Triage huge Claude Code transcript archives with grep hit-counts first

When mining ~100MB JSONL session transcripts for a topic, do a per-file grep -c hit-count pass first to rank sessions, then extract only user messages (type==user, string content) via a small python loop instead of reading files. Complex alternation greps over 100MB files can exceed 2-min timeouts - split into simple patterns or background them. Also: osascript can pull Apple Notes bodies by title match, useful when a user drafted content in Notes.