fix: handle OpenAI content array format in messagesToPrompt

[darrenwadley-ui]

Summary

• Fix messagesToPrompt to handle the case where msg.content is an array of content parts instead of a plain string
• Add extractContent() helper that normalizes both formats to a string
• Replace all three msg.content references in messagesToPrompt with extractContent(msg.content)

Bug

The OpenAI chat completions API allows message content to be either a string or an array of content part objects:

{
  "role": "user",
  "content": [{"type": "text", "text": "Hello, how are you?"}]
}

Clients like OpenClaw always send content in the array format. When the proxy receives this, JavaScript's template literal coerces the array to a string, producing [object Object] instead of the actual text. This means Claude CLI receives garbage input like:

[object Object]

instead of:

Hello, how are you?

How to reproduce

Send a request to the proxy with content as an array:

curl -X POST http://localhost:3000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-opus-4",
    "messages": [
      {"role": "user", "content": [{"type": "text", "text": "Hello!"}]}
    ]
  }'

Before this fix, Claude CLI would receive [object Object] as the prompt.

Fix

Added an extractContent() helper function that:

1. Returns the string as-is if content is already a string
2. Extracts and joins .text fields from content part arrays
3. Falls back to String(content ?? "") for any other shape

All three uses of msg.content in messagesToPrompt (system, user, assistant cases) now go through this helper.

Test plan

• Send a request with content as a plain string — should work as before
• Send a request with content as [{type: "text", text: "..."}] — should now correctly extract the text
• Send a request with multiple content parts — should join them with newlines

🤖 Generated with Claude Code