Skip to content

Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.

fix: Extend Anthropic format tool protection to IntelligentContextManager

chopratejasd7d50fef602d
diff --git a/headroom/transforms/intelligent_context.py b/headroom/transforms/intelligent_context.py
index bbfb23edf..a0bb155ed 100644
--- a/headroom/transforms/intelligent_context.py
+++ b/headroom/transforms/intelligent_context.py
@@ -683,15 +683,47 @@ class IntelligentContextManager(Transform):
i -= 1
- # Protect tool responses for protected assistant messages
+ # Also protect any tool responses that belong to protected assistant messages
for i in list(protected):
msg = messages[i]
- if msg.get("role") == "assistant" and msg.get("tool_calls"):
- tool_call_ids = {tc.get("id") for tc in msg.get("tool_calls", [])}
- for j, other_msg in enumerate(messages):
- if other_msg.get("role") == "tool":
- if other_msg.get("tool_call_id") in tool_call_ids:
- protected.add(j)
+ if msg.get("role") == "assistant":
+ tool_call_ids: set[str] = set()
+
+ # OpenAI format: tool_calls array
+ if msg.get("tool_calls"):
+ tool_call_ids.update(
+ tc.get("id") for tc in msg.get("tool_calls", []) if tc.get("id")
+ )
+
+ # Anthropic format: content blocks with type=tool_use
+ content = msg.get("content")
+ if isinstance(content, list):
+ for block in content:
+ if isinstance(block, dict) and block.get("type") == "tool_use":
+ tc_id = block.get("id")
+ if tc_id:
+ tool_call_ids.add(tc_id)
+
+ # Find and protect corresponding tool responses
+ if tool_call_ids:
+ for j, other_msg in enumerate(messages):
+ # OpenAI format: role="tool"
+ if other_msg.get("role") == "tool":
+ if other_msg.get("tool_call_id") in tool_call_ids:
+ protected.add(j)
+
+ # Anthropic format: role="user" with tool_result blocks
+ if other_msg.get("role") == "user":
+ other_content = other_msg.get("content")
+ if isinstance(other_content, list):
+ for block in other_content:
+ if (
+ isinstance(block, dict)
+ and block.get("type") == "tool_result"
+ and block.get("tool_use_id") in tool_call_ids
+ ):
+ protected.add(j)
+ break
return protected
diff --git a/tests/test_transforms/test_intelligent_context.py b/tests/test_transforms/test_intelligent_context.py
index 558c5225a..02cb76aa1 100644
--- a/tests/test_transforms/test_intelligent_context.py
+++ b/tests/test_transforms/test_intelligent_context.py
@@ -1775,3 +1775,374 @@ class TestSummarizeEdgeCases:
assert result.messages is not None
# No summaries created due to exception
assert len(result.summaries_created) == 0
+
+
+# =============================================================================
+# Test Anthropic Format Tool Protection
+# =============================================================================
+
+
+class TestAnthropicFormatToolProtection:
+ """Tests for Anthropic format tool_use/tool_result protection.
+
+ These tests verify that IntelligentContextManager correctly handles
+ Anthropic's native format where:
+ - tool_use blocks appear in assistant.content[]
+ - tool_result blocks appear in user.content[]
+
+ This is critical for Claude Code integration.
+ """
+
+ @pytest.fixture
+ def anthropic_tool_conversation(self) -> list[dict[str, Any]]:
+ """Conversation with Anthropic format tool_use/tool_result."""
