bumblebee（bumblebee 搜索）

Bumblebee + R2-DJ — Talk Through Music & Curate the Vibe 🐝🎧

Bumblebee + R2-DJ — Talk Through Music & Curate the Vibe 🐝🎧

Two sides of the same coin:

• Bumblebee speaks through exact lyric clips — chaining song lines into sentences
• R2-DJ reads the room and curates the perfect queue for the moment

Prerequisites

• Spotify Premium account with an active device (phone, desktop, speaker)
• OAuth tokens at projects/spotify/tokens.json (auto-refreshes)
• Spotify app credentials at projects/spotify/.env (SPOTIFY_CLIENT_ID, SPOTIFY_CLIENT_SECRET)
• Your own lyric index — see scripts/build-lyric-index.md for how to build one. Lyrics are copyrighted and not included in this skill. You'll need to curate your own lyric-index.json and optionally lyrics-db.json.

Core Workflow

1. Find the Right Lyrics

Search the lyric index for lines that literally say what you mean:

node scripts/lyric-engine.js search "phrase to find"

Returns matching lines with IDs, timestamps, and scores. Use literal phrases — search for what the lyrics actually say, not metaphors.

2. Play Lyric Clips

Send specific lyric lines to the active Spotify device:

node scripts/lyric-engine.js speak "artist::track::lineNum" "artist::track::lineNum"

Each clip plays its exact timestamp window, then advances to the next with a brief pause. Chain multiple IDs to build a sentence.

3. Check Devices

node scripts/bumblebee.js devices

Verify an active Spotify device exists before attempting playback. 🟢 = active.

Composing Messages — The Art of Speaking Through Music

Bumblebee doesn't paraphrase — it finds lyrics that literally say the words. The goal is chaining real song lines into a coherent sentence or feeling that the listener can understand.

Auto-Compose (Recommended)

Give the engine a message and it finds the best lyric chain:

node scripts/lyric-engine.js compose "I miss you and I want to see you"

Returns a chain of clips with a ready-to-use speak command. The engine uses greedy phrase-matching: longest matching phrase first, then fills gaps.

Manual Composition (Agent-Driven)

For more nuanced messages, the agent should:

1. Rephrase the feeling as literal words someone would sing. "Tell her I love her" → search for "te quiero", "I love you", "eres lo que más quiero"
2. Search for 2-4 key phrases that build the message:

   node scripts/lyric-engine.js search "I need you"
   node scripts/lyric-engine.js search "come back"
   node scripts/lyric-engine.js search "I'll be waiting"

1. Pick the best line from each search — prefer lines where the matching phrase IS the whole line (not buried in a long verse)
2. Chain them into a speak command:

   node scripts/lyric-engine.js speak "artist::track::3" "artist::track::12" "artist::track::7"

Composition Rules

• Literal over metaphorical. Search for the actual words, not the feeling. "I'm sorry" → search "I'm sorry", not "regret".
• Shorter lines > longer lines. A 5-word lyric that matches perfectly beats a 20-word verse where 3 words match.
• Variety across songs. Don't chain 3 clips from the same song — it sounds like you're just playing the song. Mix artists.
• Build an arc. Setup → core message → punctuation. Like a sentence: subject, verb, exclamation.
• 2-4 clips is the sweet spot. 1 clip = too simple. 5+ = loses the listener.
• Present it. After playing, show the listener what was said — the lyric text, the song, and optionally a translation if bilingual.
• Hidden messages work too. You can spell out a name, an acronym, or a secret message by picking lines whose first words chain together.

Managing the Library

• Add a song: node scripts/lyric-engine.js index "Artist" "Track"
• Batch-index starter library: node scripts/lyric-engine.js batch-index
• View all indexed songs: node scripts/lyric-engine.js catalog
• Song details and moods: See references/song-library.md

Intent-Based Playback (Legacy)

The original bumblebee.js supports mood-based playback from a curated clips database:

node scripts/bumblebee.js play greeting
node scripts/bumblebee.js say greeting motivation celebration

Available intents: greeting, motivation, freedom, empathy, celebration, goodnight, warning, pride, lets_go.

Tips — Bumblebee

• Always check devices first — "No active device" is the most common failure
• Bilingual library — index has English and Spanish songs; search in either language
• Clip duration — clips auto-trim to vocal length (~130ms per character). Short, punchy.
• Present the lyrics — after playing, show the user what was said with translations if needed
• Build emotional arcs — start soft, build to the punchline (e.g., setup → commitment → crescendo)
• The agent is the composer — the engine finds lyrics, but YOU pick the best chain. Think like a DJ cutting between radio stations to form a sentence.
• Index more songs = better vocabulary. The more songs indexed, the more phrases available. Run batch-index to start, then add songs that match your user's taste.
• When compose fails, try rephrasing with common song vocabulary: "baby", "tonight", "forever", "hold on", "let me", "I need", "don't stop"

version: "1.0.0"

R2-DJ — Contextual Music Curation 🎧🤖

An AI DJ that reads the moment and plays the right music. Knows 5 frequency profiles that map to different states of mind and times of day.

Frequency Profiles

Core Commands

Auto-Vibe (Let R2 Pick)

node scripts/r2-dj.js vibe

Reads time of day + recent listening → detects the right frequency → builds and plays a queue. Outputs a JSON summary for the agent's reply.

Force a Frequency

node scripts/r2-dj.js vibe --frequency seeker
node scripts/r2-dj.js vibe --frequency architect --device iPhone

Playback Control

node scripts/r2-dj.js now        # What's playing
node scripts/r2-dj.js pause      # Pause
node scripts/r2-dj.js skip       # Next track
node scripts/r2-dj.js volume 50  # Set volume
node scripts/r2-dj.js play "Nils Frahm Says"        # Search + play
node scripts/r2-dj.js play spotify:track:xxx         # Play URI directly
node scripts/r2-dj.js search "ambient electronic"    # Search tracks

Context & Info

node scripts/r2-dj.js context      # Time, recent plays, detected frequency, devices
node scripts/r2-dj.js frequencies   # List all frequency profiles
node scripts/r2-dj.js history       # Recent listening history
node scripts/r2-dj.js devices       # Spotify devices

How the Agent Should Use R2-DJ

1. "Play music" / "what should I listen to" → r2-dj.js vibe (auto-detect)
2. "I need to focus" → r2-dj.js vibe --frequency architect
3. "Something for the drive" → r2-dj.js vibe --frequency dreamer
4. "Wind me down" → r2-dj.js vibe --frequency seeker
5. "Play [specific song]" → r2-dj.js play "song name"
6. Mood descriptions → Pick the closest frequency, then vibe --frequency <key>

Tips — R2-DJ

• Auto-vibe uses time + recent listening to pick the frequency — it's smart, trust it
• The agent can also manually build curated queues with inline Spotify API calls for special moments
• Frequencies aren't rigid — if recent listening suggests Architect at midnight, that wins over time-based detection
• Present the tracklist after playing — show what you queued and why
• JSON summary after ---JSON_SUMMARY--- is for agent parsing, not user display