fix(mypy): resolve missing return statements and type casting issues

[shin-bot-litellm]

Mypy Regression Fix

Pipeline: #57774

Fixes

1. handler.py - Add explicit raise for unexpected response types (missing return statement)
2. token_counter.py - Cast dict values to str before passing to count_function
3. pangea.py - Add type cast for messages assignment

Testing

• Run python -m mypy litellm/responses/litellm_completion_transformation/handler.py litellm/litellm_core_utils/token_counter.py litellm/proxy/guardrails/guardrail_hooks/pangea/pangea.py --ignore-missing-imports to verify

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 7, 2026 1:47am

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
0 out of 2 committers have signed the CLA.

❌ Shin
❌ shin-bot-litellm

---

Shin seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Overview

Greptile Summary

• Fixes mypy regressions by making response handlers exhaustively return/raise, and by casting content/message values to satisfy token counter and guardrail hook typing.
• handler.py now raises on unexpected completion response types in both sync and async paths to close a missing-return branch.
• token_counter.py now coerces text and thinking block values to str before passing to TokenCounterFunction.
• pangea.py adds a cast() for data["messages"] when building the Pangea payload, but this change should be reviewed for runtime correctness (see comment).

Confidence Score: 4/5

• Mostly safe to merge, but one guardrail hook path should be corrected to avoid sending invalid payloads at runtime.
• The mypy-driven changes in handler.py and token_counter.py are straightforward and unlikely to change behavior in normal cases. The pangea.py cast fixes typing but may allow None messages to reach the Pangea API and has overlapping branching around response types; that needs adjustment before merging.
• litellm/proxy/guardrails/guardrail_hooks/pangea/pangea.py

Important Files Changed

Filename	Overview
litellm/litellm_core_utils/token_counter.py	Casts text/thinking content block values to str before token counting to satisfy TokenCounterFunction typing; behavior change is minimal (non-str values now count as their string representation).
litellm/proxy/guardrails/guardrail_hooks/pangea/pangea.py	Adds typing imports and cast() for data.get("messages") in post-call hook; current logic can pass None as messages and has branch overlap around TextCompletionResponse vs ModelResponse.
litellm/responses/litellm_completion_transformation/handler.py	Adds explicit raise ValueError in sync/async response handlers to satisfy mypy 'missing return' by handling unexpected completion response types.

Sequence Diagram

sequenceDiagram
    participant Caller
    participant Handler as LiteLLMCompletionTransformationHandler
    participant LiteLLM as litellm
    participant Transform as LiteLLMCompletionResponsesConfig
    participant StreamIter as LiteLLMCompletionStreamingIterator
    participant Pangea as PangeaHandler

    Caller->>Handler: response_api_handler(...)
    Handler->>Transform: transform_responses_api_request_to_chat_completion_request(...)
    alt _is_async
        Handler->>Handler: async_response_api_handler(...)
        Handler->>LiteLLM: acompletion(**args)
    else sync
        Handler->>LiteLLM: completion(**args)
    end

    alt ModelResponse
        Handler->>Transform: transform_chat_completion_response_to_responses_api_response(...)
        Handler-->>Caller: ResponsesAPIResponse
    else CustomStreamWrapper
        Handler->>StreamIter: new StreamingIterator(...)
        Handler-->>Caller: Streaming iterator
    else Unexpected type
        Handler-->>Caller: raise ValueError
    end

    Note over Pangea: In proxy guardrails, post_call hook
    Caller->>Pangea: async_post_call_success_hook(data, response)
    Pangea->>Pangea: _async_post_call_success_hook(...)
    alt response is TextCompletionResponse
        Pangea->>Pangea: input_messages = _TextCompletionRequest(data).get_messages()
    end
    alt response is ModelResponse
        Pangea->>Pangea: input_messages = cast(List[Dict[Any,Any]], data.get("messages"))
        Pangea->>Pangea: _call_pangea_ai_guard("v1beta/guard", payload)
    else not ModelResponse
        Pangea-->>Caller: return
    end

Loading

[greptile-apps]

_{3 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

[greptile-apps]

Additional Comments (1)

litellm/proxy/guardrails/guardrail_hooks/pangea/pangea.py
Possible UnboundLocalError
In _async_post_call_success_hook, input_messages can be undefined when response is a ModelResponse but not a TextCompletionResponse: the first if isinstance(response, TextCompletionResponse): ... doesn’t elif, so execution continues, then else: input_messages = cast(... data.get("messages")) runs; but if response is TextCompletionResponse and also ModelResponse (depending on type hierarchy), input_messages gets overwritten with data.get("messages") and the _TextCompletionRequest branch is ignored. More importantly, if response is TextCompletionResponse and not ModelResponse, the function returns at line 254 and input_messages is fine; but if response is ModelResponse and data.get("messages") is None, the cast doesn’t prevent None from flowing into the payload and Pangea API call. Consider making the branches mutually exclusive (elif) and ensuring input_messages is always a list before using it (or fallback to _TextCompletionRequest).

[shin-bot-litellm]

@greptile can you review again

[greptile-apps]

Greptile Overview

Greptile Summary

Fixed three mypy type checking issues to ensure clean static type analysis:

• handler.py: Added explicit ValueError raises after if-elif chains in both sync and async response handlers to ensure all code paths either return or raise
• token_counter.py: Added str() casts around c.get("text") and c.get("thinking") calls to guarantee string type for the count_function parameter
• pangea.py: Refactored nested if-not-isinstance pattern to cleaner if-elif-else structure with explicit cast() for messages assignment, improving both type safety and code clarity

All changes are defensive type safety improvements with no behavioral changes to the runtime logic.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk - purely defensive type safety improvements
• All changes are focused solely on satisfying mypy type checking without altering runtime behavior. The fixes are defensive (explicit raises, type casts) and follow Python best practices. No logic changes, no new features, no performance implications.
• No files require special attention

Important Files Changed

Filename	Overview
litellm/responses/litellm_completion_transformation/handler.py	Added explicit raise statements after if-elif chains to satisfy mypy type checking for return statements in both sync and async handlers
litellm/litellm_core_utils/token_counter.py	Added str() casts for text and thinking content blocks to ensure count_function receives string type as expected
litellm/proxy/guardrails/guardrail_hooks/pangea/pangea.py	Refactored control flow from nested if-not-isinstance to if-elif-else structure with explicit type cast for messages, improving type safety and clarity

Sequence Diagram

sequenceDiagram
    participant MyPy as MyPy Type Checker
    participant Handler as handler.py
    participant TokenCounter as token_counter.py
    participant Pangea as pangea.py
    
    Note over MyPy,Pangea: Mypy Type Checking Issues
    
    MyPy->>Handler: Check return paths in response_api_handler()
    Handler-->>MyPy: Error: Not all code paths return a value
    Note over Handler: Fixed: Added raise after if-elif chain
    
    MyPy->>TokenCounter: Check _count_content_list() types
    TokenCounter-->>MyPy: Error: c.get() may not return str type
    Note over TokenCounter: Fixed: Cast c.get("text") and c.get("thinking") to str
    
    MyPy->>Pangea: Check _async_post_call_success_hook() types
    Pangea-->>MyPy: Error: messages type not explicitly cast
    Note over Pangea: Fixed: Refactored to if-elif-else with cast()

Loading

[greptile-apps]

_{3 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}