+ return [
+ {"role": "system", "content": "You are a helpful assistant."},
+ {"role": "user", "content": "Take a screenshot of the page."},
+ {
+ "role": "assistant",
+ "content": [
+ {"type": "text", "text": "I'll take a screenshot for you."},
+ {
+ "type": "tool_use",
+ "id": "toolu_screenshot_1",
+ "name": "browser_screenshot",
+ "input": {},
+ },
+ ],
+ },
+ {
+ "role": "user",
+ "content": [
+ {
+ "type": "tool_result",
+ "tool_use_id": "toolu_screenshot_1",
+ "content": "Screenshot captured successfully: [base64 image data]",
+ }
+ ],
+ },
+ {
+ "role": "assistant",
+ "content": "I've captured the screenshot. The page shows a login form.",
+ },
+ {"role": "user", "content": "Now click the submit button."},
+ {
+ "role": "assistant",
+ "content": [
+ {"type": "text", "text": "Clicking the submit button."},
+ {
+ "type": "tool_use",
+ "id": "toolu_click_1",
+ "name": "browser_click",
+ "input": {"selector": "#submit"},
+ },
+ ],
+ },
+ {
+ "role": "user",
+ "content": [
+ {
+ "type": "tool_result",
+ "tool_use_id": "toolu_click_1",
+ "content": "Clicked element #submit",
+ }
+ ],
+ },
+ {"role": "assistant", "content": "Done! The form has been submitted."},
+ {"role": "user", "content": "Thanks!"},
+ ]
+
+ @pytest.fixture
+ def anthropic_multiple_tools_same_message(self) -> list[dict[str, Any]]:
+ """Multiple Anthropic tool_use blocks in same assistant message."""
+ return [
+ {"role": "system", "content": "You are a code assistant."},
+ {"role": "user", "content": "Read both config files."},
+ {
+ "role": "assistant",
+ "content": [
+ {"type": "text", "text": "I'll read both files."},
+ {
+ "type": "tool_use",
+ "id": "toolu_read_1",
+ "name": "Read",
+ "input": {"file_path": "/etc/config1.json"},
+ },
+ {
+ "type": "tool_use",
+ "id": "toolu_read_2",
+ "name": "Read",
+ "input": {"file_path": "/etc/config2.json"},
+ },
+ ],
+ },
+ {
+ "role": "user",
+ "content": [
+ {
+ "type": "tool_result",
+ "tool_use_id": "toolu_read_1",
+ "content": '{"setting1": "value1"}',
+ },
+ {
+ "type": "tool_result",
+ "tool_use_id": "toolu_read_2",
+ "content": '{"setting2": "value2"}',
+ },
+ ],
+ },
+ {"role": "assistant", "content": "Both config files have been read."},
+ {"role": "user", "content": "Great, thanks!"},
+ ]
+
+ def test_anthropic_tool_result_protected_when_tool_use_protected(
+ self,
+ anthropic_tool_conversation: list[dict[str, Any]],
+ tokenizer: Tokenizer,
+ ):
+ """Tool_result user messages should be protected when their tool_use is protected."""
+ config = IntelligentContextConfig(keep_last_turns=2)
+ manager = IntelligentContextManager(config=config)
+
+ protected = manager._get_protected_indices(anthropic_tool_conversation)
+
+ # The last tool_use is at index 6, tool_result at index 7
+ # With keep_last_turns=2, indices 6-9 should be protected
+ # The tool_result at 7 should be protected
+
+ # Check that if assistant with tool_use is protected, its tool_result is too
+ for i in protected:
+ msg = anthropic_tool_conversation[i]
+ if msg.get("role") == "assistant":
+ content = msg.get("content")
+ if isinstance(content, list):
+ tool_use_ids = set()
+ for block in content:
+ if isinstance(block, dict) and block.get("type") == "tool_use":
+ tool_use_ids.add(block.get("id"))
+
+ # Find the corresponding tool_result message
+ if tool_use_ids:
+ for j, other_msg in enumerate(anthropic_tool_conversation):
+ if other_msg.get("role") == "user":
+ other_content = other_msg.get("content")
+ if isinstance(other_content, list):
+ for block in other_content:
+ if (
+ isinstance(block, dict)
+ and block.get("type") == "tool_result"
+ and block.get("tool_use_id") in tool_use_ids
+ ):
+ assert j in protected, (
+ f"Tool_result at {j} should be protected "
+ f"because tool_use at {i} is protected"
+ )
+
+ def test_anthropic_tool_units_dropped_atomically(
+ self,
+ anthropic_tool_conversation: list[dict[str, Any]],
+ tokenizer: Tokenizer,
+ ):
+ """Anthropic tool_use and tool_result should be dropped together."""
