cache_control from content blocks (Anthropic, Bedrock)
+                # that read cache_control from content blocks (Anthropic, Bedrock),
+                # unless that block already carries its own cache_control.
                 if isinstance(msg["content"], list) and len(msg["content"]) > 0:
                     last_element = msg["content"][-1]
-                    if isinstance(last_element, dict):
+                    if isinstance(last_element, dict) and "cache_control" not in last_element:
                         last_element["cache_control"] = input_item["cache_control"]
             return [msg]
 

tests/unit/responses/litellm_completion_transformation/test_litellm_completion_responses.py

@@ -5302,3 +5302,56 @@ class TestCacheControlPreservationOnMessages:
         assert "cache_control" not in result[0]
         assert result[1].get("cache_control") == {"type": "ephemeral"}
         assert result[2].get("cache_control") == {"type": "ephemeral"}
+
+    def test_cache_control_propagated_to_last_content_block_for_list_content(self):
+        """
+        For Anthropic/Bedrock, cache_control must also be set on the last
+        content block (not just the message level) to actually activate
+        prompt caching when content is a list of blocks.
+        """
+        input_item = {
+            "role": "user",
+            "content": [
+                {"type": "input_text", "text": "first block"},
+                {"type": "input_text", "text": "last block"},
+            ],
+            "cache_control": {"type": "ephemeral"},
+        }
+
+        result = LiteLLMCompletionResponsesConfig._transform_responses_api_input_item_to_chat_completion_message(
+            input_item=input_item
+        )
+
+        assert len(result) == 1
+        msg = result[0]
+        assert msg.get("cache_control") == {"type": "ephemeral"}
+        assert isinstance(msg["content"], list)
+        assert "cache_control" not in msg["content"][0]
+        assert msg["content"][-1]["cache_control"] == {"type": "ephemeral"}
+
+    def test_cache_control_does_not_overwrite_existing_block_level_cache_control(self):
+        """
+        If the last content block already carries its own cache_control
+        (preserved by _transform_responses_api_content_to_chat_completion_content),
+        the message-level cache_control must not silently overwrite it.
+        """
+        input_item = {
+            "role": "user",
+            "content": [
+                {
+                    "type": "input_text",
+                    "text": "last block",
+                    "cache_control": {"type": "ephemeral", "ttl": "1h"},
+                },
+            ],
+            "cache_control": {"type": "ephemeral"},
+        }
+
+        result = LiteLLMCompletionResponsesConfig._transform_responses_api_input_item_to_chat_completion_message(
+            input_item=input_item
+        )
+
+        assert len(result) == 1
+        msg = result[0]
+        assert msg.get("cache_control") == {"type": "ephemeral"}
+        assert msg["content"][-1]["cache_control"] == {"type": "ephemeral", "ttl": "1h"}