terminalPipeline — How Each Book Gets Made
1. ANALYZE (Python + Haiku — cheapest)
/midi-analyze → analysis_data.json
Reads .mid files, extracts harmony/rhythm/structure stats per song.
2. SONGWRITING BOOK (Python + Haiku/Sonnet)
/midi-sonnet-guide → songwriting_book.md (13 chapters, auto-generated)
Raw statistical guide. Usable as-is but reads like a report, not a teacher.
3. SONG STYLE BOOK (Sonnet via band-book-rewrite)
/band-book-rewrite → song_style_book.md (N chapters = N songs)
Rewrites each song analysis into teacher-voice with writing rules and
texture notes. This is the book you actually READ to learn the band.
4. ARCHETYPE BOOK (Opus — expensive, use only when needed)
/midi-archetype-book → archetype_book.md (N archetypes)
Clusters songs into compositional forms. Verse/chorus templates,
band-language vocabulary. Opus quality but Opus cost.
MODEL PRIORITY: Python first → Haiku → Sonnet → Opus (troubleshooting only)
To generate a missing book: Ensure the band folder has analysis_data.json (or run /midi-analyze first).
Then run the appropriate skill. For songwriting books, Haiku is sufficient. For song-style rewrites, Sonnet produces
the teacher voice. Archetype books require Opus and should only be generated when the band has 5+ analyzed songs.