+ config = IntelligentContextConfig(keep_last_turns=1)
+ manager = IntelligentContextManager(config=config)
+
+ # Force dropping by using small limit
+ result = manager.apply(
+ anthropic_tool_conversation,
+ tokenizer,
+ model_limit=300,
+ output_buffer=50,
+ )
+
+ # Verify no orphaned tool_results
+ tool_use_ids_present = set()
+ tool_result_ids_present = set()
+
+ for msg in result.messages:
+ content = msg.get("content")
+ if isinstance(content, list):
+ for block in content:
+ if isinstance(block, dict):
+ if block.get("type") == "tool_use":
+ tool_use_ids_present.add(block.get("id"))
+ elif block.get("type") == "tool_result":
+ tool_result_ids_present.add(block.get("tool_use_id"))
+
+ # Every tool_result should have its tool_use present
+ for result_id in tool_result_ids_present:
+ assert result_id in tool_use_ids_present, (
+ f"Orphaned tool_result with tool_use_id={result_id}"
+ )
+
+ def test_anthropic_multiple_tools_same_message_atomic(
+ self,
+ anthropic_multiple_tools_same_message: list[dict[str, Any]],
+ tokenizer: Tokenizer,
+ ):
+ """Multiple tool_use blocks in same message should be handled atomically."""
+ config = IntelligentContextConfig(keep_last_turns=1)
+ manager = IntelligentContextManager(config=config)
+
+ result = manager.apply(
+ anthropic_multiple_tools_same_message,
+ tokenizer,
+ model_limit=200,
+ output_buffer=50,
+ )
+
+ # Check atomicity
+ tool_use_ids = set()
+ tool_result_ids = set()
+
+ for msg in result.messages:
+ content = msg.get("content")
+ if isinstance(content, list):
+ for block in content:
+ if isinstance(block, dict):
+ if block.get("type") == "tool_use":
+ tool_use_ids.add(block.get("id"))
+ elif block.get("type") == "tool_result":
+ tool_result_ids.add(block.get("tool_use_id"))
+
+ # All tool_results should have their tool_use
+ for result_id in tool_result_ids:
+ assert result_id in tool_use_ids, f"Orphaned tool_result: {result_id}"
+
+ def test_anthropic_format_no_api_error_scenario(
+ self,
+ anthropic_tool_conversation: list[dict[str, Any]],
+ tokenizer: Tokenizer,
+ ):
+ """Verify the specific scenario that causes 'unexpected tool_use_id' error is fixed."""
+ # This test specifically verifies the bug fix for:
+ # "unexpected tool_use_id found in tool_result blocks"
+
+ config = IntelligentContextConfig(
+ keep_last_turns=1,
+ keep_system=True,
+ )
+ manager = IntelligentContextManager(config=config)
+
+ # Use a limit that would cause dropping
+ result = manager.apply(
+ anthropic_tool_conversation,
+ tokenizer,
+ model_limit=250,
+ output_buffer=50,
+ )
+
+ # Simulate what the API would check
+ tool_use_ids_in_conversation = set()
+ tool_result_ids_in_conversation = set()
+
+ for msg in result.messages:
+ content = msg.get("content")
+ if isinstance(content, list):
+ for block in content:
+ if isinstance(block, dict):
+ if block.get("type") == "tool_use":
+ tool_use_ids_in_conversation.add(block.get("id"))
+ elif block.get("type") == "tool_result":
+ tool_result_ids_in_conversation.add(block.get("tool_use_id"))
+
+ # API error condition: tool_result references a tool_use_id that doesn't exist
+ orphaned_results = tool_result_ids_in_conversation - tool_use_ids_in_conversation
+
+ assert len(orphaned_results) == 0, (
+ f"Would cause API error! Orphaned tool_result ids: {orphaned_results}"
+ )
+
+ def test_mixed_openai_and_anthropic_formats(
+ self,
+ tokenizer: Tokenizer,
+ ):
+ """Both OpenAI and Anthropic formats should work together."""
