Skip to main content
skill CX Codex

music-supervision-scoring — Claude Code skill for video music

music-supervision-scoring is a Claude Code skill for planning music in video projects. It supports track selection, licensing, scoring briefs, and audio reviews.

MIT 263 installs ★ 252 updated 2026-07-14

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add calesthio/generative-media-skills --skill music-supervision-scoring -a claude-code
Open on GitHub

What music-supervision-scoring does

The skill helps create a music supervision plan, a composer brief, a shortlist of licensed tracks, an AI music prompt, a draft cue sheet, or a mix review. It connects musical choices to the scene’s dramatic purpose, edit points, and handoff needs.

A cue plan can record timecodes, sync points, emotional shifts, and stem requirements. The skill also turns a temp track into musical attributes such as tempo, instrumentation, and rhythm, rather than asking for a copy of a known song or artist. A composition and its particular recording are separate works for copyright and licensing purposes.

What it helps with

  • plan music cues for a trailer or product film
  • write a composer brief with timecodes and sync points
  • describe a temp track without naming an artist to imitate
  • find moments where music competes with speech or sound effects

Example prompts for the agent

  • Plan music cues for this trailer, including entrances, exits, and sync points.
  • Write a composer brief for this edit with timecodes and emotional changes.
  • Turn this temp track into an AI music prompt using musical traits, not artist references.

How to install music-supervision-scoring

1Skills CLI (Vercel)

npx skills add calesthio/generative-media-skills --skill music-supervision-scoring -a claude-code

2Manually: put the files in place

~/.claude/skills/music-supervision-scoring/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/music-supervision-scoring/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about music-supervision-scoring

How do I install music-supervision-scoring in Claude Code?

Run npx skills add calesthio/generative-media-skills --skill music-supervision-scoring -a claude-code. The skill file is placed at ~/.claude/skills/music-supervision-scoring/SKILL.md.

What does music-supervision-scoring do?

It helps plan music for audiovisual work, prepare scoring briefs and generation prompts, draft cue sheets, and review mixes.

Which rights apply to a commercial recording?

The composition and the sound recording usually need separate clearance. A cover or new recording changes the master-recording side, while rights to the composition still need attention.

Source text

SKILL.md · MIT
---
name: music-supervision-scoring
description: Provider-independent music supervision and scoring direction for AI agents producing generated videos, ads, trailers, explainers, product films, social clips, documentaries, game trailers, and branded content. Use when planning, prompting, selecting, licensing, mixing, repairing, or quality-reviewing music, score, temp tracks, needle drops, AI-generated tracks, cue sheets, stems, loudness, loopability, platform music constraints, and rights-aware audio handoff for visual media.
---

# Music supervision and scoring direction

Treat music as a production decision, not decoration. A strong music plan explains what the audience should feel, where music should enter and leave, what rights path is safe, how the cue will survive edit changes, and how the final mix will protect speech, captions, audio description, and delivery specs.

Use this skill to produce a music supervision plan, score brief, licensed-track shortlist, AI-music prompt, cue sheet draft, mix handoff, or music QA review for audiovisual work.

## Evidence labels

Use these labels when making claims or giving recommendations:

- Documented fact: supported by law, standards, official platform docs, PRO guidance, or other authoritative sources. Date volatile facts.
- Empirical observation: based on inspecting the actual edit, stems, waveform, transcript, delivery specs, or generated outputs.
- Production heuristic: a practical recommendation that often works but depends on the brief, audience, platform, and mix.

Do not turn a heuristic into a rule. For example, "trailer drums should always rise" is weak; "this trailer's second act needs rhythmic acceleration because the edit compresses from 3-second shots to 0.8-second shots" is useful.

