I've been working on connecting up the CLI to VSCode, and I'm always using the argument to ensure I get complete, structured data. In case of errors though, the command output goes to instead of and is usually (always?) a string instead of json.

I'm hoping for a more stable way to identify these error message types, rather than matching on individual words or phrases that may change in the future. For instance, when publishing a task if there is a conflict I want to detect that and offer to open up the diff editor. But right now, I'm matching on an error string that could change.

Perhaps exiting with standardized error codes? eg.

It looks like already uses exit code 2 for conflicts, so this system could possibly be expanded to cover to all errors.

I'm not sure if it would make sense to format stderr as json when the argument is passed, but maybe that's also worth considering.