Noticing that after I enable planning tools in the SDK, OpenHands really wants to make a plan, then ask me for input before proceeding. Are we hacking around that behavior in the web?

I see at least several sections in the planning prompt that send the LLM to ask the user https://github.com/OpenHands/software-agent-sdk/blob/6bf154a740b5f4257a228b327fe6868b7c4ed5c6/openhands-sdk/openhands/sdk/agent/prompts/system_prompt_planning.j2#L68

yeah when I ask the agent to do something at app.all-hands.dev, 95% of the time the agent just does it when using my custom UI built on the SDK, 95% of the time the agent investigates and then asks me a stupid question

Yes, what are the LLMs?

sonnet 4.6 for my app, opus 4.5 for web

This might be worth it, just for ease of understanding what's up https://openhands-ai.slack.com/archives/C07S376KASK/p1771712235361339

to close the loop, this was a vibecoded mistake 🙃