Merge 297a65d312 into e768ad55ce

.github/copilot-instructions.md

@@ -0,0 +1,48 @@
+# Copilot instructions for LiteLLM
+
+This repo has detailed contributor docs already; read them rather than duplicating:
+
+- `CLAUDE.md` / `AGENTS.md` (identical content) - coding conventions, lint/type-suppression rules, PR/commit conventions. **Read this in full before writing code.**
+- `ARCHITECTURE.md` - request flow diagrams for the SDK and the proxy (AI Gateway), the translation-layer pattern, and the data-access layer (`litellm/models/` + `litellm/repositories/`).
+- `CONTRIBUTING.md` - setup, testing, linting walkthrough.
+- `tests/test_litellm/readme.md` - unit test directory conventions.
+- `tests/e2e/CLAUDE.md` and `tests/e2e/CONTRIBUTING.md` - e2e harness rules (shared transport only, no raw `requests`, suite-per-folder layout).
+
+## Big picture
+
+LiteLLM is two things sharing one codebase:
+
+1. **SDK** (`litellm/`): `main.py` (`completion()`/`acompletion()`) resolves provider via `utils.py:get_llm_provider()`, then `llms/custom_httpx/llm_http_handler.py` (`BaseLLMHTTPHandler`) calls a provider's `Config.transform_request()` / `transform_response()` (in `llms/{provider}/chat/transformation.py`, subclassing `BaseConfig` from `llms/base_llm/chat/transformation.py`). You almost never modify the handler itself, just the provider's `transformation.py`.
+2. **AI Gateway / Proxy** (`litellm/proxy/`): wraps the SDK with auth (`proxy/auth/user_api_key_auth.py`), rate limiting/budgets (`proxy/hooks/`), and routing (`router.py`), then calls into the SDK's `main.py` for the actual LLM call. Cost is calculated post-call in `litellm_logging.py` -> `cost_calculator.py` and queued to Postgres via `proxy/db/db_spend_update_writer.py`.
+
+Persistence/caching: Redis (`caching/redis_cache.py`, `caching/dual_cache.py`) for rate limits/API-key cache/cooldowns; Postgres (`proxy/schema.prisma`) for keys/teams/users/spend logs, accessed through `litellm/repositories/` (`BaseRepository[T]` + entity repos) with entity Pydantic models in `litellm/models/` (re-exported from `proxy/_types.py` for backwards compat).
+
+To add/modify a provider or feature, find the relevant `transformation.py` from the table in `ARCHITECTURE.md` section 3, and add unit tests in `tests/llm_translation/test_{provider}.py` that call `transform_request`/`transform_response` directly (no live API calls needed).
+
+## Build / install
+
+- Package manager is `uv`, not pip/poetry directly.
+- `make install-dev` - base dev deps. `make install-proxy-dev` - adds proxy extras. `make install-test-deps` - full env incl. Postgres/Prisma, needed for proxy or DB-backed tests.
+- `make bootstrap` - full fresh-clone/worktree setup (deps, Prisma client, UI npm install).
+
+## Tests
+
+- Unit tests: `tests/test_litellm/` mirrors `litellm/`'s structure 1:1 (e.g. `litellm/utils.py` -> `tests/test_litellm/test_utils.py`); **mocked only, no real API calls**.
+- Run one file: `uv run pytest tests/test_litellm/test_your_file.py -v`. Run one test: append `::test_name`.
+- Run everything: `make test-unit` (parallelized, `tests/test_litellm`). Other `make test-unit-*` targets split proxy/llms/integration tests into matrix groups mirroring CI (see `make help`).
+- Integration/live-API tests live under `tests/llm_translation/`, `tests/e2e/` (real proxy + real providers, see e2e CLAUDE.md/CONTRIBUTING.md), etc. — not mocked.
+- e2e tests must use the shared transport (`e2e_http.py`), never `requests.*` directly (enforced by CI check).
+
+## Lint / type-check
+
+- `make format` - ruff format (line length 120, not 88). `make lint` - full CI-parity lint (ruff, basedpyright budget gates, circular imports, import safety). `make lint-dev` - faster, changed-files-only version for local iteration.
+- basedpyright, ruff-strict, and type-discipline rules are gated by ratcheting budget files (`basedpyright-code-budget.json`, `ruff-strict-budget.json`, `type-discipline-budget.json`). If your change fixes violations, run `make lint-budget-update` and commit the lowered budgets.
+- `make pre-commit` runs the CI-equivalent checks on staged files and writes full output to a log file in `.git` (path printed first/last line) — read that log instead of re-running.
+- Every lint/type suppression must cite the exact rule and a reason, e.g. `# pyright: ignore[reportArgumentType]  # ...`. `# type: ignore` is banned (disabled in `pyrightconfig.json`).
+
+## Key conventions (see CLAUDE.md/AGENTS.md for full list)
+
+- No mutation: don't reassign variables; annotate with `: Final`; build collections in one shot (comprehensions -> `tuple()`/`MappingProxyType()`/`frozenset()`) instead of seeding-and-mutating (`LIT001`/`LIT002`).
+- Fully typed; no bare `Any`/`dict`. Validate untyped inputs via Pydantic in the caller rather than loosening types.
+- Composition over inheritance; early returns over nesting; model failures as values rather than raising where a public contract exists.
+- Conventional Commits for commits/PR titles; branch off `litellm_internal_staging` (not `main`), branch names use `litellm_` prefix with no `/`.

