diff --git a/src/google/adk/models/anthropic_llm.py b/src/google/adk/models/anthropic_llm.py
index 7dd050a1cb..a74299ce69 100644
--- a/src/google/adk/models/anthropic_llm.py
+++ b/src/google/adk/models/anthropic_llm.py
@@ -146,6 +146,10 @@ def part_to_message_block(
         content = json.dumps(result)
       else:
         content = str(result)
+    elif response_data:
+      # Fallback: serialize any non-standard response dict (e.g. SkillToolset
+      # returns {"skill_name": ..., "instructions": ..., "frontmatter": ...})
+      content = json.dumps(response_data)
 
     return anthropic_types.ToolResultBlockParam(
         tool_use_id=part.function_response.id or "",
diff --git a/tests/unittests/models/test_anthropic_llm.py b/tests/unittests/models/test_anthropic_llm.py
index 3df9f567be..d3037c0d82 100644
--- a/tests/unittests/models/test_anthropic_llm.py
+++ b/tests/unittests/models/test_anthropic_llm.py
@@ -741,6 +741,30 @@ def test_part_to_message_block_with_multiple_content_items():
   assert result["content"] == "First part\nSecond part"
 
 
+def test_part_to_message_block_with_non_standard_response():
+  """Test that part_to_message_block serializes non-standard response dicts.
+
+  Regression test for https://github.com/google/adk-python/issues/4779.
+  SkillToolset returns dicts like {"skill_name": ..., "instructions": ...}
+  that don't contain "content" or "result" keys.  These were silently
+  dropped (empty string), causing Claude to never see the tool output.
+  """
+  import json
+
+  from google.adk.models.anthropic_llm import part_to_message_block
+
+  skill_response = {
+      "skill_name": "search_docs",
+      "instructions": "Use the search API to find documents.",
+      "frontmatter": {"version": "1.0"},
+  }
+  part = types.Part.from_function_response(
+      name="load_skill",
+      response=skill_response,
+  )
+  part.function_response.id = "test_skill_id"
+
+
 def test_part_to_message_block_with_pdf_document():
   """Test that part_to_message_block handles PDF document parts."""
   pdf_data = b"%PDF-1.4 fake pdf content"
@@ -751,6 +775,16 @@ def test_part_to_message_block_with_pdf_document():
   result = part_to_message_block(part)
 
   assert isinstance(result, dict)
+  assert result["tool_use_id"] == "test_skill_id"
+  assert result["type"] == "tool_result"
+  assert not result["is_error"]
+  # Content must be non-empty and contain the original data
+  parsed = json.loads(result["content"])
+  assert parsed["skill_name"] == "search_docs"
+  assert parsed["instructions"] == "Use the search API to find documents."
+  assert parsed["frontmatter"] == {"version": "1.0"}
+
+
   assert result["type"] == "document"
   assert result["source"]["type"] == "base64"
   assert result["source"]["media_type"] == "application/pdf"