Edit the project-level audio timeline.

One endpoint, selected by action: add-music-layer, remove-music-layer, generate-music-clip, add-music-clip-from-library, update-music-clip, remove-music-clip, generate-sfx-cue, add-sfx-cue-from-library, update-sfx-cue, update-sfx-cues, remove-sfx-cue, set-voice-duck, set-narration-level, remake-flagged-sfx-cues (set-voice-duck takes voiceDuckDb: how far the music dips under the voices in dB, -18..0, or null to restore the default dip; heard in the preview and baked into the render; set-narration-level takes sceneIndex and level: a beat's narration level, 0..2 with 1 as recorded, or null to restore the default). Music clips render with the Lyria engine, so they are never blocked by the fal balance; describe the SOUND rather than naming a real artist. A sound-effect cue (generate-sfx-cue, add-sfx-cue-from-library) takes kind (spot: one event, its own length; ambience: a bed under the picture that, with no lengthMs, fills its scene from the offset or through endSceneIndex, repeating the effect crossfaded, at 0.2 with short fades unless set), lengthMs (how long it sounds, 100..3600000; longer than the effect repeats it), fadeInMs / fadeOutMs (0..60000) and level; durationSeconds is how long the effect is generated (up to 22). update-sfx-cue changes any of them, muted silences a cue without removing it, and null lengthMs / endSceneIndex put it back on its default span. update-sfx-cues changes or removes several cues in ONE write (a re-balance): cues, 1 to 40 entries, each update-sfx-cue's fields with its cueId, or { cueId, remove: true }; the answer's results says what each did (one naming no cue changes nothing while the rest land). remake-flagged-sfx-cues makes every cue whose file came out too quiet, clipped or gappy again from its own words — the same kind, place, length, fades and mute — and re-levels it to its band when the new file lands, never by turning it up; healthy and unmeasured cues are untouched. Every cue's file is measured as it is placed and played levelled before its level, so a level is a place in the mix: the sound map's suggestedLevels and each cue's targetLevel put a bed 10–14 dB and a spot 3–8 dB under its scene's own mix. The cue in the answer, and every cue in its timeline, carries assetLufs, normGainDb, mixLufs and loudnessFlag (see the timeline read). Requires Idempotency-Key. Scope: story:write.