+ messages = [
+ {"role": "system", "content": "You are helpful."},
+ {"role": "user", "content": "Do things."},
+ # OpenAI format tool call
+ {
+ "role": "assistant",
+ "tool_calls": [
+ {
+ "id": "call_openai_1",
+ "type": "function",
+ "function": {"name": "openai_tool", "arguments": "{}"},
+ }
+ ],
+ },
+ {
+ "role": "tool",
+ "tool_call_id": "call_openai_1",
+ "content": "OpenAI tool result",
+ },
+ {"role": "assistant", "content": "OpenAI tool done."},
+ {"role": "user", "content": "Now use Anthropic format."},
+ # Anthropic format tool call
+ {
+ "role": "assistant",
+ "content": [
+ {
+ "type": "tool_use",
+ "id": "toolu_anthropic_1",
+ "name": "anthropic_tool",
+ "input": {},
+ }
+ ],
+ },
+ {
+ "role": "user",
+ "content": [
+ {
+ "type": "tool_result",
+ "tool_use_id": "toolu_anthropic_1",
+ "content": "Anthropic tool result",
+ }
+ ],
+ },
+ {"role": "assistant", "content": "All done!"},
+ {"role": "user", "content": "Thanks!"},
+ ]
+
+ config = IntelligentContextConfig(keep_last_turns=1)
+ manager = IntelligentContextManager(config=config)
+
+ result = manager.apply(
+ messages,
+ tokenizer,
+ model_limit=200,
+ output_buffer=50,
+ )
+
+ # Verify no orphaned tools of either format
+ openai_call_ids = set()
+ openai_result_ids = set()
+ anthropic_use_ids = set()
+ anthropic_result_ids = set()
+
+ for msg in result.messages:
+ # OpenAI format
+ if msg.get("tool_calls"):
+ for tc in msg["tool_calls"]:
+ openai_call_ids.add(tc.get("id"))
+ if msg.get("role") == "tool":
+ openai_result_ids.add(msg.get("tool_call_id"))
+
+ # Anthropic format
+ content = msg.get("content")
+ if isinstance(content, list):
+ for block in content:
+ if isinstance(block, dict):
+ if block.get("type") == "tool_use":
+ anthropic_use_ids.add(block.get("id"))
+ elif block.get("type") == "tool_result":
+ anthropic_result_ids.add(block.get("tool_use_id"))
+
+ # Check OpenAI format
+ for result_id in openai_result_ids:
+ assert result_id in openai_call_ids, f"Orphaned OpenAI tool: {result_id}"
+
+ # Check Anthropic format
+ for result_id in anthropic_result_ids:
+ assert result_id in anthropic_use_ids, f"Orphaned Anthropic tool: {result_id}"
diff --git a/tests/test_transforms/test_rolling_window.py b/tests/test_transforms/test_rolling_window.py
index 77096226a..5bde13feb 100644
--- a/tests/test_transforms/test_rolling_window.py
+++ b/tests/test_transforms/test_rolling_window.py
@@ -1,11 +1,20 @@
"""Tests for rolling window transform."""
+import os
+from typing import Any
+
import pytest
from headroom import OpenAIProvider, RollingWindowConfig, Tokenizer
from headroom.parser import find_tool_units
from headroom.transforms import RollingWindow
+# Skip all tests in this module if OPENAI_API_KEY is not set
+pytestmark = pytest.mark.skipif(
+ not os.environ.get("OPENAI_API_KEY"),
+ reason="OPENAI_API_KEY not set - skipping tests that require API access",
+)
+
# Create a shared provider for tests
_provider = OpenAIProvider()
@@ -846,3 +855,364 @@ class TestTransformResult:
# warnings should be a list (possibly empty)
assert isinstance(result.warnings, list)
+
+
+# =============================================================================
+# Test Anthropic Format Tool Protection
+# =============================================================================
+
+
+class TestAnthropicFormatToolProtection:
+ """Tests for Anthropic format tool_use/tool_result protection in RollingWindow.
