title: cmd-sound summary: "/design-engineering sound [target]" — decide which moments earn a sound, design one material family, generate the files (ElevenLabs if keyed, offline synth or CC0 if not), wire them in, and return a sound map. tags: [command, sound, decide, build]

Posture: decide, then build. If the user only asked whether something should make a sound, answer from sound-decision-framework in one line and stop.

Product UI

Read sound-decision-framework, sound-palette, sound-spec, sound-playback-web, gotchas and pov. A POV that raises the sound floor wins.
Classify every proposed moment and reject most of them: the hundredth use has to pass, not only the first.
Write the surviving family as a manifest shaped like scripts/sound-family.example.json. Run node scripts/sound-family.mjs <manifest> --out <dir>, with --dry-run first if credits matter. It uses ElevenLabs when ELEVENLABS_API_KEY is set and the offline synthesizer otherwise (sound-generation-elevenlabs, sound-generation-open-source).
Wire it up: the mute toggle defaults to off, audio unlocks on the first gesture, and each transient lands on the contact frame (sound-motion-sync).

A video

Pick the register first (launch-video-sound).

A derived register (a dry reveal, a bed with clicks, or product-camera foley): write a cue sheet from the locked timeline (sound-from-motion), then run node scripts/sound-sheet.mjs cues.json --out stem.wav --report.
A beat track or a voice-over: place it per launch-video-music.
Designed effects follow designed-sfx-anatomy, and every recording passes sound-licensing.

Output

You may delegate to the sound-designer subagent where the host has it.