onecli cli help is kinda unusable for a human
I think LLMs can easily read help text, can discover the command to get a json output if necessary.
a common pattern now is also to ship the cli with a skill that gives the hints an agent would need to use the cli effectively anyhow.
"should human be the default or json? and how would you want to flip between them?"
so: human default, since if i'm reaching for the cli, its probably to debug or investigate a problem.
re: flpping between them (@johnnyfish ) that i'm not sure about - i like your idea of detecting tty / env vars to make a good guess. but i do agree with --json flag would make sense (and again, easy enough for agent to discover/ know about) Otherwise, what did you mean by flipping?
onecli cli help is kinda unusable for a human
I think LLMs can easily read help text, can discover the command to get a json output if necessary.
a common pattern now is also to ship the cli with a skill that gives the hints an agent would need to use the cli effectively anyhow.
"should human be the default or json? and how would you want to flip between them?"
so: human default, since if i'm reaching for the cli, its probably to debug or investigate a problem.
re: flpping between them (@johnnyfish ) that i'm not sure about - i like your idea of detecting tty / env vars to make a good guess. but i do agree with --json flag would make sense (and again, easy enough for agent to discover/ know about) Otherwise, what did you mean by flipping?