+
+ These tests verify that RollingWindow correctly handles Anthropic's native format
+ where:
+ - tool_use blocks appear in assistant.content[]
+ - tool_result blocks appear in user.content[]
+
+ This is critical for Claude Code integration.
+ """
+
+ @pytest.fixture
+ def anthropic_tool_conversation(self) -> list[dict[str, Any]]:
+ """Conversation with Anthropic format tool_use/tool_result."""
+ return [
+ {"role": "system", "content": "You are a helpful assistant."},
+ {"role": "user", "content": "Take a screenshot of the page."},
+ {
+ "role": "assistant",
+ "content": [
+ {"type": "text", "text": "I'll take a screenshot for you."},
+ {
+ "type": "tool_use",
+ "id": "toolu_screenshot_1",
+ "name": "browser_screenshot",
+ "input": {},
+ },
+ ],
+ },
+ {
+ "role": "user",
+ "content": [
+ {
+ "type": "tool_result",
+ "tool_use_id": "toolu_screenshot_1",
+ "content": "Screenshot captured successfully: [base64 image data]",
+ }
+ ],
+ },
+ {
+ "role": "assistant",
+ "content": "I've captured the screenshot. The page shows a login form.",
+ },
+ {"role": "user", "content": "Now click the submit button."},
+ {
+ "role": "assistant",
+ "content": [
+ {"type": "text", "text": "Clicking the submit button."},
+ {
+ "type": "tool_use",
+ "id": "toolu_click_1",
+ "name": "browser_click",
+ "input": {"selector": "#submit"},
+ },
+ ],
+ },
+ {
+ "role": "user",
+ "content": [
+ {
+ "type": "tool_result",
+ "tool_use_id": "toolu_click_1",
+ "content": "Clicked element #submit",
+ }
+ ],
+ },
+ {"role": "assistant", "content": "Done! The form has been submitted."},
+ {"role": "user", "content": "Thanks!"},
+ ]
+
+ @pytest.fixture
+ def anthropic_multiple_tools_same_message(self) -> list[dict[str, Any]]:
+ """Multiple Anthropic tool_use blocks in same assistant message."""
+ return [
+ {"role": "system", "content": "You are a code assistant."},
+ {"role": "user", "content": "Read both config files."},
+ {
+ "role": "assistant",
+ "content": [
+ {"type": "text", "text": "I'll read both files."},
+ {
+ "type": "tool_use",
+ "id": "toolu_read_1",
+ "name": "Read",
+ "input": {"file_path": "/etc/config1.json"},
+ },
+ {
+ "type": "tool_use",
+ "id": "toolu_read_2",
+ "name": "Read",
+ "input": {"file_path": "/etc/config2.json"},
+ },
+ ],
+ },
+ {
+ "role": "user",
+ "content": [
+ {
+ "type": "tool_result",
+ "tool_use_id": "toolu_read_1",
+ "content": '{"setting1": "value1"}',
+ },
+ {
+ "type": "tool_result",
+ "tool_use_id": "toolu_read_2",
+ "content": '{"setting2": "value2"}',
+ },
+ ],
+ },
+ {"role": "assistant", "content": "Both config files have been read."},
+ {"role": "user", "content": "Great, thanks!"},
+ ]
+
+ def test_anthropic_tool_result_protected_when_tool_use_protected(
+ self,
+ anthropic_tool_conversation: list[dict[str, Any]],
+ ):
+ """Tool_result user messages should be protected when their tool_use is protected."""
