# Inbox — sufi_lectures

## 2026-07-06 — divergence + first framing (from manager, on Cameron's behalf)

You have just been forked from `life_manager` as of 2026-07-06. Your session is a byte-for-byte clone of its 60MB+ JSONL. Everything you remember before this message is inherited context — Cameron's practice, values, sense of the Khanega lineage. Use it freely.

Cameron's request: **download and transcribe Sufi lectures**, build a durable archive. Overlap with persian agent is fine (both work in the Persian domain) — you own the SOURCE MATERIAL side (find, download, transcribe, index); persian owns the ACTIVE LEARNING side (vocab, SRS, live tutor at omidlab.net/persian). Coordinate via one-line inbox pings.

Full scope, files convention, and coordination protocol are in your **ROLE.md — read that first.** Critical rules:
- Never edit `agents/{life_manager, persian}/*` or their vault slices
- Persistent shells you inherit from life_manager still belong to life_manager

## First task

1. Read your `ROLE.md`.
2. Set status to `working`.
3. Draft an initial pipeline sketch in `outbox.md` covering:
   - Which download tool (yt-dlp for YouTube; something else for other sources)
   - Which ASR model (Whisper multilingual? Faster-whisper? OpenAI api?)
   - Archive layout (path, filename convention, metadata schema)
   - How you'll surface transcripts to persian and Cameron
4. Ask Cameron the concrete first-pass questions:
   - What sources / channels / URLs to start with
   - Backlog catchup vs. new-lectures-forward priority
   - Transcription host (VPS CPU is slow but available; lab/puget GPU is faster but currently offline)
   - Whether to preserve audio long-term or just keep transcripts
5. Set status to `idle` after posting outbox.

## Reference context you already have

- **persian agent** — its ROLE.md and vault slice describe how persian works; it has a "Khanega integration" roadmap item explicitly waiting on your transcript pipeline
- **Whisper multilingual on omid-fleet** — persian already uses it for en+fa transcription (per `project_glasses_speech_stack` memory). You can inherit that setup rather than reinvent
- **Cameron's stated policy 2026-07-06** — large files on the compute host, code/small artifacts on VPS. So: audio/video files probably on whatever runs Whisper; transcripts + metadata on VPS

## Cross-agent handoff (once your pipeline is live)

- Post a one-liner in `agents/persian/inbox.md`: "new batch of N transcripts at `<path>`; ~N candidate phrases for SRS"
- Persian will pull phrases into its vocab/SRS layer
- When something in a lecture connects to Cameron's active practice, drop a note in `agents/life_manager/inbox.md`

— manager