litellm/llms/anthropic/chat/transformation.py

@@ -1690,6 +1690,19 @@ class AnthropicConfig(AnthropicModelInfo, BaseConfig):
         """
         return False
 
+    def should_strip_claude_code_identity(self) -> bool:
+        """
+        Whether to drop Claude Code's self-identification from the system prompt.
+
+        Distinct from ``should_strip_billing_metadata``: that drops the billing header on
+        every provider whose request shape rejects it, including Bedrock/Vertex/Azure, which
+        still serve *Claude* models. This only trips on providers whose
+        Anthropic-compatible endpoint serves a *different* model, where "You are Claude
+        Code" is a false self-description. The first-party config and the Claude-serving
+        providers keep it, so this stays False here.
+        """
+        return False
+
     def translate_system_message(self, messages: list[AllMessageValues]) -> list[AnthropicSystemMessageContent]:
         """
         Translate system message to anthropic format.

litellm/llms/anthropic/common_utils.py

@@ -85,6 +85,9 @@ _CLAUDE_CODE_OBJECT_LIST_ADAPTER: Final = TypeAdapter(list[object])
 _CLAUDE_CODE_USER_AGENT_PREFIXES: Final = ("claude-cli/", "claude-code/")
 
 
+_CLAUDE_CODE_IDENTITY: Final = "You are Claude Code, Anthropic's official CLI for Claude."
+
+
 def requires_native_compaction_beta(
     custom_llm_provider: str,
     optional_params: Mapping[str, object],
@@ -127,6 +130,65 @@ def is_claude_code_user_agent(user_agent: str) -> bool:
     return user_agent.startswith(_CLAUDE_CODE_USER_AGENT_PREFIXES)
 
