From 39e5b0c2d11945d03ce3f450b8a7dcaffc8a8640 Mon Sep 17 00:00:00 2001 From: tin-berri Date: Thu, 27 Aug 2026 21:41:56 -0700 Subject: [PATCH] fix(anthropic): drop and self-heal empty thinking blocks on /v1/messages (#38625) * fix(anthropic): drop and self-heal empty thinking blocks on /v1/messages * test(anthropic): pin early-signature carry across the blank thinking chunk skip --- litellm/llms/anthropic/common_utils.py | 46 ++++-- .../adapters/streaming_iterator.py | 8 +- .../adapters/transformation.py | 7 +- .../messages/handler.py | 27 ++-- .../anthropic_messages/transformation.py | 8 +- ...al_pass_through_adapters_transformation.py | 29 ++++ .../test_streaming_iterator_first_delta.py | 138 ++++++++++++++++++ ...erimental_pass_through_messages_handler.py | 12 +- .../anthropic/test_anthropic_common_utils.py | 122 ++++++++++++---- 9 files changed, 335 insertions(+), 62 deletions(-) diff --git a/litellm/llms/anthropic/common_utils.py b/litellm/llms/anthropic/common_utils.py index c73376ba498..681a8397f66 100644 --- a/litellm/llms/anthropic/common_utils.py +++ b/litellm/llms/anthropic/common_utils.py @@ -974,19 +974,25 @@ def strip_advisor_blocks_from_messages(messages: list[Any], replace_with_text: b return messages -def is_anthropic_invalid_thinking_signature_error(error_text: str) -> bool: +def is_anthropic_invalid_thinking_block_error(error_text: str) -> bool: """ - Detect Anthropic 400 errors caused by missing or invalid thinking signatures. + Detect Anthropic 400 errors caused by invalid thinking blocks in replayed + history: a missing or invalid signature, or a block with empty thinking text. Known error formats: {"message":"messages.2.content.0.thinking.signature.str: Input should be a valid string"} messages.N.content.M.thinking.signature.str: Input should be a valid string messages.N.content.M: Invalid `signature` in `thinking` block + messages.N.content.M.thinking: each thinking block must contain thinking """ if not error_text: return False lower: Final = error_text.lower() - return "thinking" in lower and "signature" in lower and ("invalid" in lower or "valid string" in lower) + if "thinking" not in lower: + return False + if "signature" in lower and ("invalid" in lower or "valid string" in lower): + return True + return "must contain thinking" in lower def strip_thinking_blocks_from_anthropic_messages(messages: list[Any]) -> list[Any]: @@ -1028,22 +1034,29 @@ def strip_thinking_blocks_from_anthropic_messages_request_dict( data.pop("thinking", None) -def strip_empty_text_blocks_from_anthropic_messages( +def strip_empty_content_blocks_from_anthropic_messages( messages: list[Any], ) -> list[Any]: """ Return a new message list with empty or whitespace-only ``{"type": "text"}`` - content blocks removed. + and ``{"type": "thinking"}`` content blocks removed. Anthropic's API rejects requests containing such blocks with - ``"messages: text content blocks must be non-empty"``, but assistant - messages from Anthropic routinely arrive with ``{"type": "text", "text": ""}`` - alongside ``tool_use`` blocks (see anthropics/anthropic-sdk-python#461). + ``"messages: text content blocks must be non-empty"`` and + ``"messages.N.content.M.thinking: each thinking block must contain + thinking"`` respectively. Assistant messages routinely arrive with + ``{"type": "text", "text": ""}`` alongside ``tool_use`` blocks (see + anthropics/anthropic-sdk-python#461), and a turn served by a + non-Anthropic reasoning model through the /v1/messages bridge can carry + ``{"type": "thinking", "thinking": ""}`` when the model produced no + reasoning text (e.g. it went straight to parallel tool calls). Multi-turn tool-use clients (e.g. Claude Code) loop these prior responses back as conversation history, which then causes the next request to 400 on the unified ``/v1/messages`` path. ``/v1/chat/completions`` already handles this in ``anthropic_messages_pt``; this helper provides the equivalent guarantee for the native Anthropic Messages path. + ``redacted_thinking`` blocks are never touched: they carry opaque + ``data`` instead of thinking text. Messages whose content is a list and becomes empty after stripping are omitted, matching :func:`strip_thinking_blocks_from_anthropic_messages`. @@ -1056,7 +1069,7 @@ def strip_empty_text_blocks_from_anthropic_messages( out.append(m) continue content = m["content"] - filtered = [b for b in content if not _is_empty_text_block(b)] + filtered = [b for b in content if not _is_empty_text_block(b) and not is_empty_thinking_block(b)] if len(filtered) == len(content): out.append(m) elif filtered: @@ -1071,6 +1084,21 @@ def _is_empty_text_block(block: Any) -> bool: return not isinstance(text, str) or not text.strip() +def is_empty_thinking_block(block: object) -> bool: + """ + True for a ``{"type": "thinking"}`` content block whose thinking text is + missing, not a string, or empty/whitespace-only after ``.strip()``. + Anthropic rejects such blocks with ``"each thinking block must contain + thinking"`` (whitespace-only included, verified live), regardless of any + signature they carry. ``redacted_thinking`` blocks are a different type + and always return False. + """ + if not isinstance(block, dict) or block.get("type") != "thinking": + return False + thinking: Final = block.get("thinking") + return not isinstance(thinking, str) or not thinking.strip() + + def normalize_anthropic_tool_use_id(raw_id: str) -> str: """ Normalize a tool_use / tool_result id for Anthropic's ``^[a-zA-Z0-9_-]+$`` diff --git a/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py b/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py index 30b5df1e4ee..cefd4aa2d77 100644 --- a/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py +++ b/litellm/llms/anthropic/experimental_pass_through/adapters/streaming_iterator.py @@ -1029,6 +1029,8 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): @staticmethod def _is_blank_delta(chunk: "ModelResponseStream") -> bool: + from litellm.llms.anthropic.common_utils import is_empty_thinking_block + choice: Final = chunk.choices[0] if choice.finish_reason is not None: return False @@ -1039,7 +1041,11 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper): return False if getattr(delta, "reasoning_content", None): return False - if getattr(delta, "thinking_blocks", None): + # thinking_blocks whose entries are all empty (even if signed) must not + # open a block: the emitted {"type": "thinking", "thinking": ""} gets + # replayed as history and Anthropic rejects it (LIT-6357). + thinking_blocks: Final = getattr(delta, "thinking_blocks", None) + if thinking_blocks and any(isinstance(b, dict) and not is_empty_thinking_block(b) for b in thinking_blocks): return False return True diff --git a/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py b/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py index 754f0128a8a..a9fa00c827a 100644 --- a/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py +++ b/litellm/llms/anthropic/experimental_pass_through/adapters/transformation.py @@ -89,7 +89,10 @@ from litellm.litellm_core_utils.prompt_templates.factory import ( from litellm.litellm_core_utils.reasoning_effort_utils import ( reasoning_effort_from_thinking_budget, ) -from litellm.llms.anthropic.common_utils import normalize_anthropic_tool_use_id +from litellm.llms.anthropic.common_utils import ( + is_empty_thinking_block, + normalize_anthropic_tool_use_id, +) from litellm.llms.anthropic.experimental_pass_through.context_management import ( PolyfillResult, ) @@ -1264,6 +1267,8 @@ class LiteLLMAnthropicMessagesAdapter: if hasattr(choice.message, "thinking_blocks") and choice.message.thinking_blocks: for thinking_block in choice.message.thinking_blocks: if thinking_block.get("type") == "thinking": + if is_empty_thinking_block(thinking_block): + continue thinking_value = thinking_block.get("thinking", "") signature_value = thinking_block.get("signature", "") new_content.append( diff --git a/litellm/llms/anthropic/experimental_pass_through/messages/handler.py b/litellm/llms/anthropic/experimental_pass_through/messages/handler.py index 283c706e45e..3e314f76a3e 100644 --- a/litellm/llms/anthropic/experimental_pass_through/messages/handler.py +++ b/litellm/llms/anthropic/experimental_pass_through/messages/handler.py @@ -17,7 +17,7 @@ from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging from litellm.llms.anthropic.common_utils import ( flatten_unencrypted_web_search_results_in_anthropic_messages, sanitize_tool_use_ids_in_anthropic_messages, - strip_empty_text_blocks_from_anthropic_messages, + strip_empty_content_blocks_from_anthropic_messages, ) from litellm.llms.base_llm.anthropic_messages.transformation import ( BaseAnthropicMessagesConfig, @@ -242,17 +242,20 @@ async def anthropic_messages( """ Async: Make llm api request in Anthropic /messages API spec. - Runs the empty-text-block sanitizer before any backend dispatch. + Runs the empty-content-block sanitizer before any backend dispatch. """ # Anthropic's API rejects requests containing empty / whitespace-only - # text content blocks with "messages: text content blocks must be - # non-empty". Multi-turn tool-use clients (e.g. Claude Code) routinely - # loop assistant responses that contain {"type": "text", "text": ""} - # alongside tool_use blocks back as conversation history, which then - # causes the next /v1/messages call to 400. /v1/chat/completions - # already handles this in anthropic_messages_pt; sanitize the native - # Anthropic Messages path here for the same guarantee. See #22930. - messages = strip_empty_text_blocks_from_anthropic_messages(messages) + # text content blocks ("messages: text content blocks must be + # non-empty") and empty thinking blocks ("each thinking block must + # contain thinking"). Multi-turn tool-use clients (e.g. Claude Code) + # routinely loop assistant responses that contain such blocks — an empty + # text block alongside tool_use, or an empty thinking block from a turn + # a non-Anthropic reasoning model served through the bridge — back as + # conversation history, which then causes the next /v1/messages call to + # 400. /v1/chat/completions already handles this in + # anthropic_messages_pt; sanitize the native Anthropic Messages path + # here for the same guarantee. See #22930. + messages = strip_empty_content_blocks_from_anthropic_messages(messages) # Replay of cross-provider tool history (e.g. kimi -> Anthropic) may carry # ids like ``functions.Bash:0`` that violate Anthropic's id pattern. messages = sanitize_tool_use_ids_in_anthropic_messages(messages) @@ -374,7 +377,7 @@ async def anthropic_messages( api_base=api_base, client=client, custom_llm_provider=custom_llm_provider, - # messages were already empty-text-block sanitized at the top of this + # messages were already empty-content-block sanitized at the top of this # function and are NOT reassigned before this dispatch, so the handler # can skip its (otherwise redundant) second full-messages scan. Passed # explicitly (not via **kwargs) so it only affects this direct @@ -451,7 +454,7 @@ def anthropic_messages_handler( # ``_litellm_messages_presanitized`` to skip this redundant second # full-messages scan. Pop it so it never leaks into provider params. if not kwargs.pop("_litellm_messages_presanitized", False): - messages = strip_empty_text_blocks_from_anthropic_messages(messages) + messages = strip_empty_content_blocks_from_anthropic_messages(messages) messages = sanitize_tool_use_ids_in_anthropic_messages(messages) messages = flatten_unencrypted_web_search_results_in_anthropic_messages(messages) diff --git a/litellm/llms/base_llm/anthropic_messages/transformation.py b/litellm/llms/base_llm/anthropic_messages/transformation.py index 6455bb010f4..8e7c22930fa 100644 --- a/litellm/llms/base_llm/anthropic_messages/transformation.py +++ b/litellm/llms/base_llm/anthropic_messages/transformation.py @@ -159,20 +159,20 @@ class BaseAnthropicMessagesConfig(ABC): and issue one more attempt (bounded by max_retry_on_anthropic_messages_http_error). """ from litellm.llms.anthropic.common_utils import ( - is_anthropic_invalid_thinking_signature_error, + is_anthropic_invalid_thinking_block_error, ) - return e.response.status_code == 400 and is_anthropic_invalid_thinking_signature_error(e.response.text) + return e.response.status_code == 400 and is_anthropic_invalid_thinking_block_error(e.response.text) def transform_anthropic_messages_request_on_http_error(self, e: