Fix AttributeError when Anthropic response leads with a thinking block

AnthropicClient.completion()/acompletion() indexed response.content[0]
directly, assuming it was always the text block. When extended thinking
is enabled, Anthropic returns a ThinkingBlock first, which has no .text
attribute and crashes the call. Extract the first block with type=="text"
instead, and raise a clear error if none is present.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Author: potsju

rlm/clients/anthropic.py

@@ -7,6 +7,15 @@
 from rlm.core.types import ModelUsageSummary, UsageSummary
 
 
+def _extract_text(content: list[Any]) -> str:
+    """Return the first text block's content, skipping any thinking blocks
+    that may precede it (e.g. when extended thinking is enabled)."""
+    for block in content:
+        if block.type == "text":
+            return block.text
+    raise ValueError("Anthropic response contained no text block")
+
+
 class AnthropicClient(BaseLM):
     """
     LM Client for running models with the Anthropic API.
@@ -44,7 +53,7 @@ def completion(self, prompt: str | list[dict[str, Any]], model: str | None = Non
 
         response = self.client.messages.create(**kwargs)
         self._track_cost(response, model)
-        return response.content[0].text
+        return _extract_text(response.content)
 
     async def acompletion(
         self, prompt: str | list[dict[str, Any]], model: str | None = None
@@ -61,7 +70,7 @@ async def acompletion(
 
         response = await self.async_client.messages.create(**kwargs)
         self._track_cost(response, model)
-        return response.content[0].text
+        return _extract_text(response.content)
 
     def _prepare_messages(
         self, prompt: str | list[dict[str, Any]]

