agentleFS
Razee4315Razee4315/coldframe/.claude/skills/sound-design/SKILL.md

sound-design

Skill4 starsChanged 39 days agoMITView on GitHub

Sound for product, launch and social videos in Remotion - a soft music bed, a real sound effect for every click, keystroke, toggle, success and transition, voiceover if asked, mixing levels and loudness. Use whenever a video gets audio, together with video-rules and motion-direction.

12 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

4 rules. 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 10 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Keep every cue frame in the same config object as the visuals (scene-relative frames), so sound and picture can't drift.

    line 90
  2. Very specific:

    Start with volume 1 for effects and 0.3–0.4 for the music bed, then adjust by measuring (§6).

    line 91
  3. Distinctive:

    No clipping, no pumping, no sound effect louder than the one impact moment.

    line 102
  4. Very specific:

    Only then render the video in the cloud with coldframe.

    line 108

“Common” is measured: kept by at least 6 in 10 of the 10 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Sound design

Sound is half of what makes a film feel expensive. Every visible action gets a sound that matches it, a soft music bed carries the whole film, and the mix is clean and loud enough without clipping.

Never synthesise the soundtrack in code (numpy noise, sine pads, beeps). It sounded bad and was cut from the coldframe launch film. Use real recorded sound effects and real music.

1. Layers

  1. Music bed (always, unless the deliverable is silent): soft, instrumental, steady tempo, no vocals. Ambient/electronic/soft piano for dev tools; it sits under everything.
  2. UI sound effects: one for every visible interaction.
  3. Transition sound effects: whooshes, swishes, risers, impacts for camera moves and cuts.
  4. Voiceover (only if the user asks): recorded or high-quality TTS the user approves; always with burned-in captions.

2. Sources

WhatWhereNotes
UI + transition SFXpublic/sfx/: the coldframe library (§5)CC0 (Kenney, Remotion @remotion/sfx); safe to commit and redistribute
More SFXKenney audio packs (CC0), @remotion/sfx (per-sound license on its page), Freesound (filter CC0)Record the source + license in library.json
MusicThe user picks: Pixabay Music (free, no attribution), Chosic CC0 filter, YouTube Audio Library, or a track they ownOffer 2–3 options that fit the mood; don't commit Pixabay/YouTube tracks to a public repo (their licenses forbid redistributing the file itself); keep them in the project's public/music/ of a private repo or add to .gitignore and upload for the render
AI-generated SFX / music / voiceElevenLabs (sound effects, music, TTS) or similar, only with the user's API key and OKListen and approve before use

3. Match each sound to its action

Pick by what the element is, how heavy it is, and what just happened. Never reuse the exact same sample twice in a row: alternate variants and vary playbackRate 0.96–1.04 and gain ±2 dB.

On screenSoundLevel vs. music
Cursor click on a small control (tab, link, icon)soft click / ticksubtle, +3 dB
Primary button press (Run, Render, Deploy, CTA)firmer click with a little body+6 dB
Toggle / switch / checkboxswitch (two-state), on ≠ off variant+4 dB
Typing in a terminal/inputkey taps, one per 1–3 characters, varied; slightly louder on Entersubtle
Hover / focus / selection movingvery soft tick, or nothingbarely audible
Card or panel lands / snaps into placesoft drop / tap+4 dB
Notification, badge, counter completespop / soft bell+5 dB
Success / verified / doneconfirm chime (the one "reward" sound)+6 dB
Error / failure (only if shown)low muted buzz, short+4 dB
Camera move, fly-in, whip panwhoosh / swish whose peak lines up with the fastest frame+4 dB
Palette impact / big revealimpact (low hit), with a riser before it if one has been added to the library+8 dB, the loudest moment
Logo / end cardsoft hit + the music's resolving phrase+6 dB

Group sounds so the film has 3–4 recognisable families (clicks, whooshes, one reward chime, one impact), not 30 random ones.

4. Sync

  • Clicks and taps: on the frame the button visibly depresses (or 1 frame before; sound feels late otherwise).
  • Whooshes: start early so the peak lands on the fastest frame of the move.
  • Impacts: exactly on the impact/flash frame. Leave 3–8 frames of near-silence before a big reveal (pull the music down).
  • Cut on the music's beat where you can: pick the bpm and snap scene starts to beats (frames per beat = fps * 60 / bpm).

