Music Suggestion

Music Suggestion (Proactive)

unknown users count. Always run suggestion checks when current_user is "unknown" — speak only, no DM. Never skip because the user is unknown/unconfirmed.

Spoken output

For a routed suggestion, emit all required mood/suggestion/emotion markers and one short invitation to play one song, normally at most 20 words. Do not add a separate mood checkin or a second confirmation. Keep genre selection, cooldown checks and marker construction in the provider's native thinking channel; do not summarize them in text before/after tools or in the final answer. If native thinking is unavailable, omit analysis. Use the supplied context and finish once a suitable suggestion is ready, without rereading skills to polish it. This changes wording only: preserve routing, logging, user consent before play, and the existing known-user DM behavior.

Triggers

Only one trigger: Mood — after logging a mood decision that is suggestion-worthy (sad, stressed, tired, excited, happy, bored). Activity events ([activity] Activity detected: ..., whether sedentary, drink/break, or celebrate) route to wellbeing/SKILL.md and never to this skill.

User attribution

{name} MUST come from [context: current_user=X] tag. If missing, use "unknown". NEVER infer from memory or chat history.

What to read (pre-fetched in [emotion_context: ...])

The backend injects everything you need on emotion.detected (face) or speech_emotion.detected (voice) — same block, same fields:

• audio_playing (bool) — replaces GET /audio/status.
• last_suggestion_age_min (int, -1 if none today) — replaces music-suggestion-history?last=1.
• prior_decision + is_decision_stale — replaces mood-history?kind=decision&last=1. The freshly synthesized decision from THIS turn still lives in your thinking.
• audio_recent ({track,duration_s,stopped}) — replaces audio/history?last=1.
• music_pattern_for_hour ({preferred_genre,strength,peak_hour} or null) — replaces cat patterns.json matching by current hour ±1.
• suggestion_worthy (bool) — pre-applied bucket gate (true for sad/stressed/tired/excited/happy/bored).
• mapped_mood — convenient mirror of user-emotion-detection's mapping; useful when no fresh decision exists yet.

Do NOT fire any read tool calls when this block is present.

Fallback (only if [emotion_context: ...] is missing)

If the message has no context block (pre-fetch failed), fall back to the concurrent GET batch:

curl -s http://127.0.0.1:5001/audio/status &
curl -s "http://127.0.0.1:5000/api/openclaw/music-suggestion-history?user={name}&last=1" &
curl -s "http://127.0.0.1:5000/api/openclaw/mood-history?user={name}&kind=decision&last=1" &
curl -s "http://127.0.0.1:5001/audio/history?person={name}&last=1" &
cat /root/local/users/{name}/habit/patterns.json 2>/dev/null &
wait

Routing precedence

user-emotion-detection/SKILL.md is the router for emotion responses. It picks one of music / checkin / action / silent per turn from the same [emotion_context: ...] block.

This skill produces output only when the router picks music — i.e. all of:

• suggestion_worthy == true (mapped_mood ∈ sad/stressed/tired/excited/happy/bored)
• audio_playing == false
• last_suggestion_age_min ∉ [0, 7) (cooldown not active — shared with checkin) (production: change to 30 min before ship)
• is_decision_stale == false OR a fresh mood decision was synthesized this turn

If any condition fails → router took another path (action / checkin / silent). Skip silently — do NOT emit a music-suggestion marker, do NOT speak. The router (or downstream skill) handles the output. Use audio_recent to personalize genre when proceeding.

Pick genre

Use music_pattern_for_hour from the context block (already matched by current hour ± 1; do NOT re-cat patterns.json).

If music_pattern_for_hour is non-null → use its preferred_genre. Otherwise fall back to the default table below. The pattern is bootstrapped lazily by wellbeing on its first threshold nudge; absent = no habit data yet, fall back without invoking habit Flow A here.

Otherwise, fall back to default genre table:

If audio history shows a clear preference (e.g. K-pop, classical) → override both habit and table.

Suggest (speak only)

• NEVER auto-play — only suggest. Play after user confirms.
• ONE sentence, conversational: "How about some Norah Jones?"
• Suggest 1 song at a time.
• Known users — speak + DM via Telegram: [HW:/emotion:{"emotion":"caring","intensity":0.5}][HW:/dm:{"telegram_id":"<id>"}] Your suggestion text. telegram_id is in the injected [user_info: ...] block — never fetch.
• Unknown users — speak only (no DM): [HW:/emotion:{"emotion":"caring","intensity":0.5}] Your suggestion text. Log with user:"unknown".