hermes - 💡(How to fix) Fix fix(run_agent): non-vision models get JSON error instead of text summary for computer_use captures [1 pull requests]

Description: When a non-vision model (DeepSeek, Qwen, etc.) calls computer_use with action='capture', the method _tool_result_content_for_active_model in run_agent.py returns a JSON object with an error field instead of silently falling back to the text summary (AX tree).

{
  "error": "computer_use returned screenshot/image content, but the active model/provider does not support image input. Switch to a vision-capable model...",
  "text_summary": "..."
}

The model sees the error message and gets confused — it complains about not being vision-capable instead of using the text summary it already received.

Expected behaviour: Non-vision models should receive the text summary (AX tree) silently, without the error block. The logger.warning that follows is sufficient for debugging.

Reproduction:

1. Use a non-vision model (e.g. deepseek-v4-flash)
2. Call computer_use(action='capture')
3. Model receives JSON with error field and complains

Local fix: Removed the if tool_name == "computer_use": return json.dumps({error: ...}) block from _tool_result_content_for_active_model (~line 3892 in run_agent.py). The remaining code path returns the text summary directly.

v0.15.1 status: Still reproducible. The block was removed in a previous version but has been re-introduced.