This comment on Hacker News inspired me to investigate patterns for using my LLM CLI tool in a shebang line: