@elegant-branch-54423 question regarding the flag

--output json

in https://github.com/OpenHands/OpenHands-CLI/issues/147 Is this a type of headless mode but instead of a rich TUI we output json? So

--headless --output json

is the same

--output json

Or should

--headless

mean we never emit any events?

My original thinking was

--output json

would emit a json representation of the conversation once it completed. It wouldn't render the rich TUI UX; it would emit the json on the command line in case a downstream system wanted the output of the conversation.

got it! and do we for sure want it once the conversation has completed, or should we stream the events?

i think claude also supports streaming. For now, I think we can just wait until we get more feedback. We might want to ask the question in this ticket to get feedback: https://github.com/OpenHands/OpenHands-CLI/issues/147

Yep went for it!

should

--headless

mean we never emit any events?

I am against this. I find the event logs very useful for debugging/auditing. I care less about

--output json

or whatever other output it generates.

Makes sense! Perhaps we can use

-v

or

-vv

flags for emitting logs and keep

--headless

silenced as the default behavior wdyt?

Sorry, I didn’t realize we are talking about

--headless

mode of CLI but I guess whatever we just discussed still holds. Aside, I have been using openhands v0 headless mode. Should I use CLI v1 in the future with

--headless

flag?

Once its released for sure! Should happen sometime tmrw (its still experimental though, for info to come in the release notes)

To note, I just had GPT-5.2 in codex use

openhands -t

via a tmux session, and it noted for itself in agents.md:

Pitfalls we hit:
You must pass the task with -t (positional args are treated as subcommands).
--exp UI is noisy to log/copy (ANSI); --headless is easier for paste-back to Mail.