 
+def strip_claude_code_identity(text: str) -> str | None:
+    """Remove Claude Code's self-identification sentence from a single system text block.
+
+    Claude Code prefixes its system prompt with ``You are Claude Code, Anthropic's
+    official CLI for Claude.``. When that prompt is forwarded to a non-Claude model
+    (e.g. a DeepSeek-compatible or OpenAI-like Anthropic-compatible endpoint), the
+    sentence is a false self-description and should be dropped before sending
+    upstream. Returns ``None`` when the text was nothing but the identity sentence
+    (so callers can drop the whole block), otherwise the text with the sentence and
+    its framing whitespace removed.
+    """
+    stripped: Final = text.strip()
+    if stripped == _CLAUDE_CODE_IDENTITY:
+        return None
+    if stripped.startswith(_CLAUDE_CODE_IDENTITY):
+        return text.replace(_CLAUDE_CODE_IDENTITY, "", 1).lstrip("\n")
+    return text
+
+
+def strip_claude_code_identity_from_system(
+    system_param: str | list[object] | None,
+) -> str | list[object] | None:
+    """Strip Claude Code's self-identification sentence from a full system parameter.
+
+    Unlike :func:`strip_claude_code_identity`, which operates on a single text
+    block, this handles the whole ``system`` value of an Anthropic Messages
+    request -- either a plain string or a list of system content blocks.
+    Non-text blocks are kept as-is; text blocks have the identity sentence
+    removed, and are dropped entirely when it was the only content.
+
+    Returns ``None`` when every block was dropped so callers can remove the whole
+    ``system`` parameter.
+    """
+    if isinstance(system_param, str):
+        return strip_claude_code_identity(system_param)
+    if isinstance(system_param, list):
+        filtered_list: Final = []  # mutable-ok: API message payload
+        for content_block in system_param:
+            if isinstance(content_block, dict):
+                text = content_block.get("text", "")
+                content_type = content_block.get("type", "")
+                if content_type != "text":
+                    filtered_list.append(content_block)
+                    continue
+                stripped_text = strip_claude_code_identity(text)
+                if stripped_text is None:
+                    continue
+                if stripped_text == text:
+                    filtered_list.append(content_block)
+                else:
+                    rewritten = {**content_block, "text": stripped_text}  # mutable-ok: API message payload
+                    filtered_list.append(rewritten)
+            else:
+                # Keep non-dict items as-is.
+                filtered_list.append(content_block)
+        return filtered_list if len(filtered_list) > 0 else None
+    return system_param
+
+
 def _validated_claude_code_mapping(value: object) -> dict[object, object] | None:
     try:
         return _CLAUDE_CODE_OBJECT_MAPPING_ADAPTER.validate_python(value)

litellm/llms/anthropic/pass_through/adapters/transformation.py

@@ -8,6 +8,7 @@ from typing import TYPE_CHECKING, Any, Final, Literal, TypeAlias, TypeVar, cast
 from pydantic import JsonValue, TypeAdapter
 
 import litellm
+from litellm.llms.anthropic.common_utils import strip_claude_code_identity_from_system
 from litellm.llms.anthropic.pass_through.utils import (
     is_reasoning_auto_summary_enabled,
     prompt_cache_key_from_user_id,
@@ -1230,6 +1231,23 @@ class LiteLLMAnthropicMessagesAdapter:
         new_messages: list[AllMessageValues] = []
         tool_name_mapping: dict[str, str] = {}
 
+        # Strip Claude Code's self-identification from the system prompt before
+        # translating to an OpenAI chat-completions request. This bridge only runs
+        # for non-Anthropic models (first-party servers take the native
+        # AnthropicMessagesConfig path), so "You are Claude Code" is always a
+        # false self-description here and must not reach the target model.
+        system_param: Final = anthropic_message_request.get("system")
+        if system_param is not None:
+            stripped_system = strip_claude_code_identity_from_system(system_param)
+            if stripped_system is None or stripped_system != system_param:
+                # This adapter is also invoked with a read-only wire-body mapping
+                # (shadow evaluation); mutate a copy, never the caller's request.
+                anthropic_message_request = cast(AnthropicMessagesRequest, dict(anthropic_message_request))
+                if stripped_system is None:
+                    anthropic_message_request.pop("system", None)
+                else:
+                    anthropic_message_request["system"] = stripped_system
+
         ## CONVERT ANTHROPIC MESSAGES TO OPENAI
         messages_list: Final[list[AllAnthropicPassThroughMessageValues]] = cast(
             list[AllAnthropicPassThroughMessageValues],

litellm/llms/anthropic/pass_through/messages/transformation.py

@@ -28,6 +28,7 @@ from ...common_utils import (
     optionally_handle_anthropic_oauth,
     requires_native_compaction_beta,
     strip_advisor_blocks_from_messages,
+    strip_claude_code_identity_from_system,
     strip_encrypted_reasoning_blocks_from_anthropic_messages,
 )
 from .mid_conversation_system import (
@@ -130,6 +131,32 @@ class AnthropicMessagesConfig(BaseAnthropicMessagesConfig):
         """
         return False
 
