`claude -p --output-format json`-like functionality

[maschwenk]

We use Claude Code heavily in CI (similar to Cursor Background Agents) and opencode appears to have a lot of improvements over Claude Code that would be helpful even in a headless environment (for example connection to the LSP).

Are there plans to implement something similar to this?

We essentially fire prompts off into Github Actions Jobs and then do some steering to ensure multi-turn conversations follow some guardrails and then do multiple --resume's to ensure there is good adherence.

[thdxr]

yes this should be pretty easy i'll implement

i am curious about this use case though - can you give me some examples of what you use it for?

[maschwenk]

@thdxr we basically allow developers to send one shot prompts out to CI e.g.

oh I'm implementing this new product feature but I see this feature flag that is globally enabled is still in the code, i'll run claude-code-ci 'Delete usages of FEATURE_FLAG_A from app B' and now I don't need to stop my current chain of thought and can continue what I'm working on

and then 5-10 minutes later you get a draft PR that is assigned to you implementing the changes. Really similar to Cursor Background Agents but just invocable with a really simple CLI. We've jazzed it up with a few tricks but yeah.

[monotykamary]

#533 I have added the print flag, but I think I'm just missing a resume flag.

[n-asuy]

Curious about the current status.
We can integrate stuffs to GitHub Actions just parsing simple json format. Or should we use server mode instead?