## First pass: define the music job

Before choosing music, answer these production questions:

1. Deliverable: format, duration, target platform, aspect ratio, paid/organic use, territory, term, client/brand sensitivity, and whether this is editorial, commercial, internal, broadcast, festival, game, or social.
2. Narrative function: what music must do that picture, edit, voice, and sound design cannot do alone.
3. Rights path: original score, generated music, stock/library track, licensed commercial recording, public-domain composition with newly licensed recording, client-supplied track, or no music.
4. Mix priority: dialogue/narration, on-screen sounds, sound effects, music, captions, audio description, and silence.
5. Music identity: genre, tempo range, instrumentation, era, geography, performance texture, harmonic color, density, vocal/instrumental choice, and brand fit.
6. Deliverables: full mix, instrumental, no-drums, no-lead, stems, loops, cutdowns, stings, alt endings, and cue sheet data.

Documented fact, verified 2026-07-10: a music supervisor's professional role includes shaping musical vision, tone, and style with the creative team. Source: Guild of Music Supervisors, https://www.guildofmusicsupervisors.com/what-is-a-music-supervisor

## Spotting: decide where music belongs

Create a spotting map before prompting or licensing. For every cue, capture:

| Field | What to decide |
|---|---|
| Cue ID | Stable number such as M01, M02, M03; do not renumber casually after revisions. |
| Timecode | In, out, duration, and edit lock status. |
| Cue type | Score, source/diegetic, licensed song, trailerized hit, bumper, logo sting, loop, silence. |
| Dramatic job | Tension, release, irony, credibility, intimacy, propulsion, reveal, memory, transition, button. |
| Sync points | Words, cuts, camera moves, title reveals, product moments, jokes, impacts, game mechanics. |
| Emotional change | Start state, turn, end state; avoid one static mood over a dynamic scene. |
| Avoid list | Genres, cliches, emotional overstatement, cultural mismatches, temp-track traps, legal risks. |
| Handoff | Stems, bars/beats, loop point, fade handles, alt ending, ducking notes. |

Production heuristics:

- Spot silence deliberately. If music starts at frame 1 by default, the film may lose contrast.
- Score subtext rather than literal action when the scene already says the surface emotion.
- Put music exits on story changes, not only on scene cuts.
- Leave 6-12 frame edit handles for short stings and 1-2 seconds of pre/post-roll for longer music when the edit is still moving.
- In explainers, use fewer harmonic surprises under dense narration; motion can come from rhythm, pulse, and arrangement.
- In trailers and game trailers, define escalation in acts: identity, threat/problem, acceleration, peak, button.

Empirical observation to make after reviewing the edit: identify where music is fighting a spoken keyword, visual gag, product sound, or accessibility narration. Repair by thinning arrangement, moving the cue, changing register, reducing side energy, or using silence.

## Temp tracks without temp-love

Temp tracks are useful for communicating energy, density, contour, and editorial rhythm. They are dangerous when the team becomes attached to an uncleared track, a famous artist identity, or a specific melody.

When using a temp, translate it into attributes:

- "Tempo around 92 BPM, dry close drums, pulsing eighth-note synth bass, restrained harmony."
- "Female vocal sample feeling is desired, but no soundalike, lyric lift, melody lift, artist imitation, or copyrighted hook."
- "The cue should hit the title at 00:13:08, then drop to a half-time bed under narration."

Never ask an AI music generator or composer for a track "like" a named living artist, a specific copyrighted song, or a recognizable recording. Request musical attributes instead.

## Choose the right music source

### Original score

Use for bespoke emotional arc, recurring themes, precise sync, brand ownership, unusual structure, or edit flexibility. Specify:

- story premise and audience;
- cue list with timecodes and sync points;
- thematic material or leitmotifs: who/what gets a motif and when it changes;
- instrumentation and performance feel;
- harmonic language and emotional restraint;
- tempo map or pacing language;
- deliverable stems and cutdowns;
- revision priorities.

### Licensed needle drop

Use when the cultural meaning, lyric, artist voice, or era of an existing recording is essential. Require rights clearance before final use. Avoid using a recognizable temp as if clearance is guaranteed.

Documented fact, verified 2026-07-10: in U.S. copyright guidance, a musical work and a sound recording are separate copyright-protected works and are commonly owned/licensed separately. Source: U.S. Copyright Office, https://www.copyright.gov/engage/musicians/

Production implication: a commercial recording normally requires clearing the composition/publishing side and the master recording side. If the plan uses a cover or re-record, the master clearance changes, but composition rights still need attention. Consult qualified counsel or a music clearance professional for legal decisions.