tests/clients/test_anthropic.py

@@ -0,0 +1,165 @@
+"""Tests for the Anthropic client."""
+
+import os
+from unittest.mock import MagicMock, patch
+
+import pytest
+from dotenv import load_dotenv
+
+from rlm.clients.anthropic import AnthropicClient, _extract_text
+from rlm.core.types import ModelUsageSummary, UsageSummary
+
+load_dotenv()
+
+
+def _text_block(text: str) -> MagicMock:
+    block = MagicMock()
+    block.type = "text"
+    block.text = text
+    return block
+
+
+def _thinking_block() -> MagicMock:
+    block = MagicMock(spec=["type", "thinking"])
+    block.type = "thinking"
+    block.thinking = "reasoning..."
+    return block
+
+
+class TestExtractText:
+    """Unit tests for the module-level _extract_text helper."""
+
+    def test_text_only(self):
+        assert _extract_text([_text_block("hello")]) == "hello"
+
+    def test_thinking_then_text(self):
+        content = [_thinking_block(), _text_block("hello")]
+        assert _extract_text(content) == "hello"
+
+    def test_no_text_block_raises(self):
+        with pytest.raises(ValueError, match="no text block"):
+            _extract_text([_thinking_block()])
+
+
+class TestAnthropicClientUnit:
+    """Unit tests that don't require API calls."""
+
+    def test_init_with_api_key(self):
+        with (
+            patch("rlm.clients.anthropic.anthropic.Anthropic"),
+            patch("rlm.clients.anthropic.anthropic.AsyncAnthropic"),
+        ):
+            client = AnthropicClient(api_key="test-key", model_name="claude-sonnet-5")
+            assert client.model_name == "claude-sonnet-5"
+
+    def test_completion_requires_model(self):
+        with (
+            patch("rlm.clients.anthropic.anthropic.Anthropic"),
+            patch("rlm.clients.anthropic.anthropic.AsyncAnthropic"),
+        ):
+            client = AnthropicClient(api_key="test-key", model_name=None)
+            with pytest.raises(ValueError, match="Model name is required"):
+                client.completion("Hello")
+
+    def test_completion_with_mocked_response(self):
+        mock_response = MagicMock()
+        mock_response.content = [_text_block("Hello from Claude!")]
+        mock_response.usage.input_tokens = 10
+        mock_response.usage.output_tokens = 5
+
+        with (
+            patch("rlm.clients.anthropic.anthropic.Anthropic") as mock_client_class,
+            patch("rlm.clients.anthropic.anthropic.AsyncAnthropic"),
+        ):
+            mock_client = MagicMock()
+            mock_client.messages.create.return_value = mock_response
+            mock_client_class.return_value = mock_client
+
+            client = AnthropicClient(api_key="test-key", model_name="claude-sonnet-5")
+            result = client.completion("Hello")
+
+            assert result == "Hello from Claude!"
+            assert client.model_call_counts["claude-sonnet-5"] == 1
+            assert client.model_input_tokens["claude-sonnet-5"] == 10
+            assert client.model_output_tokens["claude-sonnet-5"] == 5
+
+    def test_completion_with_leading_thinking_block(self):
+        """Regression test: a ThinkingBlock preceding the text block must not crash."""
+        mock_response = MagicMock()
+        mock_response.content = [_thinking_block(), _text_block("Hello from Claude!")]
+        mock_response.usage.input_tokens = 10
+        mock_response.usage.output_tokens = 5
+
+        with (
+            patch("rlm.clients.anthropic.anthropic.Anthropic") as mock_client_class,
+            patch("rlm.clients.anthropic.anthropic.AsyncAnthropic"),
+        ):
+            mock_client = MagicMock()
+            mock_client.messages.create.return_value = mock_response
+            mock_client_class.return_value = mock_client
+
+            client = AnthropicClient(api_key="test-key", model_name="claude-sonnet-5")
+            result = client.completion("Hello")
+
+            assert result == "Hello from Claude!"
+
+    @pytest.mark.anyio
+    async def test_acompletion_with_leading_thinking_block(self):
+        mock_response = MagicMock()
+        mock_response.content = [_thinking_block(), _text_block("Hello from Claude!")]
+        mock_response.usage.input_tokens = 10
+        mock_response.usage.output_tokens = 5
+
+        with (
+            patch("rlm.clients.anthropic.anthropic.Anthropic"),
+            patch("rlm.clients.anthropic.anthropic.AsyncAnthropic") as mock_async_client_class,
+        ):
+            mock_async_client = MagicMock()
+
+            async def _create(**kwargs):
+                return mock_response
+
+            mock_async_client.messages.create = _create
+            mock_async_client_class.return_value = mock_async_client
+
+            client = AnthropicClient(api_key="test-key", model_name="claude-sonnet-5")
+            result = await client.acompletion("Hello")
+
+            assert result == "Hello from Claude!"
+
+    def test_get_usage_summary_empty(self):
+        with (
+            patch("rlm.clients.anthropic.anthropic.Anthropic"),
+            patch("rlm.clients.anthropic.anthropic.AsyncAnthropic"),
+        ):
+            client = AnthropicClient(api_key="test-key", model_name="claude-sonnet-5")
+            summary = client.get_usage_summary()
+            assert isinstance(summary, UsageSummary)
+            assert summary.model_usage_summaries == {}
+
+    def test_get_last_usage(self):
+        with (
+            patch("rlm.clients.anthropic.anthropic.Anthropic"),
+            patch("rlm.clients.anthropic.anthropic.AsyncAnthropic"),
+        ):
+            client = AnthropicClient(api_key="test-key", model_name="claude-sonnet-5")
+            client.last_prompt_tokens = 100
+            client.last_completion_tokens = 50
+            usage = client.get_last_usage()
+            assert isinstance(usage, ModelUsageSummary)
+            assert usage.total_calls == 1
+            assert usage.total_input_tokens == 100
+            assert usage.total_output_tokens == 50
+
+
+class TestAnthropicClientIntegration:
+    """Integration tests that require a real API key."""
+
+    @pytest.mark.skipif(
+        not os.environ.get("ANTHROPIC_API_KEY"),
+        reason="ANTHROPIC_API_KEY not set",
+    )
+    def test_simple_completion(self):
+        client = AnthropicClient(model_name="claude-sonnet-5")
+        result = client.completion("What is 2+2? Reply with just the number.")
+        assert "4" in result