+    def should_strip_claude_code_identity(self) -> bool:
+        """
+        Whether to drop Claude Code's self-identification from the system prompt.
+
+        Distinct from ``should_strip_billing_metadata``: that drops the billing header on
+        every provider whose request shape rejects it, including Bedrock/Vertex/Azure, which
+        still serve *Claude* models. This only trips on providers whose
+        Anthropic-compatible endpoint serves a *different* model (DeepSeek, MiniMax,
+        Tencent, OpenAI-like passthrough), where "You are Claude Code" is a false
+        self-description. The first-party config keeps it.
+        """
+        return False
+
+    @staticmethod
+    def _strip_claude_code_identity_from_system(system_param: str | list[object] | None) -> str | list[object] | None:
+        """
+        Strip Claude Code's self-identification sentence from system parameter.
+
+        Args:
+            system_param: Can be a string or a list of system message content blocks
+
+        Returns:
+            System parameter with the identity sentence removed, or None if all content was removed
+        """
+        return strip_claude_code_identity_from_system(system_param)
+
     @staticmethod
     def _filter_billing_headers_from_system(system_param):
         """
@@ -537,6 +564,14 @@ class AnthropicMessagesConfig(BaseAnthropicMessagesConfig):
             else:
                 anthropic_messages_optional_request_params.pop("system", None)
 
+        stripped_identity_system: Final = anthropic_messages_optional_request_params.get("system")
+        if self.should_strip_claude_code_identity() and stripped_identity_system is not None:
+            identity_filtered: Final = self._strip_claude_code_identity_from_system(stripped_identity_system)
+            if identity_filtered is not None and len(identity_filtered) > 0:
+                anthropic_messages_optional_request_params["system"] = identity_filtered
+            else:
+                anthropic_messages_optional_request_params.pop("system", None)
+
         # Transform context_management from OpenAI format to Anthropic format if needed
         context_management_param: Final = anthropic_messages_optional_request_params.get("context_management")
         if context_management_param is not None:

litellm/llms/deepseek/messages/transformation.py

@@ -29,6 +29,9 @@ class DeepSeekAnthropicMessagesConfig(AnthropicMessagesConfig):
     def should_strip_billing_metadata(self) -> bool:
         return True
 
+    def should_strip_claude_code_identity(self) -> bool:
+        return True
+
     @staticmethod
     def get_api_key(api_key: str | None = None) -> str | None:
         return api_key or get_secret_str("DEEPSEEK_API_KEY") or litellm.api_key

litellm/llms/minimax/messages/transformation.py

@@ -31,6 +31,9 @@ class MinimaxMessagesConfig(AnthropicMessagesConfig):
     def should_strip_billing_metadata(self) -> bool:
         return True
 
+    def should_strip_claude_code_identity(self) -> bool:
+        return True
+
     @staticmethod
     def get_api_key(api_key: str | None = None) -> str | None:
         """

litellm/llms/openai_like/messages/transformation.py

@@ -32,6 +32,9 @@ class OpenAILikeAnthropicMessagesConfig(AnthropicMessagesConfig):
         super().__init__()
         self._cache_control_ttl: Final = cache_control_ttl
 
+    def should_strip_claude_code_identity(self) -> bool:
+        return True
+
     def validate_anthropic_messages_environment(
         self,
         headers: dict[str, str],

litellm/llms/tencent/messages/transformation.py

@@ -30,6 +30,9 @@ class TencentAnthropicMessagesConfig(AnthropicMessagesConfig):
     def should_strip_billing_metadata(self) -> bool:
         return True
 
+    def should_strip_claude_code_identity(self) -> bool:
+        return True
+
     @staticmethod
     def get_api_key(api_key: str | None = None) -> str | None:
         return api_key or get_secret_str("TENCENT_API_KEY") or litellm.api_key

tests/unit/llms/anthropic/chat/test_anthropic_chat_transformation.py

@@ -5832,6 +5832,118 @@ def test_should_strip_billing_metadata_by_provider(module_path, class_name, expe
     assert config_cls().should_strip_billing_metadata() is expected_strip
 
 
+@pytest.mark.parametrize(
+    "module_path, class_name, expected_strip",
+    [
+        # First-party and Claude-serving providers keep the identity sentence: the
+        # model genuinely *is* Claude, even via Bedrock/Vertex/Azure.
+        ("litellm.llms.anthropic.chat.transformation", "AnthropicConfig", False),
+        (
+            "litellm.llms.anthropic.pass_through.messages.transformation",
+            "AnthropicMessagesConfig",
+            False,
+        ),
+        (
+            "litellm.llms.bedrock.claude_platform.transformation",
+            "BedrockClaudePlatformConfig",
+            False,
+        ),
+        (
+            "litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.transformation",
+            "VertexAIAnthropicConfig",
+            False,
+        ),
+        ("litellm.llms.azure_ai.anthropic.transformation", "AzureAnthropicConfig", False),
+        # Non-Claude Anthropic-compatible endpoints strip the false self-description.
+        ("litellm.llms.minimax.messages.transformation", "MinimaxMessagesConfig", True),
+        (
+            "litellm.llms.deepseek.messages.transformation",
+            "DeepSeekAnthropicMessagesConfig",
+            True,
+        ),
+        (
+            "litellm.llms.tencent.messages.transformation",
+            "TencentAnthropicMessagesConfig",
+            True,
+        ),
+    ],
+)
+def test_should_strip_claude_code_identity_by_provider(module_path, class_name, expected_strip):
+    import importlib
+
+    config_cls = getattr(importlib.import_module(module_path), class_name)
+    assert config_cls().should_strip_claude_code_identity() is expected_strip
+
+
+def _system_with_identity_block(identity_sentence: str) -> list:
+    return [
+        {
+            "role": "system",
+            "content": [
+                # Claude Code sends the identity sentence as its own text block.
+                {"type": "text", "text": identity_sentence},
+                {"type": "text", "text": "real system prompt"},
+            ],
+        }
+    ]
+
+
+def test_messages_request_strips_claude_code_identity_for_minimax():
+    from litellm.llms.minimax.messages.transformation import MinimaxMessagesConfig
+    from litellm.types.router import GenericLiteLLMParams
+
+    config = MinimaxMessagesConfig()
+    assert config.should_strip_claude_code_identity() is True
+
+    optional_params = {
+        "max_tokens": 16,
+        "system": [
+            {"type": "text", "text": "You are Claude Code, Anthropic's official CLI for Claude."},
+            {"type": "text", "text": "real system prompt"},
+        ],
+    }
+    result = config.transform_anthropic_messages_request(
+        model="MiniMax-M2",
+        messages=[{"role": "user", "content": "hi"}],
+        anthropic_messages_optional_request_params=optional_params,
+        litellm_params=GenericLiteLLMParams(),
+        headers={},
+    )
+
+    texts = [block["text"] for block in result.get("system", [])]
+    assert "You are Claude Code, Anthropic's official CLI for Claude." not in texts
+    assert "real system prompt" in texts
+
+
+def test_messages_request_keeps_claude_code_identity_for_first_party():
+    from litellm.llms.anthropic.pass_through.messages.transformation import (
+        AnthropicMessagesConfig,
+    )
+    from litellm.types.router import GenericLiteLLMParams
+
+    config = AnthropicMessagesConfig()
+    assert config.should_strip_claude_code_identity() is False
+
+    optional_params = {
+        "max_tokens": 16,
+        "system": [
+            {"type": "text", "text": "You are Claude Code, Anthropic's official CLI for Claude."},
+            {"type": "text", "text": "real system prompt"},
+        ],
+    }
+    result = config.transform_anthropic_messages_request(
+        model="claude-3-5-sonnet-latest",
+        messages=[{"role": "user", "content": "hi"}],
+        anthropic_messages_optional_request_params=optional_params,
+        litellm_params=GenericLiteLLMParams(),
+        headers={},
+    )
+
+    texts = [block["text"] for block in result.get("system", [])]
+    assert "You are Claude Code, Anthropic's official CLI for Claude." in texts
+    assert "real system prompt" in texts
+
+
 def test_namespace_tool_flat_nested_tools_are_extracted():
     """Codex sends nested tools in flat format {type, name, description, parameters} with no 'function' wrapper.
     These must be normalized and mapped without raising KeyError: 'function'."""
@@ -6379,6 +6491,29 @@ def test_eager_input_streaming_reaches_anthropic_request_tools():
     assert result["tools"][0]["name"] == "write_file"
 