### Library or platform-safe music

Use when budget, schedule, or licensing simplicity matters more than uniqueness. Check license scope: platform, media type, territory, duration, paid ads, organic posts, client work, broadcast, sublicensing, editing, and attribution.

Documented facts, verified 2026-07-10:

- YouTube's Audio Library is accessible through YouTube Studio and YouTube says its Audio Library music and sound effects are copyright-safe for YouTube videos, with attribution requirements depending on the license type. Source: YouTube Help, https://support.google.com/youtube/answer/3376882
- TikTok says its Commercial Music Library is pre-cleared for businesses to soundtrack TikTok content; TikTok also warns that the General Music Library is not pre-cleared for brand/business commercial content. Sources: https://ads.tiktok.com/help/article/commercial-music-library and https://ads.tiktok.com/business/en-US/blog/audio-library-royalty-free-music

Production implication: a track that is safe inside one platform library may not be safe for another platform, an offline ad, a TV buy, a trade-show booth, or a client's website. Do not assume portability.

### AI-generated music

Use when the project needs custom timing, many variations, no famous-artist dependency, rapid drafts, loops, or stem-based iteration. Treat provider terms and output rights as volatile.

Documented facts, verified 2026-07-10:

- The U.S. Copyright Office has an AI report series including copyrightability of generative AI outputs, with Part 2 published January 29, 2025. Source: https://www.copyright.gov/ai/
- ElevenLabs documents Music v2 capabilities such as section-based composition and transitions, and its API docs caution that seed-based consistency is not exact across system updates. Sources: https://elevenlabs.io/docs/overview/capabilities/music and https://elevenlabs.io/docs/api-reference/music/stream
- Suno's terms, as checked 2026-07-10, contain user-risk, non-infringement disclaimer, and non-unique-output language. Source: https://suno.com/terms-of-service

Production implication: verify the active provider plan, commercial scope, prohibited prompts, indemnity, attribution, retention, similarity, and output terms on the generation date. Preserve prompt, settings, seed if used, model/provider version if available, generation timestamp, and license/terms snapshot.

## AI music prompt anatomy

A good AI music prompt is a score brief with constraints. Include:

- deliverable: duration, instrumental/vocal, loop/sting/full cue, stems if supported;
- edit function: where it enters/exits and what it must hit;
- structure: sections with approximate timestamps;
- instrumentation: lead, rhythm, bass, texture, percussion, vocals/no vocals;
- performance: live, programmed, intimate, aggressive, restrained, imperfect, polished;
- tempo and meter when important;
- harmony/mode: bright, ambiguous, minor, modal, static, resolving, dissonant, consonant;
- mix: leave dialogue space, low-end restraint, no busy lead under VO, no harsh cymbals for mobile;
- negatives: no artist imitation, no recognizable melody, no lyrics unless cleared and requested, no sudden genre jump unless planned;
- export request: full mix plus stems/alt endings/loop points if the tool supports them.

Example: 30-second SaaS launch ad underscore

Intent: confident product launch under narration, not a hype trailer.

Prompt:

"Create a 30-second instrumental underscore for a B2B SaaS launch video. Structure: 0:00-0:05 quiet tactile pulse and soft piano motif for the problem statement; 0:05-0:18 add warm analog bass, muted kick, and subtle arpeggiated synth for workflow momentum; 0:18-0:26 lift with restrained strings and brighter harmony as the product UI resolves; 0:26-0:30 clean logo button with a short two-note motif and natural tail. Tempo 104 BPM, 4/4, modern premium minimal, warm, precise, optimistic but not childish. Keep 1-4 kHz sparse for narration intelligibility, no vocals, no whistling, no celebrity/artist references, no recognizable song imitation. Provide a full mix and, if possible, drums, harmonic bed, lead motif, and bass stems."

Why it works: it maps the edit arc, leaves room for voice, requests a reusable motif, and avoids risky soundalike language.

Example: documentary transition cue

Intent: move from archival loss to cautious present-day hope without manipulating the audience.

Prompt:

"Compose a 52-second instrumental documentary cue. Emotional arc: unresolved grief at the start, patient curiosity in the middle, small humane lift at the end. Instrumentation: felt piano, low cello harmonics, faint room tone-like granular texture, no drums until a very soft pulse after 0:32. Structure: 0:00-0:14 sparse piano notes with long decays; 0:14-0:32 cello enters and harmony stays suspended; 0:32-0:46 introduce a quiet heartbeat-like pulse; 0:46-0:52 cadence upward without full resolution for transition. Avoid melodrama, trailer impacts, heroic brass, and sentimental pop progressions. Mix should sit below interview dialogue and permit audio description gaps."

Why it works: it scores the ethical stance, not just the sadness.

Example: game trailer act map

Intent: 75-second action trailer with loopable gameplay bed and final logo hit.

Direction:

| Time | Music direction |
|---|---|
| 0:00-0:10 | World identity: processed bowed metal, distant choir pad, no drums. |
| 0:10-0:28 | Mechanics reveal: 140 BPM hybrid percussion, short bass ostinato, hits on UI ability reveals. |
| 0:28-0:48 | Combat escalation: add taiko-like low drums, distorted synth pulses, risers into match cuts. |
| 0:48-1:06 | Boss reveal: drop to half-time, brass/synth motif states antagonist interval. |
| 1:06-1:15 | Logo/button: full ensemble hit, 1-second silence, dry final sting. |

Music request:

"Create a 75-second instrumental hybrid action trailer cue following the act map above. No vocals. Keep the middle combat section loopable from 0:28-0:48 with a clean bar-aligned loop point. Avoid copyrighted franchise references, named composers, and recognizable game themes. Deliver full mix, percussion stem, synth/bass stem, tonal/orchestral stem, risers/hits stem, and final sting if supported."

## Cue sheets and metadata

Draft cue sheet data as the music plan becomes real. Do not wait until the final export if many cues are changing.

Documented fact, verified 2026-07-10: BMI describes cue sheets as a primary way PROs track music in film/TV and lists fields such as title, air date, music length, production company, cue title, composer, publisher, PRO, timing, and usage. Source: https://www.bmi.com/creators/detail/what_is_a_cue_sheet

Documented fact, verified 2026-07-10: ASCAP describes cue sheets as schedules of music in film/TV and points to RapidCue as a standardized digital submission format developed with BMI. Source: https://www.ascap.com/help/royalties-and-payment/cue-sheets

Minimum cue log for an AI agent to maintain:

| Field | Notes |
|---|---|
| Production title/version | Include episode/ad/cutdown variant if applicable. |
| Cue ID and title | Stable ID plus human-readable title. |
| Source | Original score, AI-generated, library, licensed master, public domain, client-supplied. |
| Writer/composer/publisher/PRO | Include "unknown/pending" rather than guessing. |
| Recording owner/master info | Required for needle drops and many libraries. |
| ISRC/ISWC/IPI/CAE | Include when available; do not invent. |
| In/out/duration | Use final picture timecode. |
| Usage type | Background, feature, theme, visual vocal, source/diegetic, logo, promo, etc. |
| Rights notes | License scope, attribution, restrictions, platform, term, territory. |
| Asset path/version | Exact file used in the final mix. |

## Mix and edit handoff

For each approved cue, hand off:

- WAV preferred for finishing, commonly 48 kHz for video workflows unless the project spec says otherwise;
- full mix plus stems: drums/percussion, bass, harmonic bed, lead/melody, vocals, risers/hits, ambience, stings;
- exact duration, BPM, meter, start bar, edit points, loop points, and ending tail;
- suggested fade in/out, ducking notes, and "do not cut before" points;
- loudness readout and true-peak readout after final mix;
- rights and attribution metadata.