httpx.HTTPStatusError, request_data: dict) -> dict: """ Mutates request_data in place when retrying after a recoverable HTTP error. """ from litellm.llms.anthropic.common_utils import ( - is_anthropic_invalid_thinking_signature_error, + is_anthropic_invalid_thinking_block_error, strip_thinking_blocks_from_anthropic_messages_request_dict, ) - if e.response.status_code == 400 and is_anthropic_invalid_thinking_signature_error(e.response.text): + if e.response.status_code == 400 and is_anthropic_invalid_thinking_block_error(e.response.text): strip_thinking_blocks_from_anthropic_messages_request_dict(request_data) return request_data diff --git a/tests/test_litellm/llms/anthropic/experimental_pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py b/tests/test_litellm/llms/anthropic/experimental_pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py index e6e5cd02a45..ea1813acb82 100644 --- a/tests/test_litellm/llms/anthropic/experimental_pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py +++ b/tests/test_litellm/llms/anthropic/experimental_pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py @@ -1013,6 +1013,35 @@ def test_translate_openai_content_to_anthropic_thinking_and_redacted_thinking(): assert result[1]["data"] == "REDACTED" +def test_translate_openai_content_to_anthropic_drops_empty_thinking_blocks(): + """LIT-6357 non-streaming producer half: a bridged reasoning model whose + thinking_blocks entry has empty or whitespace-only text (signed or not) + must not surface as {"type": "thinking", "thinking": ""} — clients replay + it as history and Anthropic 400s with "each thinking block must contain + thinking". Non-empty thinking and redacted_thinking pass through.""" + openai_choices = [ + Choices( + message=Message( + role="assistant", + content="the answer", + thinking_blocks=[ + {"type": "thinking", "thinking": "", "signature": "sig_abc"}, + {"type": "thinking", "thinking": " \n "}, + {"type": "thinking", "thinking": "real plan", "signature": "sigsig"}, + {"type": "redacted_thinking", "data": "REDACTED"}, + ], + ) + ) + ] + + adapter = LiteLLMAnthropicMessagesAdapter() + result = adapter._translate_openai_content_to_anthropic(choices=openai_choices) + + assert [b["type"] for b in result] == ["thinking", "redacted_thinking", "text"] + assert result[0]["thinking"] == "real plan" + assert result[1]["data"] == "REDACTED" + + def test_translate_streaming_openai_chunk_to_anthropic_thinking_delta(): choices = [ StreamingChoices( diff --git a/tests/test_litellm/llms/anthropic/experimental_pass_through/adapters/test_streaming_iterator_first_delta.py b/tests/test_litellm/llms/anthropic/experimental_pass_through/adapters/test_streaming_iterator_first_delta.py index f64ffb6d233..6268cd01efe 100644 --- a/tests/test_litellm/llms/anthropic/experimental_pass_through/adapters/test_streaming_iterator_first_delta.py +++ b/tests/test_litellm/llms/anthropic/experimental_pass_through/adapters/test_streaming_iterator_first_delta.py @@ -1027,3 +1027,141 @@ async def test_tool_block_start_flush_does_not_duplicate_or_drop_events(is_async ] assert _input_json_deltas(events) == ['{"file_text":', ' "hello"}'] _assert_deltas_match_their_block_type(events) + + +def _thinking_block_starts(events: List[dict]) -> List[dict]: + return [ + e["content_block"] + for e in events + if e.get("type") == "content_block_start" and e["content_block"].get("type") == "thinking" + ] + + +def _empty_thinking_then_tool_chunks(thinking: str = "", signature: str = "") -> List[MagicMock]: + return [ + _thinking_chunk(thinking, signature=signature), + _tool_chunk("call_paris", "get_weather", '{"city": "Paris"}'), + _make_chunk(Delta(content=None), finish_reason="tool_calls"), + ] + + +@pytest.mark.parametrize("is_async", [False, True]) +@pytest.mark.parametrize( + "thinking,signature", + [("", ""), (" \n\t ", ""), ("", "sig_abc")], + ids=["empty", "whitespace-only", "empty-but-signed"], +) +@pytest.mark.asyncio +async def test_contentless_thinking_chunk_opens_no_thinking_block(is_async: bool, thinking: str, signature: str): + """LIT-6357 producer half: a reasoning model that goes straight to tool + calls