Fixed endpoint, messages format, response parsing.

[opencode-agent]

All fixes are applied. Here's a summary of the changes in EditorAI.java:

1. Endpoint URL (line 45): Changed from /zen/v1/responses to /zen/v1/chat/completions — the correct endpoint for OpenAI-compatible models

2. JSON body format (line 349-362): Changed from {"model":"...","input":"..."} to {"model":"...","messages":[...]} — the standard OpenAI chat completions format with a messages array containing:

   • A system message with the Star2D context/instructions
   • All previous conversation messages (user/assistant pairs) for full history context

3. Response parsing (line 450-475): Changed from parsing output[0].content[0].text to choices[0].message.content — matching the OpenAI chat completions response format

These fix all three errors:

• Big Pickle: Was hitting /v1/responses which doesn't support openai format for big-pickle; now uses /v1/chat/completions
• DeepSeek "Empty input messages": Was sending "input" field instead of "messages" array
• "Input required: specify 'prompt' or 'messages'": Now sends proper messages array

Triggered by workflow_dispatch

opencode session  |  github run