Skill file: ~/Documents/Claude/Scheduled/transcript-scan/SKILL.md
Phase 1 (local sweep) — recursive search of:
~/Downloads,~/Desktop,~/Documents(esp.Zoom/,WebEx/,Sound recordings/)/SkyRun/Meeting Notes/,/SkyRun/Onboarding & Reference/~/Onedrive,~/OneDrive-Personal
File types: .txt, .md, .vtt, .srt, .docx. Filename patterns: transcript, meeting, recording, otter, fathom, fireflies, gemini, spinach, grain, gong, zoom, plus participant names (rachel, mandy, jasmine, cathy, wayne, adam, isabelle, tyler, paul, julie).
Phase 2 (gmail sweep): Chrome MCP at https://mail.google.com/mail/u/0/#inbox. Sender searches:
from:noreply@otter.aifrom:Spinach AI OR from:noreply@spinach.iofrom:meetings-noreply@google.com(Gemini)from:no-reply@zoom.us recordingfrom:fred@fireflies.ai OR from:noreply@fireflies.aifrom:noreply@grain.com OR from:noreply@grain.cofrom:noreply@fathom.videofrom:hello@read.aifrom:noreply@gong.io- Plus subject contains "transcript" (added 2026-04-15 per Joseph's direct request) — catches copy-pasted transcripts in email bodies.
Phase 3: Update /SkyRun/Call Transcripts/index.json with new entries + _meta.total_files + _meta.total_chars.
File convention: YYYY-MM-DD_meeting-slug_type.txt → transcripts/ (verbatim, word-for-word) or notes/ (summaries, recaps, placeholders).
How to apply: When modifying, preserve the verbatim-vs-notes split. Dedupe by date+slug. Don't re-import — index.json entries are source of truth.