streams a ``thinking_blocks`` entry with no real thinking text; the + wrapper used to open ``{"type": "thinking", "thinking": ""}`` for it and + close the block with no delta. Clients (Claude Code) replay that block as + history and Anthropic rejects the next tool-loop request with + "each thinking block must contain thinking" — empty-but-signed included. + The contentless chunk must open nothing; the tool_use block must be + unaffected.""" + chunks = _empty_thinking_then_tool_chunks(thinking, signature) + if is_async: + wrapper = AnthropicStreamWrapper(completion_stream=_AsyncStream(chunks), model="claude-x") + events = await _drain_async(wrapper) + else: + wrapper = AnthropicStreamWrapper(completion_stream=iter(chunks), model="claude-x") + events = _drain_sync(wrapper) + + assert _thinking_block_starts(events) == [] + assert _thinking_deltas(events) == [] + tool_starts = [ + e["content_block"] + for e in events + if e.get("type") == "content_block_start" and e["content_block"].get("type") == "tool_use" + ] + assert [b["name"] for b in tool_starts] == ["get_weather"] + _assert_deltas_match_their_block_type(events) + + +@pytest.mark.parametrize("is_async", [False, True]) +@pytest.mark.asyncio +async def test_empty_first_thinking_chunk_then_real_text_still_opens_one_block(is_async: bool): + """The contentless-chunk skip must not eat a thinking stream whose first + chunk is empty but whose later chunks carry real text: exactly one thinking + block opens and the text flows into it.""" + chunks = [ + _thinking_chunk(""), + _thinking_chunk("Let me think"), + _thinking_chunk("", signature="sig123"), + _make_chunk(Delta(content="Hello")), + _make_chunk(Delta(content=None), finish_reason="stop"), + ] + if is_async: + wrapper = AnthropicStreamWrapper(completion_stream=_AsyncStream(chunks), model="claude-x") + events = await _drain_async(wrapper) + else: + wrapper = AnthropicStreamWrapper(completion_stream=iter(chunks), model="claude-x") + events = _drain_sync(wrapper) + + assert len(_thinking_block_starts(events)) == 1 + assert _thinking_deltas(events) == ["Let me think"] + assert _signature_deltas(events) == ["sig123"] + assert _text_deltas(events) == ["Hello"] + _assert_deltas_match_their_block_type(events) + + +@pytest.mark.parametrize("is_async", [False, True]) +@pytest.mark.asyncio +async def test_early_signature_on_blank_thinking_chunk_is_carried_to_the_opened_block(is_async: bool): + """Pins that the blank-chunk skip does not lose an early signature: the + classifier captures the skipped chunk's signature into the pending block + start body, so when real thinking text follows, the opened block still + carries it. Guards the LIT-6357 blank-skip against regressing signature + replay.""" + chunks = [ + _thinking_chunk("", signature="sig_early"), + _thinking_chunk("Let me think"), + _make_chunk(Delta(content="Hello")), + _make_chunk(Delta(content=None), finish_reason="stop"), + ] + if is_async: + wrapper = AnthropicStreamWrapper(completion_stream=_AsyncStream(chunks), model="claude-x") + events = await _drain_async(wrapper) + else: + wrapper = AnthropicStreamWrapper(completion_stream=iter(chunks), model="claude-x") + events = _drain_sync(wrapper) + + starts = _thinking_block_starts(events) + assert len(starts) == 1 + assert starts[0].get("signature") == "sig_early" + assert _thinking_deltas(events) == ["Let me think"] + assert _text_deltas(events) == ["Hello"] + _assert_deltas_match_their_block_type(events) + + +@pytest.mark.parametrize("is_async", [False, True]) +@pytest.mark.asyncio +async def test_early_signature_discarded_when_first_block_is_not_thinking(is_async: bool): + """An early signature from a skipped blank thinking chunk must not leak + into a text or tool_use first block, and must not resurrect an empty + thinking block on its own (an empty-but-signed block is exactly what + Anthropic rejects).""" + chunks = [ + _thinking_chunk("", signature="sig_early"), + _make_chunk(Delta(content="Hello")), + _make_chunk(Delta(content=None), finish_reason="stop"), + ] + if is_async: + wrapper = AnthropicStreamWrapper(completion_stream=_AsyncStream(chunks), model="claude-x") + events = await _drain_async(wrapper) + else: + wrapper = AnthropicStreamWrapper(completion_stream=iter(chunks), model="claude-x") + events = _drain_sync(wrapper) + + assert _thinking_block_starts(events) == [] + text_starts = [ + e["content_block"] + for e in events + if e.get("type") == "content_block_start" and e["content_block"].get("type") == "text" + ] + assert len(text_starts) == 1 + assert "signature" not in text_starts[0] + assert _text_deltas(events) == ["Hello"] + _assert_deltas_match_their_block_type(events) diff --git a/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py b/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py index 5fc4a361e78..ad4c3d6bfbb 100644 --- a/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py +++ b/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py @@ -709,8 +709,8 @@ def test_handler_strips_when_no_presanitized_flag(): with patch.object( handler, - "strip_empty_text_blocks_from_anthropic_messages", - wraps=handler.strip_empty_text_blocks_from_anthropic_messages, + "strip_empty_content_blocks_from_anthropic_messages", + wraps=handler.strip_empty_content_blocks_from_anthropic_messages, ) as spy: result = handler.anthropic_messages_handler( max_tokens=10, @@ -729,8 +729,8 @@ def test_handler_skips_strip_when_presanitized(): with patch.object( handler, - "strip_empty_text_blocks_from_anthropic_messages", - wraps=handler.strip_empty_text_blocks_from_anthropic_messages, + "strip_empty_content_blocks_from_anthropic_messages", + wraps=handler.strip_empty_content_blocks_from_anthropic_messages, ) as spy: result = handler.anthropic_messages_handler( max_tokens=10, @@ -849,8 +849,8 @@ async def test_async_wrapper_sets_presanitized_and_sanitizes_once(): patch("asyncio.get_event_loop", return_value=fake_loop), patch.object( handler, - "strip_empty_text_blocks_from_anthropic_messages", - wraps=handler.strip_empty_text_blocks_from_anthropic_messages, + "strip_empty_content_blocks_from_anthropic_messages", + wraps=handler.strip_empty_content_blocks_from_anthropic_messages, ) as spy, ): await handler.anthropic_messages( diff --git a/tests/test_litellm/llms/anthropic/test_anthropic_common_utils.py b/tests/test_litellm/llms/anthropic/test_anthropic_common_utils.py index b2984795c1c..a2da2cccb7c 100644 --- a/tests/test_litellm/llms/anthropic/test_anthropic_common_utils.py +++ b/tests/test_litellm/llms/anthropic/test_anthropic_common_utils.py @@ -1292,11 +1292,12 @@ class TestPassthroughAuthToken: class TestAnthropicThinkingSignatureSelfHeal: - """Helpers for retrying after invalid encrypted thinking signatures.""" + """Helpers for retrying after invalid thinking blocks in replayed history: + invalid encrypted signatures, and blocks with empty thinking text.""" - def test_is_anthropic_invalid_thinking_signature_error_positive(self): + def test_is_anthropic_invalid_thinking_block_error_positive(self): from litellm.llms.anthropic.common_utils import ( - is_anthropic_invalid_thinking_signature_error, + is_anthropic_invalid_thinking_block_error, ) raw = ( @@ -1304,34 +1305,97 @@ class TestAnthropicThinkingSignatureSelfHeal: '"message":"messages.3.content.3: Invalid `signature` in `thinking` block"},' '"request_id":"req_011Ca2EtQDxp7x6RGUY2jVn9"}' ) - assert is_anthropic_invalid_thinking_signature_error(raw) is True + assert is_anthropic_invalid_thinking_block_error(raw) is True - def test_is_anthropic_invalid_thinking_signature_error_positive_bedrock(self): + def test_is_anthropic_invalid_thinking_block_error_positive_bedrock(self): from litellm.llms.anthropic.common_utils import ( - is_anthropic_invalid_thinking_signature_error, + is_anthropic_invalid_thinking_block_error, ) # Real user-reported Bedrock scenario raw = '{"message":"messages.2.content.0.thinking.signature.str: Input should be a valid string"}' - assert is_anthropic_invalid_thinking_signature_error(raw) is True + assert is_anthropic_invalid_thinking_block_error(raw) is True - def test_is_anthropic_invalid_thinking_signature_error_positive_vertex(self): + def test_is_anthropic_invalid_thinking_block_error_positive_vertex(self): from litellm.llms.anthropic.common_utils import ( - is_anthropic_invalid_thinking_signature_error, + is_anthropic_invalid_thinking_block_error, ) raw = "messages.4.content.1.thinking.signature.str: Input should be a valid string" - assert is_anthropic_invalid_thinking_signature_error(raw) is True + assert is_anthropic_invalid_thinking_block_error(raw) is True - def test_is_anthropic_invalid_thinking_signature_error_negative(self): + def test_is_anthropic_invalid_thinking_block_error_negative(self): from litellm.llms.anthropic.common_utils import ( - is_anthropic_invalid_thinking_signature_error, + is_anthropic_invalid_thinking_block_error, ) - assert is_anthropic_invalid_thinking_signature_error("") is False - assert is_anthropic_invalid_thinking_signature_error("rate limit exceeded") is False - assert is_anthropic_invalid_thinking_signature_error("invalid_request_error: model not found") is False - assert is_anthropic_invalid_thinking_signature_error("thinking signature is malformed") is False + assert is_anthropic_invalid_thinking_block_error("") is False + assert is_anthropic_invalid_thinking_block_error("rate limit exceeded") is False + assert is_anthropic_invalid_thinking_block_error("invalid_request_error: model not found") is False + assert is_anthropic_invalid_thinking_block_error("thinking signature is malformed") is False + + def test_is_anthropic_invalid_thinking_block_error_positive_empty_thinking(self): + """LIT-6357: replayed history holding {"type": "thinking", "thinking": ""} + (produced when a non-Anthropic reasoning model's turn is bridged to the + Anthropic surface with no reasoning text) 400s with a message that names + no signature, so the pre-rename matcher missed it and the strip-and-retry + never fired. Raw string captured live on 2026-08-27.""" + from litellm.llms.anthropic.common_utils import ( + is_anthropic_invalid_thinking_block_error, + ) + + raw = ( + '{"type":"error","error":{"type":"invalid_request_error",' + '"message":"messages.1.content.0.thinking: each thinking block must contain thinking"},' + '"request_id":"req_011CeUTxhJj2rTUkK61qtbJ8"}' + ) + assert is_anthropic_invalid_thinking_block_error(raw) is True + + def test_is_empty_thinking_block(self): + from litellm.llms.anthropic.common_utils import is_empty_thinking_block + + assert is_empty_thinking_block({"type": "thinking", "thinking": ""}) is True + assert is_empty_thinking_block({"type": "thinking", "thinking": " \n\t "}) is True + assert is_empty_thinking_block({"type": "thinking", "thinking": None}) is True + assert is_empty_thinking_block({"type": "thinking"}) is True + assert is_empty_thinking_block({"type": "thinking", "thinking": "", "signature": "sig_abc"}) is True + assert is_empty_thinking_block({"type": "thinking", "thinking": "plan", "signature": "sig"}) is False + assert is_empty_thinking_block({"type": "redacted_thinking", "data": "opaque"}) is False + assert is_empty_thinking_block({"type": "text", "text": ""}) is False + assert is_empty_thinking_block("not a dict") is False + + def test_strip_empty_content_blocks_drops_empty_thinking_blocks(self): + """LIT-6357 ingestion half: an assistant tool-loop turn carrying an + empty (even signed) thinking block keeps its tool_use blocks and loses + the poison; whitespace-only counts as empty; a non-empty thinking block + and redacted_thinking are untouched.""" + from litellm.llms.anthropic.common_utils import ( + strip_empty_content_blocks_from_anthropic_messages, + ) + + tu = {"type": "tool_use", "id": "toolu_01A", "name": "get_weather", "input": {"city": "Paris"}} + msgs = [ + {"role": "user", "content": "weather?"