Bug

When using api: "openai-responses", the gateway sends messages where content is null in the input array. Strict OpenAI-compatible providers (e.g., weelinking) reject these requests with 400, while OpenAI's own API tolerates null content.

Error

400 Invalid type for 'input[3].content': expected one of an array of objects or string, but got null instead.

Gateway log:

[responses] error provider=wlkgpt api=openai-responses model=gpt-5.5 name=Error status=400 code=bad_response_status_code type=bad_response_status_code message=400 openai_error
providerRuntimeFailureKind: schema
failoverReason: format

Configuration

{
  "wlkgpt": {
    "baseUrl": "https://api.weelinking.com/v1",
    "api": "openai-responses",
    "models": [{ "id": "gpt-5.5", "reasoning": true }]
  }
}

Environment

• openclaw version: 2026.6.2-beta.1
• Provider: weelinking (only supports openai-responses format)
• Model: gpt-5.5
• Channel: feishu

Expected Behavior

The gateway should sanitize the input array before sending: replace content: null with either an empty string "" or omit the message. This matches OpenAI's documented schema which requires content to be string | array.

Actual Behavior

Messages with content: null (likely from tool_use / tool_result entries in the conversation history) are passed through verbatim, causing strict providers to reject the entire request.

Workaround

None found. The error persists across session resets and gateway restarts.

Suggested Fix

In the openai-responses transport layer, before serializing the request payload, filter/transform messages:

input = input.map(msg => ({
  ...msg,
  content: msg.content ?? (Array.isArray(msg.tool_calls) || msg.tool_call_id ? msg.content : "")
}));

Or strip messages where content is null and the message type doesn't require it.