Summary

• Add --model option to ask and send commands for selecting ChatGPT model/mode before sending
• Add new opencli chatgpt model <model> standalone command for switching models
• Fix ask command returning partial/thinking intermediate text instead of waiting for the full response

Supported models

auto, instant, thinking, 5.2-instant, 5.2-thinking

Usage

# Switch model standalone
opencli chatgpt model thinking

# Ask with model selection
opencli chatgpt ask "hello" --model thinking --timeout 60

# Send with model selection
opencli chatgpt send "hello" --model instant

Implementation details

Model selection (ax.ts)

• Clicks the "Options" button (stable desc="Options") to open the model popover
• Searches only within the AXPopover element to avoid matching sidebar items (e.g. "Auto" prefix previously matched "Automatic Trading on Chain")
• For legacy models (5.2-instant, 5.2-thinking): clicks "Legacy models" first to expand the submenu, then selects the model
• Uses desc prefix matching within the popover for robustness

Response polling fix (ask.ts)

• Before: Broke out of the poll loop as soon as any new text appeared — captured thinking intermediate text in thinking mode
• After: Polls isGenerating() which checks if the "Stop generating" button exists in the AX tree. Only reads the final response after generation completes

Files changed

Test plan

• opencli chatgpt model auto — switches to Auto
• opencli chatgpt model thinking — switches to Thinking
• opencli chatgpt model 5.2-thinking — opens Legacy submenu, selects GPT-5.2 Thinking
• opencli chatgpt model auto after legacy model — correctly selects Auto (not sidebar items)
• opencli chatgpt ask "what is 2+2?" --model thinking — waits for full response, not thinking text
• opencli chatgpt ask "explain quantum computing in 3 sentences" --model thinking --timeout 60 — returns complete response