fix(proxy): forward extra headers in chat

[nehaaprasad]

Type: #20373

Changes

• Proxy extra_headers were passed into the chat completions - responses bridge but never added to the request, so they were not sent upstream (e.g. Cloudflare auth).

• The bridge now sets request_data["extra_headers"] = headers in transform_request() so those headers are forwarded to litellm.responses().

• Added a test that transform_request includes the given headers as extra_headers in the returned request data.

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 4, 2026 3:31am

[greptile-apps]

Greptile Overview

Greptile Summary

This PR fixes a bug where proxy extra_headers (like Cloudflare authentication tokens) were not being forwarded to upstream providers when using the chat completions to Responses API bridge.

Changes:

• Added request_data["extra_headers"] = headers in transform_request() to forward headers to litellm.responses()
• Added test to verify headers are properly included in the transformed request data

Impact:
The fix is minimal (3 lines) and directly addresses the issue. Headers are now correctly passed through the transformation bridge to the upstream API call.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The change is straightforward, well-tested, and directly fixes a documented bug. The 3-line addition properly forwards headers that were previously being ignored. Test coverage validates the fix.
• No files require special attention

Important Files Changed

Filename	Overview
litellm/completion_extras/litellm_responses_transformation/transformation.py	Added 3 lines to forward headers as extra_headers in transform_request(). Simple, safe change that fixes header forwarding bug.
tests/test_litellm/completion_extras/test_litellm_responses_transformation_transformation.py	Added test to verify headers are included in extra_headers field. Test is simple, focused, and validates the fix correctly.

Sequence Diagram

sequenceDiagram
    participant Client
    participant Proxy
    participant Handler as ResponsesToCompletionBridgeHandler
    participant Transform as LiteLLMResponsesTransformationHandler
    participant ResponsesAPI as litellm.responses()
    participant Upstream as LLM Provider

    Client->>Proxy: POST /chat/completions<br/>(headers: cf-aig-authorization, etc.)
    Proxy->>Proxy: Extract headers from request
    Proxy->>Handler: completion(messages, headers, ...)
    Handler->>Transform: transform_request(messages, headers, ...)
    Transform->>Transform: Convert messages to Responses API format
    Transform->>Transform: Build request_data dict
    Note over Transform: NEW: request_data["extra_headers"] = headers
    Transform-->>Handler: return request_data
    Handler->>ResponsesAPI: responses(**request_data)
    Note over ResponsesAPI: extra_headers unpacked from request_data
    ResponsesAPI->>Upstream: HTTP request with extra_headers
    Note over Upstream: Headers like cf-aig-authorization<br/>now forwarded correctly
    Upstream-->>ResponsesAPI: Response
    ResponsesAPI-->>Handler: ResponsesAPIResponse
    Handler->>Handler: Transform response to chat completion format
    Handler-->>Proxy: ModelResponse
    Proxy-->>Client: Chat completion response

Loading

[greptile-apps]

_{2 files reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[Sameerlite]

LGTM