+ config = RollingWindowConfig(keep_last_turns=2)
+ window = RollingWindow(config)
+
+ protected = window._get_protected_indices(anthropic_tool_conversation)
+
+ # Check that if assistant with tool_use is protected, its tool_result is too
+ for i in protected:
+ msg = anthropic_tool_conversation[i]
+ if msg.get("role") == "assistant":
+ content = msg.get("content")
+ if isinstance(content, list):
+ tool_use_ids = set()
+ for block in content:
+ if isinstance(block, dict) and block.get("type") == "tool_use":
+ tool_use_ids.add(block.get("id"))
+
+ # Find the corresponding tool_result message
+ if tool_use_ids:
+ for j, other_msg in enumerate(anthropic_tool_conversation):
+ if other_msg.get("role") == "user":
+ other_content = other_msg.get("content")
+ if isinstance(other_content, list):
+ for block in other_content:
+ if (
+ isinstance(block, dict)
+ and block.get("type") == "tool_result"
+ and block.get("tool_use_id") in tool_use_ids
+ ):
+ assert j in protected, (
+ f"Tool_result at {j} should be protected "
+ f"because tool_use at {i} is protected"
+ )
+
+ def test_anthropic_tool_units_dropped_atomically(
+ self,
+ anthropic_tool_conversation: list[dict[str, Any]],
+ ):
+ """Anthropic tool_use and tool_result should be dropped together."""
+ config = RollingWindowConfig(keep_last_turns=1)
+ window = RollingWindow(config)
+ tokenizer = get_tokenizer()
+
+ # Force dropping by using small limit
+ result = window.apply(
+ anthropic_tool_conversation,
+ tokenizer,
+ model_limit=300,
+ output_buffer=50,
+ )
+
+ # Verify no orphaned tool_results
+ tool_use_ids_present = set()
+ tool_result_ids_present = set()
+
+ for msg in result.messages:
+ content = msg.get("content")
+ if isinstance(content, list):
+ for block in content:
+ if isinstance(block, dict):
+ if block.get("type") == "tool_use":
+ tool_use_ids_present.add(block.get("id"))
+ elif block.get("type") == "tool_result":
+ tool_result_ids_present.add(block.get("tool_use_id"))
+
+ # Every tool_result should have its tool_use present
+ for result_id in tool_result_ids_present:
+ assert result_id in tool_use_ids_present, (
+ f"Orphaned tool_result with tool_use_id={result_id}"
+ )
+
+ def test_anthropic_multiple_tools_same_message_atomic(
+ self,
+ anthropic_multiple_tools_same_message: list[dict[str, Any]],
+ ):
+ """Multiple tool_use blocks in same message should be handled atomically."""
+ config = RollingWindowConfig(keep_last_turns=1)
+ window = RollingWindow(config)
+ tokenizer = get_tokenizer()
+
+ result = window.apply(
+ anthropic_multiple_tools_same_message,
+ tokenizer,
+ model_limit=200,
+ output_buffer=50,
+ )
+
+ # Check atomicity
+ tool_use_ids = set()
+ tool_result_ids = set()
+
+ for msg in result.messages:
+ content = msg.get("content")
+ if isinstance(content, list):
+ for block in content:
+ if isinstance(block, dict):
+ if block.get("type") == "tool_use":
+ tool_use_ids.add(block.get("id"))
+ elif block.get("type") == "tool_result":
+ tool_result_ids.add(block.get("tool_use_id"))
+
+ # All tool_results should have their tool_use
+ for result_id in tool_result_ids:
+ assert result_id in tool_use_ids, f"Orphaned tool_result: {result_id}"
+
+ def test_anthropic_format_no_api_error_scenario(
+ self,
+ anthropic_tool_conversation: list[dict[str, Any]],
+ ):
+ """Verify the specific scenario that causes 'unexpected tool_use_id' error is fixed."""
