CrateBot5 — Development skill for Claude Code

CrateBot 2.0

Audio fingerprinting and intelligent music analysis suite for DJs.

**CrateBot** uses machine learning to analyze your music library and suggest songs you might want to use while playing live. Think of it as a DJ's intelligent crate-digging assistant.

Built with Claude Code by someone who has basically never written a line of code in his life. See: [The Road Runner Economy](https://nraford7.github.io/road-runner-economy/)

What It Does

• ML-powered tagging: Predict Genre, Timing, Mood, and Descriptive tags for your tracks
• 2192-dimensional feature vectors: EffNet (1280) + Jamendo genres (400) + CLAP (512) embeddings
• Native Swift training: Train custom models on your own tagged library using CoreML
• Vibe generation: Claude API-powered tags that capture what makes each track distinctive
• Mnemonic anchors: Album-art-like memory hooks (2-3 word phrases that feel like the track)
• Hook detection: Native Speech framework for detecting memorable vocal moments
• Real-time progress: Live updates for long analysis tasks
• Checkpoint recovery: Resume interrupted training

Vibe System

CrateBot generates two complementary tags for each track:

Short Vibe Tag

**Format:** `[ENERGY] [DISTINCTIVE THING] [MOMENT]`

Identifies what makes the track unique - the thing you'd tell a friend to listen for.

DARK FLUTE MELODY PEAK
HARD ACID 303 SQUELCH PEAK
JOYFUL KALIMBA GROOVE OPENER
DREAMY STRINGS PIANO BLEND FLOATER

Mnemonic Anchor

**Format:** `[synesthetic modifier] + [concrete anchor]`

A 2-3 word phrase that works like album art in text form - triggers recall through association, not description.

sweating serpent
chrome shaman
golden grandmother
velvet cathedral

The modifier translates sonic qualities to other senses (warm, dusty, chrome, velvet). The anchor is something you can picture (wizard, panther, cathedral, shaman).

Architecture

Built with **SwiftUI** and **CoreML** for fully na