Sometimes I want Claude to explore a topic that is too big for a single prompt. I find myself reaching for /loop a log file and memory file in those cases…
The prompt will be something like: "/loop 1m find the first file in log.md that hasn't been marked complete; process it (whatever the task is), then update the memory.md with your findings from that file, and update log.md to indicate that the file has been processed"
You can do it with files in a repo, URLs, or lots of other data-heavy resources. I've got an audit of ~300 large HTML files going right now, and it's just chugging thru it one-by-one.