Lecture Notes Pipeline — Data skill for Claude Code

Lecture Notes Pipeline

Record a lecture on a Mac, transcribe it locally with whisper.cpp while the class is still going, and let three Claude Code agents turn the transcript and your own live notes into study notes in an Obsidian vault.

• Local transcription. Audio never leaves the machine; whisper.cpp runs on Apple Silicon with Metal.
• Live drafts. Every 10 minutes a finished chunk is transcribed and two agents update a draft of the notes and a list of slides to check.
• Your notes stay the backbone. If you type notes during class, the transcript fills your gaps instead of replacing them.
• Study-ready output. Notes follow rules from learning research (retrieval questions, concept map, spaced review plan) and land in the vault as a topic note plus a short session log.
• One command per class. classe NET r builds the workspace if needed and starts recording.

An animated walk-through of the flow is in [`docs/index.html`](docs/index.html) (publish it with GitHub Pages, see below).

How it works

flowchart LR
  subgraph Start
    A[classe NET r] --> B[lecture.py --launch] --> C[Herdr workspace
5 panes, 3 agents] --> D[Recorder pane]
  end
  subgraph Live["While recording (every 10 min)"]
    E[Audio: mic / system / both] --> F[10-min WAV chunks] --> G[whisper-cli per chunk] --> H[partial.srt]
    H --> I([notes agent
INCREMENTAL]) --> J[running.md] --> K([research agent
LIVE FLAGS]) --> L[slide-flags.md]
  end
  subgraph Final["After Enter"]
    M[Join chunks] --> N[whisper-cli once
whole lecture] --> O[transcript .srt]
    O --> P([notes agent
ENRICH / NOTES]) --> Q([research agent
FINAL FLAGS]) --> R([Obsidian agent
LECTURE INGEST])
    R --> S[(Topic note + session log)]
  end
  D --> E
  L -. Enter .-> M

**1. Start.** `classe NET r` runs `lecture.py NET --launch`. It looks for an open Herdr workspace for that course (Herdr sets `HERDR_WORKSPACE_ID` in its panes). If there is none it builds o