Claude Code now injects an unknown_command_fallback reminder when a typed slash command doesn't exist
What
When a user's message starts with a slash command that isn't available in the current session, Claude Code now injects a new unknown_command_fallback reminder into the transcript. It tells the model the command name (when it's safe to show), suggests the closest available command, lists the commands that ARE available in the session, and instructs the model to treat the message as a plain request rather than claim the command ran.
Why
This stops the model from silently pretending a nonexistent command succeeded, and instead points the user toward a command that does exist or has it handle the message as ordinary text.