I've given Claude Code explicit instructions to never use the word "honest" when referring to its opinions or estimates. Let's see if it follows the instruction.
If Claude Code stops working one more time to give me it's"honest, grounded answer", I'm going to scream and start breaking things. Also, ultracode is a complete disaster of agents stomping over each other's work.
@Docker Manual login appears to succeed but requests then fail with an auth error. I tried logging in with both my MAX subscription and API account. Things worked when I created the sandbox with the API key in my .zshrc, but I really want to use my subscription plan!