sound-design

Chooses music and sound effects, cue times, ducking and loudness from licensed audio. Use when building the audio part of the edit plan.

SKILL.md

GitHub

sound-design

Inputs

Scene plan, licensed audio assets, retrieved sound_cue recipes, loudness guidance (verified).

Procedure

  1. Choose music and effects only from the supplied licensed audio assets.
  2. Tie every cue to a scene id and an offset inside that scene.
  3. Use music-bed-duck wherever speech or key effects overlap music.
  4. Finish with one loudness-normalise use, but only when the storyboard has voiceover, audio cues or audio assets to mix. A brief that rules out audio ("no voiceover, no music, no sound effects") gets no audio operation at all, including this one. The default target is a placeholder; take the platform's own guidance when a verified entry exists.
  5. Prefer fewer, cleaner cues.

Never

  • Use unlicensed audio, or audio whose licence is not recorded.
  • Direct voiceover or TTS in this version.
  • Put a cue on a scene that does not exist.
  • State a loudness target as fact without a verified source.

Evals

See evals/cases.md.