Skip to content

Conversation

@adam91holt
Copy link

Summary

  • Adds --agent CLI flag to enable ChatGPT's Agent mode for agentic tasks (web browsing, code execution)
  • Agent mode activation clicks the "+" button and selects "Agent" from the menu before submitting
  • Response capture waits for the send button to reappear (not just stop button disappearing), ensuring full agent output is captured after all actions complete

Test plan

  • Manual test: oracle --engine browser --agent --prompt "What is the current Bitcoin price? Search the web." returns actual search results
  • Unit tests added for config propagation (agent: trueagentMode: true)
  • TypeScript compiles without errors
  • Test on different ChatGPT UI languages (agent mode uses text matching for "agent")

🤖 Generated with Claude Code

Enables ChatGPT's agentic task mode (web browsing, code execution) via
the browser engine. Key changes:

- New `--agent` CLI flag for browser mode
- Agent mode activation clicks "+" button and selects "Agent" from menu
- Response capture waits for send button to reappear (not just stop
  button disappearing) ensuring full agent output is captured
- Longer settle windows and stable cycles for agent mode since agents
  pause between actions

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <[email protected]>
@steipete
Copy link
Owner

steipete commented Dec 8, 2025

Interesting. That's not really Pro then anymore? How different is this to Pro, which also works agentic?

@adam91holt
Copy link
Author

It can use the full web browser with JavaScript and perform tasks etc.

However, clawd will love this. He will essentially get a whole containerised environment to run a browser running on OpenAI compute.

Your ability is better than mine to code so you can probs do it cleaner haha. Maybe it's a different package?

@steipete
Copy link
Owner

steipete commented Dec 9, 2025

Clawd uses playwright for that. I'm not super convinced this is needed - oracle is for code.

Do you have concrete examples how this helps you with agentic coding tasks?

Main gripe: Every option I add is one more to confuse models and new automatic flows that can break.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants