music

Step 6 of the content pipeline, optional. Adds a flat music bed under the already-polished dialogue in a single mix pass — no ducking automation, no per-segment adjustments. Use only when the user ask

SKILL.md

GitHub

Music (optional)

Skip this entirely unless the user wants a music bed. When they do:

1. Get the track

If the user supplies a file, use it. If they need one sourced or generated, that's /media-use (BGM resolution/generation) — come back here once you have a local audio file.

2. Mix it in, once

.claude/skills/music/scripts/mix_music_bed.sh <job-name> <music-file> [bed_db_below_dialogue]

Default is 18dB under the dialogue. This is deliberately a flat bed — one static volume offset for the whole track, mixed once against whatever the job's furthest-along render is (captioned > graphics > rough cut). No ducking curve, no per-segment volume automation — those would violate the "never per-segment audio FX" lock and reintroduce the click/pump risks that lock exists to avoid. If a scene genuinely needs the music to duck under a specific line, that's a deliberate creative call to make with /hyperframes-audio's automation envelopes inside the HyperFrames composition itself (before this flattening mix), not a patch applied here.

Writes outputs/<job>-music.mp4 with dialogue untouched and the mix re-encoded at AAC 256k (matches, never drops below, the polish pass's bitrate).

3. Continue

Straight to /export.