}, + { + "role": "assistant", + "content": [{"type": "thinking", "thinking": "", "signature": "sig_abc"}, tu], + }, + { + "role": "assistant", + "content": [ + {"type": "thinking", "thinking": " \n "}, + {"type": "thinking", "thinking": "real plan", "signature": "sig"}, + {"type": "redacted_thinking", "data": "opaque"}, + ], + }, + {"role": "assistant", "content": [{"type": "thinking", "thinking": ""}]}, + ] + out = strip_empty_content_blocks_from_anthropic_messages(msgs) + assert len(out) == 3 + assert [b["type"] for b in out[1]["content"]] == ["tool_use"] + assert [b["type"] for b in out[2]["content"]] == ["thinking", "redacted_thinking"] + assert out[2]["content"][0]["thinking"] == "real plan" + assert len(msgs[1]["content"]) == 2 def test_strip_thinking_blocks_from_anthropic_messages(self): from litellm.llms.anthropic.common_utils import ( @@ -1398,14 +1462,14 @@ class TestAnthropicThinkingSignatureSelfHeal: assert "thinking" not in data assert data["messages"] == [] - def test_strip_empty_text_blocks_from_anthropic_messages(self): + def test_strip_empty_content_blocks_from_anthropic_messages(self): """Covers #22930. The core regression scenario: an assistant message with an empty text block alongside ``tool_use`` loses the empty block and keeps the ``tool_use``; a whole message that reduces to no blocks is dropped; whitespace-only text counts as empty; the caller's list is never mutated.""" from litellm.llms.anthropic.common_utils import ( - strip_empty_text_blocks_from_anthropic_messages, + strip_empty_content_blocks_from_anthropic_messages, ) tu = {"type": "tool_use", "id": "x", "name": "Bash", "input": {}} @@ -1414,14 +1478,14 @@ class TestAnthropicThinkingSignatureSelfHeal: {"role": "assistant", "content": [{"type": "text", "text": " \n "}, tu]}, {"role": "assistant", "content": [{"type": "text", "text": ""}]}, ] - out = strip_empty_text_blocks_from_anthropic_messages(msgs) + out = strip_empty_content_blocks_from_anthropic_messages(msgs) assert len(out) == 2 and out[0] is msgs[0] assert [b["type"] for b in out[1]["content"]] == ["tool_use"] assert len(msgs[1]["content"]) == 2 # caller's content unchanged def test_strip_empty_text_blocks_preserves_thinking_blocks(self): from litellm.llms.anthropic.common_utils import ( - strip_empty_text_blocks_from_anthropic_messages, + strip_empty_content_blocks_from_anthropic_messages, ) msgs = [ @@ -1433,12 +1497,12 @@ class TestAnthropicThinkingSignatureSelfHeal: ], } ] - out = strip_empty_text_blocks_from_anthropic_messages(msgs) + out = strip_empty_content_blocks_from_anthropic_messages(msgs) assert [b["type"] for b in out[0]["content"]] == ["thinking"] def test_strip_empty_text_blocks_treats_null_text_as_empty(self): from litellm.llms.anthropic.common_utils import ( - strip_empty_text_blocks_from_anthropic_messages, + strip_empty_content_blocks_from_anthropic_messages, ) msgs = [ @@ -1450,12 +1514,12 @@ class TestAnthropicThinkingSignatureSelfHeal: ], } ] - out = strip_empty_text_blocks_from_anthropic_messages(msgs) + out = strip_empty_content_blocks_from_anthropic_messages(msgs) assert [b["type"] for b in out[0]["content"]] == ["tool_result"] def test_strip_empty_text_blocks_treats_missing_text_key_as_empty(self): from litellm.llms.anthropic.common_utils import ( - strip_empty_text_blocks_from_anthropic_messages, + strip_empty_content_blocks_from_anthropic_messages, ) msgs = [ @@ -1467,21 +1531,21 @@ class TestAnthropicThinkingSignatureSelfHeal: ], } ] - out = strip_empty_text_blocks_from_anthropic_messages(msgs) + out = strip_empty_content_blocks_from_anthropic_messages(msgs) assert [b["type"] for b in out[0]["content"]] == ["tool_result"] def test_strip_empty_text_blocks_leaves_non_empty_text_alone(self): from litellm.llms.anthropic.common_utils import ( - strip_empty_text_blocks_from_anthropic_messages, + strip_empty_content_blocks_from_anthropic_messages, ) msgs = [{"role": "assistant", "content": [{"type": "text", "text": "hi"}]}] - out = strip_empty_text_blocks_from_anthropic_messages(msgs) + out = strip_empty_content_blocks_from_anthropic_messages(msgs) assert out[0] is msgs[0] # untouched messages keep identity def test_strip_empty_text_blocks_treats_non_string_text_value_as_empty(self): from litellm.llms.anthropic.common_utils import ( - strip_empty_text_blocks_from_anthropic_messages, + strip_empty_content_blocks_from_anthropic_messages, ) msgs = [ @@ -1493,7 +1557,7 @@ class TestAnthropicThinkingSignatureSelfHeal: ], } ] - out = strip_empty_text_blocks_from_anthropic_messages(msgs) + out = strip_empty_content_blocks_from_anthropic_messages(msgs) assert [b["type"] for b in out[0]["content"]] == ["tool_result"] def test_flatten_unencrypted_web_search_results_keeps_snippet_evidence(self):