5. Build it in Remotion

coldframe setup installs the library in public/sfx/ (37 CC0 sounds in 17 roles, trimmed so they start on the frame, loudness-matched; sources in public/sfx/LICENSE.md, details in library.json) and the helpers in src/coldframe-sound.tsx.

RoleUse
click-softclick on a small control (tab, link, icon)
click-firmpress on a primary button (Run, Render, CTA)
tickhover / focus / selection moving
key, key-entertyping; Enter at the end of a command
toggle-on, toggle-offswitches, checkboxes
selectoption picked in a list
dropcard/panel lands and snaps into place
popnotification, badge, small item appears
confirmsuccess / verified / done (the one reward sound)
errora failure shown on screen
open, closepanel/window expands or collapses (peak-aligned)
whooshcamera move, fly-in, whip pan (peak-aligned)
impact-softelement lands with weight, beat accent
impactthe one big moment: palette impact, reveal, logo
import { MusicBed, Sfx, typingCues } from "./coldframe-sound";

<MusicBed src="music/bed.mp3" volume={0.35} duckAt={[REVEAL]} />
<Sfx role="click-firm" at={PRESS} />                 // the frame the button visibly depresses
<Sfx role="whoosh" at={FASTEST_FRAME} />             // peak lands on `at` automatically
{typingCues(TYPE_START, TYPE_END, cmd.length).map((f) => <Sfx key={f} role="key" at={f} />)}
<Sfx role="key-enter" at={ENTER} />
<Sfx role="confirm" at={RESULT} />
<Sfx role="impact" at={REVEAL} />
  • Variants rotate and playbackRate varies ±3% automatically, so repeated clicks never sound identical.
  • Keep every cue frame in the same config object as the visuals (scene-relative frames), so sound and picture can't drift.
  • Start with volume 1 for effects and 0.3–0.4 for the music bed, then adjust by measuring (§6).
  • Audition the whole library: the example project's SfxAudition composition (npx remotion render SfxAudition out/sfx-audition.wav --codec=wav).
  • Need a sound the library doesn't have (riser, deep sub hit, ambience)? Add a CC0 file to public/sfx/, note its source in LICENSE.md, and ask the user to approve it.

coldframe's chunks carry their audio slice as lossless PCM, so many overlapping <Audio> tags render and join sample-exactly in the cloud.

6. Levels and loudness

  • Final mix: about -14 LUFS integrated (YouTube, social), true peak ≤ -1 dBTP. Web hero/README versions ship silent (strip audio: ffmpeg -i in.mp4 -an -c:v copy out.mp4).
  • Music bed sits well under effects (roughly -24 to -20 LUFS on its own); duck it 6–10 dB under voiceover and before the big reveal.
  • Measure: ffmpeg -i out.mp4 -af loudnorm=print_format=json -f null - (integrated, true peak) and -af volumedetect (peak). Use the system ffmpeg; Remotion's bundled one lacks these filters.
  • No clipping, no pumping, no sound effect louder than the one impact moment.

7. Approval before the full render

  1. Render the audio only, locally (cheap, no video frames): npx remotion render <Comp> out/audio-preview.wav --codec=wav. For long films, cut a 10–15 s sample around the busiest part.
  2. Send it to the user with a one-line cue list ("0:03 fly-in whoosh, 0:07 click, 0:09 success chime…") and wait for their OK.
  3. Only then render the video in the cloud with coldframe.

8. Checklist

  • Music bed chosen by the user (2–3 options offered), license noted, fades in and out cleanly.
  • Every visible click, keystroke, toggle, landing, success and camera move has a matching sound; nothing visible happens in silence unless on purpose.
  • 3–4 sound families, variants rotated, no identical repeats back to back.
  • Impacts on the frame, whoosh peaks on the fastest frame, clicks on the press frame.
  • ~-14 LUFS integrated, true peak ≤ -1 dBTP, no clipping (measured).
  • Audio sample approved by the user before the cloud render.
  • Silent version exported for README / website hero.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.