From 888686d858571b102391bab7ff4dc3ee2819ddf3 Mon Sep 17 00:00:00 2001
From: "devin-ai-integration[bot]"
 <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Date: Sat, 10 Oct 2026 17:29:46 -0700
Subject: [PATCH] test(headroom): pin that SDK-echoed null tool_calls never
 reach /v1/compress (#39667)

* fix(headroom): drop null tool_calls before calling /v1/compress

headroom-ai 0.27.0 through 0.30.0 iterate msg.get("tool_calls", []) and fail
with 'NoneType' object is not iterable when an assistant row carries an explicit
tool_calls: null, which the fail-closed guardrail then surfaces as a 502. Strip
the null key from rows sent to the compression service; real tool_calls lists
are forwarded unchanged

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

* test(headroom): pin that SDK-echoed null tool_calls never reach /v1/compress

Main now withholds every assistant row from /v1/compress (#39974), so the
null tool_calls strip added here never runs on a row headroom-ai 0.27-0.30
fails on. Drop the strip and pin the customer's history shape through the
public OpenAI chat translation handler instead.

---------

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: yucheng-berri <293926326+yucheng-berri@users.noreply.github.com>
---
 .../guardrail_hooks/test_headroom.py          | 39 +++++++++++++++++++
 1 file changed, 39 insertions(+)

diff --git a/tests/unit/proxy/guardrails/guardrail_hooks/test_headroom.py b/tests/unit/proxy/guardrails/guardrail_hooks/test_headroom.py
index 6f4b1816150..5bdcdddf4b7 100644
--- a/tests/unit/proxy/guardrails/guardrail_hooks/test_headroom.py
+++ b/tests/unit/proxy/guardrails/guardrail_hooks/test_headroom.py
@@ -2008,6 +2008,45 @@ async def test_apply_guardrail_sends_textless_parts_rows_unflattened(
     assert wire_messages[1]["content"] == "D" * 5000
 
 
+@pytest.mark.asyncio
+async def test_sdk_echoed_null_tool_calls_never_reach_compression_service(guardrail: HeadroomGuardrail):
+    """LIT-6921: OpenAI SDK assistant messages echoed back as history carry
+    ``"tool_calls": null``, which headroom-ai 0.27-0.30 answer with a 503. The
+    chat handler sends no row carrying that null and forwards the echoed rows
+    unchanged."""
+    from openai.types.chat import ChatCompletionMessage
+
+    from litellm.llms.openai.chat.guardrail_translation.handler import OpenAIChatCompletionsHandler
+
+    def _sdk_assistant(**fields: object) -> dict:
+        return ChatCompletionMessage.model_validate({"role": "assistant", **fields}).model_dump()
+
+    tool_call = {"id": "call_1", "type": "function", "function": {"name": "lookup", "arguments": "{}"}}
+    history = [
+        {"role": "user", "content": "E" * 5000},
+        _sdk_assistant(content=None, tool_calls=[tool_call]),
+        {"role": "tool", "tool_call_id": "call_1", "content": "R" * 5000},
+        _sdk_assistant(content="summary"),
+        {"role": "user", "content": "F" * 5000},
+        _sdk_assistant(content="noted"),
+        {"role": "user", "content": "and now?"},
+    ]
+    assert history[3]["tool_calls"] is None
+    sent: dict = {}
+
+    def _echo(**kwargs):
+        sent["messages"] = kwargs["json"]["messages"]
+        return _make_compress_response(json.loads(json.dumps(sent["messages"])))
+
+    data = {"model": "claude-sonnet-5", "messages": json.loads(json.dumps(history))}
+    with patch.object(guardrail.async_handler, "post", new_callable=AsyncMock, side_effect=_echo):
+        result = await OpenAIChatCompletionsHandler().process_input_messages(data=data, guardrail_to_apply=guardrail)
+
+    assert [row["role"] for row in sent["messages"]] == ["user", "tool", "user"]
+    assert all(row.get("tool_calls", []) is not None for row in sent["messages"])
+    assert [result["messages"][index] for index in (1, 3, 5)] == [history[index] for index in (1, 3, 5)]
+
+
 @pytest.mark.asyncio
 async def test_fail_open_returns_original_parts_shapes():
     guardrail = _make_guardrail(unreachable_fallback="fail_open")