Production heuristics for dialogue-first video:

- Keep lead melodies away from the primary speech band during narration, especially 1-4 kHz.
- Use arrangement ducking before heavy broadband compression when the cue feels crowded.
- For mobile/social, protect midrange clarity and avoid low-end-only musical information.
- For explainers, prefer short motifs and stable beds over song-like hooks that compete with learning.
- For documentaries, avoid music that instructs the viewer how to judge a real person unless that is an explicit ethical/directorial choice.

## Loudness, headroom, and platform delivery

Documented facts:

- ITU-R BS.1770-5 specifies algorithms for subjective programme loudness and true-peak signal level; verified 2026-07-10. Source: https://www.itu.int/rec/R-REC-BS.1770
- EBU R 128 recommends programme loudness normalization to -23.0 LUFS with generally +/-0.5 LU tolerance and a maximum permitted true peak of -1 dBTP for linear production; verified 2026-07-10. Source: https://tech.ebu.ch/docs/r/r128-2014.pdf

Production heuristics:

- Always follow the client/platform/broadcaster spec when one exists.
- For web/social drafts without a formal spec, deliver a final mix that leaves true-peak headroom, preserves speech intelligibility, and avoids over-limiting for artificial loudness. Many online services normalize playback, so crushing dynamics rarely improves the listener experience.
- Keep music stems with headroom so the editor can rebalance after voiceover, subtitles, or localization.
- For short ads and trailers, measure the finished programme, not only the music file. A loud music bed plus quiet voiceover can still fail intelligibility.

## Accessibility and inclusive audio

Documented facts, verified 2026-07-10:

- Section508.gov explains that captions include dialogue and other sounds, and are designed to enable participation for people who are deaf or have hearing loss. Source: https://www.section508.gov/create/synchronized-media/
- WCAG 2.1 Success Criterion 1.2.3 addresses audio description or a media alternative for prerecorded synchronized media; W3C explains that audio description adds important visual information to the audio. Source: https://www.w3.org/WAI/WCAG21/Understanding/audio-description-or-media-alternative-prerecorded.html

Production implications:

- If music communicates story information, represent relevant music or sound events in captions, e.g. "[tense strings build]" or "[music cuts out]" when meaningful.
- Do not place dense music under required audio description. If the video needs AD, spot AD pauses alongside music cues.
- Avoid relying on music alone to convey critical information such as danger, success, or button timing; pair with visual/textual cues when accessibility matters.

## Iteration and repair

When music is "wrong," diagnose before regenerating:

| Symptom | Likely cause | Repair |
|---|---|---|
| Feels cheesy | Over-explicit emotion, generic harmony, too many trailer devices | Reduce arrangement, change harmony, score subtext, use silence. |
| Fights VO | Busy lead, vocal layer, same frequency range as speech | Remove vocals/lead, lower density, duck around keywords, rewrite prompt for narration space. |
| Misses edit | No spotting map, wrong tempo, no sync points | Add timecoded hits and structure; request bar-aligned sections. |
| Sounds like a famous track | Temp-love or soundalike prompt | Translate to neutral attributes; change melody, instrumentation, rhythm, and references. |
| Loop clicks or drifts | Non-bar loop, reverb tail mismatch, tempo instability | Ask for loopable bed with clean bar boundary; crossfade; separate tail. |
| Rights unclear | Missing source/license data | Stop final use; request documentation or replace with cleared/original/generated music. |
| Mix pumps | Over-limiting or aggressive sidechain | Rebalance stems, lower limiter input, automate manually around dialogue. |

## Music QA checklist

Before final delivery, verify:

- The music plan matches the story arc and does not flatten emotional contrast.
- Every cue has source, rights notes, version, and timecode.
- No prompt or cue asks for a living artist imitation, copyrighted melody, uncleared lyric, or famous recording soundalike.
- Needle drops have documented clearance status for composition and master/recording sides, or are blocked from final use.
- Platform-library tracks are only used within their licensed scope.
- AI-generated tracks have provider, plan/terms snapshot date, prompt, settings, output ID/path, and usage restrictions recorded.
- Cue sheet fields are complete or marked pending; no metadata is invented.
- Full mix and stems are named clearly and match the final edit.
- Dialogue/narration remains intelligible on phone speakers and headphones.
- Captions include meaningful music/non-speech sound cues; audio description is not masked.
- Loudness and true peak are measured against the actual delivery spec.
- Cutdowns and localization versions have updated cue timings.

