Mix and master by measurement: sdsp-mix
An agent mixes the way an engineer would: render, measure against a commercial reference, change one thing. Mix map of who masks whom, kick and bass ducking, mastering to a loudness target.
Use the sdsp-mix skill on {mix.wav}. 1. Measure first with tools/mixcheck.py against this reference in the same style: {reference.wav}. Report crest factor, band deltas, stereo correlation and the loudness of each section. 2. Say what is wrong in numbers (a build louder than its drop, a band 4 dB off the reference, correlation under +0.2). 3. Change ONE thing, render again, measure again, and show me the before and after numbers. Do not guess settings by name; every change needs the reading that justified it.
In Claude Code, paste the prompt as a message.
Install
npx skills add fortunto2/superduper-dsp --skill sdsp-mix# or the whole set as a Claude Code plugin (commands then look like /superduper-dsp:sdsp-mix):
claude plugin marketplace add https://github.com/fortunto2/superduper-dsp
claude plugin install superduper-dsp@superduper-dsp --scope userSteps
- Install the skill (npx line or the plugin). It calls tools/mixcheck.py from the superduper-dsp repo, so clone the repo or use the plugin's cache folder, and have Python 3 with numpy.
- Render your mix to a WAV and pick a commercial reference in the same genre: absolute band targets differ by up to 10 dB between genres.
- Paste the prompt. Read the first report before you let the agent change anything.
- For a live session, the skill also describes a mix map (who masks whom) that reads level and spectrum from the SuperDuper plugins while the DAW plays.
Tips
- Always pass a reference. A 'dark' or 'harsh' verdict is a band delta of 4 dB or more against it.
- Healthy crest factor is 10 to 13 dB. Under 8 means the chain flattened the performance.
- Wants the SuperDuper plugins for the live bus and mix map; the measuring half (mixcheck.py) works on any WAV.
| works with | Claude Code: Runs as a skill or a prompt in the terminal; can read your files and write the result. |
| cost | Free. No API, no key. |
| status | Unverified: written from the technique, not yet run end to end by us. |
| source | sdsp-mix skill in superduper-dsp (MIT) |
| license | CC BY 4.0 for the text here; skills are MIT. |
| updated | 2026-11-10 |
| for agents | sdsp-mix-measure-then-mix-and-master.md |
Related
Nine agent skills for mixing, mastering and making music (superduper-dsp)
A Claude Code plugin that teaches an agent to mix by measurement, build songs in REAPER, render plugin chains headless, turn audio into MIDI and tear down a reference track.
Level every chapter to -16 LUFS with ffmpeg loudnorm
One ffmpeg line that brings a file to -16 LUFS integrated and -1.5 dBTP true peak without touching the video, the two-pass version for exact results, and the timestamp trap on upload.
Tear down a reference track by numbers: tempo, key, groove, balance
The agent reads a spectrogram instead of listening, splits stems with Demucs, fits tempo and phase to the kick, and rebuilds the vibe until the numbers match the reference.
Compare two microphones by numbers, not impressions: mic-compare
Record the same phrase on two mics at matched gain and let the agent report response, noise floor, per-band SNR, plosives and sensitivity, and what EQ can still fix.
Render plugin chains from a text file, no DAW: sdsp-chain
One TOML file describes tracks, effects, sidechains, automation and playable instruments; a Rust command renders it to WAV and prints LUFS and true peak per stage.
A vocal chain slot by slot: plosives, gate, compress, air, de-ess
The vocal-mix skill lays out a nine-slot channel strip in REAPER with the reasoning for the order. The numbers are tuned to one voice and one mic; the order carries over.