Cal Rueb from the Anthropic team just revealed the Claude Code features that 90% of users will never find on their own
he writes the system prompt that powers Claude Code - and spent 25 minutes showing how to use it properly
> claude md files = memory that persists across every session
> Escape twice = jump back in your conversation history
> run 4 Claude Codes in parallel and orchestrate them like a team
> Claude 4 now thinks between tool calls - throw "think hard" at every bug
save this - it's the closest thing to an internal manual for Claude Code you'll ever find for free