## Source list

Key sources independently checked for this skill:

- Guild of Music Supervisors, "What is a Music Supervisor," verified 2026-07-10: https://www.guildofmusicsupervisors.com/what-is-a-music-supervisor
- U.S. Copyright Office, "What Musicians Should Know about Copyright," verified 2026-07-10: https://www.copyright.gov/engage/musicians/
- U.S. Copyright Office, "Copyright and Artificial Intelligence," verified 2026-07-10: https://www.copyright.gov/ai/
- BMI, "What is a Cue Sheet?", verified 2026-07-10: https://www.bmi.com/creators/detail/what_is_a_cue_sheet
- ASCAP, "Cue Sheet Corner," verified 2026-07-10: https://www.ascap.com/help/royalties-and-payment/cue-sheets
- ITU-R BS.1770, verified 2026-07-10: https://www.itu.int/rec/R-REC-BS.1770
- EBU R 128, verified 2026-07-10: https://tech.ebu.ch/docs/r/r128-2014.pdf
- YouTube Help, "Use music and sound effects from the Audio Library," verified 2026-07-10: https://support.google.com/youtube/answer/3376882
- TikTok Business Help, "About the Commercial Music Library," verified 2026-07-10: https://ads.tiktok.com/help/article/commercial-music-library
- TikTok for Business, "Explore Royalty-Free Music in the Commercial Music Library," verified 2026-07-10: https://ads.tiktok.com/business/en-US/blog/audio-library-royalty-free-music
- Section508.gov, "Video and Other Synchronized Media," verified 2026-07-10: https://www.section508.gov/create/synchronized-media/
- W3C WAI, WCAG 2.1 Understanding SC 1.2.3, verified 2026-07-10: https://www.w3.org/WAI/WCAG21/Understanding/audio-description-or-media-alternative-prerecorded.html
- ElevenLabs Music documentation and API reference, verified 2026-07-10: https://elevenlabs.io/docs/overview/capabilities/music and https://elevenlabs.io/docs/api-reference/music/stream
- Suno Terms of Service, verified 2026-07-10: https://suno.com/terms-of-service

© calesthio · MIT · calesthio/generative-media-skills

Similar skills

All skills for Marketing and content →

From the same set149

The whole set →
cinematic-shot-direction The cinematic-shot-direction skill for Claude Code translates story intent into framing, shot size, optics, and camera movement for media production. CX Codex +1 328installs character-design-continuity character-design-continuity is a Claude Code skill for shaping recurring characters and keeping their identity consistent across generated media. It comes into play for character briefs, reference sets, and continuity checks across scenes. CX Codex +1 326installs storyboard-previsualization Storyboard-previsualization is a Claude Code skill that builds provider-independent storyboards and previsualization plans for video, film, and media projects. CX Codex +1 301installs title-kinetic-typography title-kinetic-typography is a Claude Code skill for planning animated text in video, ads, and title sequences. It covers hierarchy, legibility, timing, and safe placement. CX Codex +1 277installs social-short-production The social-short-production skill for Claude Code plans, produces, adapts, and finishes platform-ready short vertical social videos. CX Codex +1 263installs generated-media-qa The generated-media-qa skill for Claude Code runs quality assurance checks on AI-generated and AI-assisted media assets. Runs when reviewing, accepting, and reporting on images, video, and audio. CX Codex +1 261installs ffmpeg-media-finishing ffmpeg-media-finishing is a Claude Code skill for preparing video, audio, and images for delivery. It steps in during final export and quality checks. CX Codex +1 258installs still-image-retouching-finishing The still-image-retouching-finishing skill for Claude Code manages photo post-production and prepares images for print, web, or archive delivery. CX Codex +1 251installs

Project sponsors

Companies that keep this analytics open