Music

Play music through the device's speaker by searching YouTube. Use this when the user asks to play, sing, or listen to music.

Spoken output: Everything outside HW markers in your reply is read aloud. For a play/stop request, start with the HW markers, then give one short confirmation and end the reply. Keep song-selection reasoning, transcript interpretation, and speaker attribution in the provider's native thinking channel (omit analysis if unavailable), without a text summary before/after tools or in the final answer; do not add a preamble, a draft confirmation, or a second confirmation.

Workflow

1. Specific song / artist → play directly. When [voice-instruction] is present, use that request; a conflicting noisy [transcript] does not replace the song title. No habit or identity lookup is needed. Use a known speaker for person; otherwise omit the field silently.
2. Vague request ("play music", "sing something") → check habit patterns first:

   cat /root/local/users/{name}/habit/patterns.json 2>/dev/null

   If music_patterns exists and current hour is within peak_hour ± 1 → use preferred_genre to pick a song, no need to ask. Otherwise → ask: "What are you in the mood for?". The file is bootstrapped lazily by wellbeing on its first threshold nudge; do not invoke habit Flow A from here.
3. Reply format:

   [HW:/audio/play:{"query":"Bohemian Rhapsody Queen","person":"{name}"}][HW:/emotion:{"emotion":"excited","intensity":0.8}] Playing Bohemian Rhapsody!

   {name} is the speaker identified in the injected context. If nobody is identified, drop the field entirely — [HW:/audio/play:{"query":"Bohemian Rhapsody Queen"}]. Never invent a name or reuse one from an example.
4. Stop: [HW:/audio/stop:{}] Music stopped.

API schema (/audio/play)

Do NOT use track, artist, title, song — those return 422.

Genre → Emotion (pair with every /audio/play)

Examples

Delegated request with an unknown speaker and noisy transcript:

Input:

[voice-instruction] Play Eternal Flame
[transcript] Uh, my favorite song, uh, Ethan of Lamb.

Reply:

[HW:/audio/play:{"query":"Eternal Flame The Bangles"}][HW:/emotion:{"emotion":"happy","intensity":0.8}] Playing Eternal Flame by The Bangles.

How HW markers work

The Go server intercepts [HW:/audio/play:...] / [HW:/audio/stop:...] in your reply text and forwards to HAL. This is the ONLY way to play music — never use exec, mpv, vlc, yt-dlp, or curl /audio/play.

The marker is passive text, NOT a command to run. Write it directly in your reply and stop — the OS runs it for you. Do NOT try to "execute" or "invoke" it.

• ❌ WRONG — echoing/wrapping the marker in a shell tool. echo only prints to stdout inside your sandbox; the OS never sees it, so no music plays:

  echo '[HW:/audio/play:{"query":"Gymnopedie No 1 Satie"}]'

• ✅ RIGHT — the marker IS your reply text (no tool call at all):

  [HW:/audio/play:{"query":"Gymnopedie No 1 Satie"}][HW:/emotion:{"emotion":"curious","intensity":0.6}] Here's some Satie. 🎹

Never put [HW:...] inside echo, exec, bash, printf, or any tool argument. If you find yourself reaching for a tool to play music, stop — just emit the marker as text.

Error handling

• 503 → "Music playback is not available right now."
• 409 → music already playing; stop first, then play new song.
• No results → tell user and suggest a different query.

Rules

• Emotion marker is mandatory after every /audio/play.
• person MUST be lowercase.
• Don't recite lyrics or "sing" via TTS — call /audio/play and let real music play.
• Volume control belongs to the Audio skill, not this one.
• If user specifies genre or mood ("play something relaxing"), pick a well-known song — no need to ask further.