test(context_management): cover a client compaction summary counting toward the trigger

The trigger also skips the system prompt count when messages and tools alone already pass it

litellm/llms/anthropic/messages/context_management/editors/clear_tool_uses.py

@@ -223,6 +223,8 @@ def _trigger_met(
         messages=messages,
         tools=cast("list[ChatCompletionToolParam] | None", tools),
     )
+    if message_tokens > threshold:
+        return True, message_tokens
     current_tokens: Final = message_tokens + count_system_tokens(model, system)
     verbose_logger.debug("context_management polyfill: current_tokens: %s", current_tokens)
     verbose_logger.debug("context_management polyfill: threshold: %s", threshold)

tests/unit/llms/anthropic/messages/context_management/test_clear_tool_uses.py

@@ -10,7 +10,7 @@ from unittest.mock import AsyncMock, MagicMock
 import pytest
 
 import litellm
-from litellm.llms.anthropic.messages.adapters.handler import _run_polyfill_if_enabled
+from litellm.llms.anthropic.messages.adapters.handler import _run_polyfill_if_enabled, prepare_context_managed_request
 from litellm.llms.anthropic.messages.context_management import (
     AnthropicContextManagementError,
     PolyfillResult,
@@ -724,6 +724,36 @@ def test_input_tokens_trigger_counts_the_system_prompt(system: str | list[dict[s
     assert with_system["cleared_input_tokens"] == met_without_system["cleared_input_tokens"] > 0
 
 
+async def test_input_tokens_trigger_counts_a_client_compaction_summary() -> None:
+    """A client compaction summary moves into the system prompt, so the trigger counts it too."""
+    tail: Final = _make_history(n_pairs=3)
+    edit_spec: Final = {
+        "type": "clear_tool_uses_20250919",
+        "trigger": {"type": "input_tokens", "value": litellm.token_counter(model=MODEL, messages=tail) + 100},
+        "keep": {"type": "tool_uses", "value": 1},
+    }
+    compacted_turn: Final = {"role": "assistant", "content": [{"type": "compaction", "content": "Earlier work. " * 400}]}
+
+    async def prepare(messages: list[dict[str, object]]) -> PolyfillResult | None:
+        return await prepare_context_managed_request(
+            model=MODEL,
+            messages=messages,
+            tools=None,
+            system=None,
+            context_management_spec={"edits": [edit_spec]},
+            litellm_metadata=None,
+            additional_drop_params=None,
+            llm_router=None,
+        )
+
+    without_summary: Final = await prepare(deepcopy(tail))
+    with_summary: Final = await prepare([compacted_turn, *deepcopy(tail)])
+
+    assert without_summary is None or not without_summary.applied_edits
+    assert with_summary is not None
+    assert [edit.get("cleared_tool_uses") for edit in with_summary.applied_edits] == [2]
+
+
 @pytest.mark.parametrize(("clear_tool_inputs", "triggers"), [(False, False), (True, True)])
 def test_tool_uses_trigger_counts_tiny_results_only_when_their_inputs_are_cleared(
     clear_tool_inputs: bool, triggers: bool