 
+def test_messages_request_strips_identity_only_system_for_minimax():
+    """Identity-only system drops the whole system param (pop path)."""
+    from litellm.llms.minimax.messages.transformation import MinimaxMessagesConfig
+    from litellm.types.router import GenericLiteLLMParams
+
+    config = MinimaxMessagesConfig()
+    optional_params = {
+        "max_tokens": 16,
+        "system": [
+            {"type": "text", "text": "You are Claude Code, Anthropic's official CLI for Claude."},
+        ],
+    }
+    result = config.transform_anthropic_messages_request(
+        model="MiniMax-M2",
+        messages=[{"role": "user", "content": "hi"}],
+        anthropic_messages_optional_request_params=optional_params,
+        litellm_params=GenericLiteLLMParams(),
+        headers={},
+    )
+
+    assert "system" not in result
+
+
 # ---------------------------------------------------------------------------
 # Mid-conversation ``role: "system"`` on the chat completions path.
 #

tests/unit/llms/anthropic/pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py

@@ -2862,6 +2862,68 @@ def test_translate_completion_input_params_keeps_provider_native_tools():
     assert translated["tools"] == [{"googleMaps": {}}]
 
 
+CLAUDE_CODE_IDENTITY = "You are Claude Code, Anthropic's official CLI for Claude."
+
+
+@pytest.mark.parametrize(
+    "system,expected_system_content",
+    [
+        # Identity sentence alone -> the whole system message is dropped. The bridge
+        # only serves non-Anthropic models, so the first-party case is never here.
+        (CLAUDE_CODE_IDENTITY, None),
+        # Identity followed by the real prompt -> the real prompt survives.
+        (f"{CLAUDE_CODE_IDENTITY}\nYou are an interactive agent.", "You are an interactive agent."),
+        # Non-identity system prompt is passed through untouched.
+        ("You are a helpful assistant.", "You are a helpful assistant."),
+    ],
+)
+def test_translate_anthropic_to_openai_strips_claude_code_identity(system, expected_system_content):
+    """The chat-completions bridge must not forward Claude Code's self-identification upstream."""
+    from litellm.types.llms.anthropic import AnthropicMessagesRequest
+
+    adapter = LiteLLMAnthropicMessagesAdapter()
+    openai_request, _ = adapter.translate_anthropic_to_openai(
+        anthropic_message_request=AnthropicMessagesRequest(
+            model="hosted_vllm/kimi-k3",
+            max_tokens=1024,
+            messages=[{"role": "user", "content": "hi"}],
+            system=system,
+        ),
+        custom_llm_provider="hosted_vllm",
+    )
+
+    system_messages = [m for m in openai_request["messages"] if m["role"] == "system"]
+    if expected_system_content is None:
+        assert system_messages == []
+    else:
+        assert len(system_messages) == 1
+        assert system_messages[0]["content"] == expected_system_content
+
+
+def test_translate_anthropic_to_openai_strips_claude_code_identity_read_only_request():
+    """Identity stripping must not mutate a read-only request (shadow evaluation passes a
+    MappingProxyType, so the adapter must copy before rewriting ``system``)."""
+    from types import MappingProxyType
+
+    adapter = LiteLLMAnthropicMessagesAdapter()
+    request = MappingProxyType(
+        {
+            "model": "hosted_vllm/kimi-k3",
+            "max_tokens": 1024,
+            "messages": [{"role": "user", "content": "hi"}],
+            "system": f"{CLAUDE_CODE_IDENTITY}\nYou are an interactive agent.",
+        }
+    )
+
+    openai_request, _ = adapter.translate_anthropic_to_openai(
+        anthropic_message_request=request,
+        custom_llm_provider="hosted_vllm",
+    )
+
+    system_messages = [m for m in openai_request["messages"] if m["role"] == "system"]
+    assert [m["content"] for m in system_messages] == ["You are an interactive agent."]
+
+
 def test_translate_openai_content_to_anthropic_reasoning_content_without_thinking_blocks():
     """
     Test that reasoning_content is converted to thinking block when thinking_blocks is not present.

