Skip to content

fix(compaction): also filter out server_tool_use blocks before compaction #1151

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
stephaniepang97 wants to merge 3 commits into
base: main
Choose a base branch
from
+36 −10
Open

fix(compaction): also filter out server_tool_use blocks before compaction #1151

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
77ebc79
fix(compaction): also filter out server_tool_use blocks before compac…
stephaniepang97 Jan 24, 2026
21b733b
fix(compaction): include current message with server tool results
stephaniepang97 Jan 24, 2026
f2f840d
fix(compaction): convert current_message content to dicts
stephaniepang97 Jan 24, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
46 changes: 36 additions & 10 deletions src/anthropic/lib/tools/_beta_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -155,10 +155,14 @@ def _handle_request(self) -> Iterator[RunnerItemT]:
raise NotImplementedError()
yield # type: ignore[unreachable]

def _check_and_compact(self) -> bool:
def _check_and_compact(self, current_message: ParsedBetaMessage[ResponseFormatT] | None = None) -> bool:
"""
Check token usage and compact messages if threshold exceeded.
Returns True if compaction was performed, False otherwise.

Args:
current_message: The current assistant message that hasn't been appended yet.
This is needed to include server tool results in compaction.
"""
if self._compaction_control is None or not self._compaction_control["enabled"]:
return False
Expand All @@ -185,13 +189,21 @@ def _check_and_compact(self) -> bool:

messages = list(self._params["messages"])

# Include current message if provided (contains server tool results not yet appended)
# Convert content to dicts so the filter logic below works correctly
if current_message is not None:
messages.append({
"role": current_message.role,
"content": [block.model_dump() for block in current_message.content]
})

if messages[-1]["role"] == "assistant":
# Remove tool_use blocks from the last message to avoid 400 error
# (tool_use requires tool_result, which we don't have yet)
# Remove tool_use and server_tool_use blocks from the last message to avoid 400 error
# (tool_use/server_tool_use requires corresponding result blocks, which we don't have yet)
non_tool_blocks = [
block
for block in messages[-1]["content"]
if isinstance(block, dict) and block.get("type") != "tool_use"
if isinstance(block, dict) and block.get("type") not in ("tool_use", "server_tool_use")
]

if non_tool_blocks:
Expand Down Expand Up @@ -249,7 +261,8 @@ def __run__(self) -> Iterator[RunnerItemT]:
self._iteration_count += 1

# If the compaction was performed, skip tool call generation this iteration
if not self._check_and_compact():
# Pass current message so compaction can include server tool results
if not self._check_and_compact(message):
response = self.generate_tool_call_response()
if response is None:
log.debug("Tool call was not requested, exiting from tool runner loop.")
Expand Down Expand Up @@ -406,10 +419,14 @@ async def _handle_request(self) -> AsyncIterator[RunnerItemT]:
raise NotImplementedError()
yield # type: ignore[unreachable]

async def _check_and_compact(self) -> bool:
async def _check_and_compact(self, current_message: ParsedBetaMessage[ResponseFormatT] | None = None) -> bool:
"""
Check token usage and compact messages if threshold exceeded.
Returns True if compaction was performed, False otherwise.

Args:
current_message: The current assistant message that hasn't been appended yet.
This is needed to include server tool results in compaction.
"""
if self._compaction_control is None or not self._compaction_control["enabled"]:
return False
Expand All @@ -436,13 +453,21 @@ async def _check_and_compact(self) -> bool:

messages = list(self._params["messages"])

# Include current message if provided (contains server tool results not yet appended)
# Convert content to dicts so the filter logic below works correctly
if current_message is not None:
messages.append({
"role": current_message.role,
"content": [block.model_dump() for block in current_message.content]
})

if messages[-1]["role"] == "assistant":
# Remove tool_use blocks from the last message to avoid 400 error
# (tool_use requires tool_result, which we don't have yet)
# Remove tool_use and server_tool_use blocks from the last message to avoid 400 error
# (tool_use/server_tool_use requires corresponding result blocks, which we don't have yet)
non_tool_blocks = [
block
for block in messages[-1]["content"]
if isinstance(block, dict) and block.get("type") != "tool_use"
if isinstance(block, dict) and block.get("type") not in ("tool_use", "server_tool_use")
]

if non_tool_blocks:
Expand Down Expand Up @@ -500,7 +525,8 @@ async def __run__(self) -> AsyncIterator[RunnerItemT]:
self._iteration_count += 1

# If the compaction was performed, skip tool call generation this iteration
if not await self._check_and_compact():
# Pass current message so compaction can include server tool results
if not await self._check_and_compact(message):
response = await self.generate_tool_call_response()
if response is None:
log.debug("Tool call was not requested, exiting from tool runner loop.")
Expand Down
Loading