+ config = RollingWindowConfig(
+ keep_last_turns=1,
+ keep_system=True,
+ )
+ window = RollingWindow(config)
+ tokenizer = get_tokenizer()
+
+ # Use a limit that would cause dropping
+ result = window.apply(
+ anthropic_tool_conversation,
+ tokenizer,
+ model_limit=250,
+ output_buffer=50,
+ )
+
+ # Simulate what the API would check
+ tool_use_ids_in_conversation = set()
+ tool_result_ids_in_conversation = set()
+
+ for msg in result.messages:
+ content = msg.get("content")
+ if isinstance(content, list):
+ for block in content:
+ if isinstance(block, dict):
+ if block.get("type") == "tool_use":
+ tool_use_ids_in_conversation.add(block.get("id"))
+ elif block.get("type") == "tool_result":
+ tool_result_ids_in_conversation.add(block.get("tool_use_id"))
+
+ # API error condition: tool_result references a tool_use_id that doesn't exist
+ orphaned_results = tool_result_ids_in_conversation - tool_use_ids_in_conversation
+
+ assert len(orphaned_results) == 0, (
+ f"Would cause API error! Orphaned tool_result ids: {orphaned_results}"
+ )
+
+ def test_mixed_openai_and_anthropic_formats(self):
+ """Both OpenAI and Anthropic formats should work together."""
+ messages = [
+ {"role": "system", "content": "You are helpful."},
+ {"role": "user", "content": "Do things."},
+ # OpenAI format tool call
+ {
+ "role": "assistant",
+ "tool_calls": [
+ {
+ "id": "call_openai_1",
+ "type": "function",
+ "function": {"name": "openai_tool", "arguments": "{}"},
+ }
+ ],
+ },
+ {
+ "role": "tool",
+ "tool_call_id": "call_openai_1",
+ "content": "OpenAI tool result",
+ },
+ {"role": "assistant", "content": "OpenAI tool done."},
+ {"role": "user", "content": "Now use Anthropic format."},
+ # Anthropic format tool call
+ {
+ "role": "assistant",
+ "content": [
+ {
+ "type": "tool_use",
+ "id": "toolu_anthropic_1",
+ "name": "anthropic_tool",
+ "input": {},
+ }
+ ],
+ },
+ {
+ "role": "user",
+ "content": [
+ {
+ "type": "tool_result",
+ "tool_use_id": "toolu_anthropic_1",
+ "content": "Anthropic tool result",
+ }
+ ],
+ },
+ {"role": "assistant", "content": "All done!"},
+ {"role": "user", "content": "Thanks!"},
+ ]
+
+ config = RollingWindowConfig(keep_last_turns=1)
+ window = RollingWindow(config)
+ tokenizer = get_tokenizer()
+
+ result = window.apply(
+ messages,
+ tokenizer,
+ model_limit=200,
+ output_buffer=50,
+ )
+
+ # Verify no orphaned tools of either format
+ openai_call_ids = set()
+ openai_result_ids = set()
+ anthropic_use_ids = set()
+ anthropic_result_ids = set()
+
+ for msg in result.messages:
+ # OpenAI format
+ if msg.get("tool_calls"):
+ for tc in msg["tool_calls"]:
+ openai_call_ids.add(tc.get("id"))
+ if msg.get("role") == "tool":
+ openai_result_ids.add(msg.get("tool_call_id"))
+
+ # Anthropic format
+ content = msg.get("content")
+ if isinstance(content, list):
+ for block in content:
+ if isinstance(block, dict):
+ if block.get("type") == "tool_use":
+ anthropic_use_ids.add(block.get("id"))
+ elif block.get("type") == "tool_result":
+ anthropic_result_ids.add(block.get("tool_use_id"))
+
+ # Check OpenAI format
+ for result_id in openai_result_ids:
+ assert result_id in openai_call_ids, f"Orphaned OpenAI tool: {result_id}"
+
+ # Check Anthropic format
+ for result_id in anthropic_result_ids:
+ assert result_id in anthropic_use_ids, f"Orphaned Anthropic tool: {result_id}"