tests/unit/llms/anthropic/test_anthropic_common_utils.py

@@ -54,6 +54,85 @@ def test_is_claude_code_one_shot_subagent_request(messages, system, expected):
     )
 
 
+@pytest.mark.parametrize(
+    "text,expected",
+    [
+        # Identity sentence alone -> drop the whole block.
+        ("You are Claude Code, Anthropic's official CLI for Claude.", None),
+        # Identity sentence followed by the rest of the system prompt -> keep the rest.
+        (
+            "You are Claude Code, Anthropic's official CLI for Claude.\nYou are an interactive agent.",
+            "You are an interactive agent.",
+        ),
+        # Leading/trailing whitespace around the bare sentence is treated as drop-only.
+        ("  You are Claude Code, Anthropic's official CLI for Claude.  ", None),
+        # Non-Claude-Code prompts are untouched.
+        ("You are a helpful assistant.", "You are a helpful assistant."),
+        # A sentence that merely mentions Claude Code is untouched.
+        ("You are Claude Code's helper.", "You are Claude Code's helper."),
+    ],
+)
+def test_strip_claude_code_identity(text, expected):
+    from litellm.llms.anthropic.common_utils import strip_claude_code_identity
+
+    assert strip_claude_code_identity(text) == expected
+
+
+CLAUDE_CODE_IDENTITY = "You are Claude Code, Anthropic's official CLI for Claude."
+
+
+@pytest.mark.parametrize(
+    "system_param,expected",
+    [
+        # String form: identity sentence alone -> drop the whole system param.
+        (CLAUDE_CODE_IDENTITY, None),
+        # String form: identity followed by the real prompt -> keep the real prompt.
+        (f"{CLAUDE_CODE_IDENTITY}\nYou are an interactive agent.", "You are an interactive agent."),
+        # String form: non-identity text is untouched.
+        ("You are a helpful assistant.", "You are a helpful assistant."),
+        # List form: identity text block is dropped, non-text blocks and the
+        # surviving text block are preserved.
+        (
+            [
+                {"type": "text", "text": CLAUDE_CODE_IDENTITY},
+                {"type": "text", "text": "real system prompt"},
+            ],
+            [{"type": "text", "text": "real system prompt"}],
+        ),
+        # List form: identity-only text means every block is dropped.
+        ([{"type": "text", "text": CLAUDE_CODE_IDENTITY}], None),
+        # List form: non-text blocks survive identity stripping untouched.
+        (
+            [
+                {"type": "text", "text": CLAUDE_CODE_IDENTITY},
+                {"type": "text", "text": "real system prompt", "cache_control": {"type": "ephemeral"}},
+            ],
+            [{"type": "text", "text": "real system prompt", "cache_control": {"type": "ephemeral"}}],
+        ),
+        # List form: identity + real content in the same block -> rewritten block.
+        (
+            [{"type": "text", "text": f"{CLAUDE_CODE_IDENTITY}\nreal system prompt"}],
+            [{"type": "text", "text": "real system prompt"}],
+        ),
+        # List form: a non-text block (type "other") is kept as-is.
+        (
+            [{"type": "text", "text": CLAUDE_CODE_IDENTITY}, {"type": "other", "foo": "bar"}],
+            [{"type": "other", "foo": "bar"}],
+        ),
+        # List form: non-dict items are kept as-is.
+        ([{"type": "text", "text": CLAUDE_CODE_IDENTITY}, "loose item"], ["loose item"]),
+        # Non-string, non-list system payload is passed through untouched.
+        (None, None),
+    ],
+)
+def test_strip_claude_code_identity_from_system(system_param, expected):
+    from litellm.llms.anthropic.common_utils import (
+        strip_claude_code_identity_from_system,
+    )
+
+    assert strip_claude_code_identity_from_system(system_param) == expected
+
+
 class TestOptionallyHandleAnthropicOAuth:
     """Tests for optionally_handle_anthropic_oauth function."""