# [Bug] `/v1/responses` endpoint forwards unsupported `user` parameter to upstream, causing 400 error

[Polaris-F]

[Bug] /v1/responses endpoint forwards unsupported user parameter to upstream, causing 400 error

Description

When clients (e.g., LobeHub v2) send requests to the /v1/responses endpoint with a user parameter, Sub2API forwards this parameter to the upstream OpenAI API without filtering it out. The upstream API rejects the request with a 400 Bad Request error: Unsupported parameter: user.

This issue affects models like gpt-5.4, gpt-5.3-codex, etc., when accessed through the Responses API endpoint.

Steps to Reproduce

1. Send a POST request to /v1/responses (or /responses) with a body that includes the user field:

{
  "model": "gpt-5.4",
  "stream": true,
  "input": [
    {"role": "user", "content": "hello"}
  ],
  "user": "user_123"
}

2. Sub2API forwards the request to the upstream OpenAI API, including the user parameter.

3. The upstream API returns: 400 Bad Request - Unsupported parameter: user

Error Log

2026-03-24T19:44:46.925+0800  WARN  handler/openai_gateway_handler.go:337  openai.forward_failed
{
  "component": "handler.openai_gateway.responses",
  "model": "gpt-5.4",
  "stream": true,
  "error": "upstream error: 400 message=Unsupported parameter: user"
}

Expected Behavior

The user parameter should be stripped from the request body before forwarding to the upstream API, similar to how other unsupported parameters (prompt_cache_retention, safety_identifier) are already handled.

Suggested Fix

In internal/service/openai_gateway_service.go, add "user" to the unsupportedFields list:

// Current code (line ~1914):
unsupportedFields := []string{"prompt_cache_retention", "safety_identifier"}

// Suggested fix:
unsupportedFields := []string{"prompt_cache_retention", "safety_identifier", "user"}

Environment

• Sub2API version: 0.1.104 (commit: 4f7629a, built: 2026-03-20)
• Client: LobeHub v2.1.44
• Upstream: OpenAI API (OAuth accounts)
• Affected endpoint: POST /v1/responses and POST /responses
• Affected models: All models accessed via Responses API (gpt-5.4, gpt-5.3-codex, etc.)

Additional Context

• The /v1/chat/completions endpoint is NOT affected — only /v1/responses.
• LobeHub v2 sends the user parameter for user tracking purposes. The upstream OpenAI Responses API does not support this parameter.
• Similar issues have been fixed before: max_output_tokens was handled in a similar way (ref: /v1/responses 不兼容 max_output_tokens，导致 OpenClaw + gpt-5.4 请求返回 400 #831, PR fix(openai): normalize OAuth passthrough /responses for OpenClaw clients #1107).