Autosound Tuning Skill
ayukhno/autosound-tuning-skill
Otimizador de áudio automotivo via Claude + REW com método estruturado
Ver no GitHubSkill Claude para otimizar áudio automotivo integrado com REW, utilizando um fluxo de revisão Generator↔Critic↔Arbiter. Encapsula método portável (intake, crossovers, fase, EQ, staging, voicing) compatível com qualquer carro e qualquer DSP, mantendo segurança dos dados do usuário.
Instalação
npx convez-ai ayukhno/autosound-tuning-skill1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Ajuste de EQ e crossovers em sistemas de áudio automotivo
- Análise e revisão estruturada de medições REW em loop colaborativo
- Documentação e acumulação de conhecimento sobre setups de DSP específicos
Veredito

Skill altamente especializado para nicho técnico (tuning de áudio automotivo). Implementa metodologia robusta com integração REW e orquestração de revisão multi-agente, exigindo conhecimento aprofundado do domínio e acesso a Claude Code.
README
A tradução para o português deste README ainda está na fila.
Autosound Tuning Skill
🇬🇧 English · 🇩🇪 Deutsch · 🇵🇱 Polski
In one line: a Claude skill that helps you tune car audio — it reads your REW measurements, analyzes them, recommends settings, and can push EQ back into REW.
- 📊 Works with REW over its API — pulls measurements, loads EQ filters back in
- 🎯 Knows the craft — target curves, tuning practices, and a built-in step-by-step tuning process
- 🎧 Test tracks — what to listen for and on which track (the descriptions, not the audio)
- 🚗 Learns your setup — accumulates car & gear knowledge from a tuning angle (only with your consent)
A Claude skill for tuning a car-audio DSP with REW — from a brand-new project (equipment + goals interview, target-curve choice, install verification) through crossovers, time alignment, phase, per-channel and summed EQ, imaging/staging, to client-preference voicing. It encodes a measured method (Generator ↔ Critic ↔ Arbiter review loop) plus a growing library of hard-won diagnostic techniques.
The skill is method, not machine. No car's measurements or DSP state live here — those stay in the tuner's own project. This repo ships the reusable process so it works on any car / any DSP.
Built and battle-tested on a VW Passat B8 / Helix DSP Ultra S (an AYA competition win), but the car/DSP specifics are isolated into a profile + a
knowledge/library so a new project is just "pick the inputs."
What's in here
skills/
├── autosound-tuning/ the main skill
│ ├── SKILL.md entry point (process map, session lifecycle, roles)
│ ├── references/ on-demand docs: phases, diagnostics, EQ patterns,
│ │ filter types, staging/depth, competition, test tracks,
│ │ voicing-by-ear, REW API, Helix specifics, intake, feedback
│ ├── knowledge/ accumulated car & DSP profiles (cars/, dsp/)
│ └── scripts/ example Critic-channel implementation (optional rig tooling)
└── review-loop/ sibling skill: independent-review orchestration (vendor-agnostic)
The two skills are a pair — autosound-tuning references review-loop for the review protocol. Install both.
Getting started — new to Claude Code / the terminal?
This skill runs in Claude Code, a terminal tool. If you've only used desktop/web chat before, here's the one-time setup. (Other languages: Deutsch · Polski.)
Why Claude Code (not the web app / Cowork): tuning is iterative — measure → analyze → change → re-measure — over text/numeric data (REW CSV/text exports, DSP configs, iteration logs). A terminal + git gives you parsing, scripting, and a full change history. Cowork is oriented to file operations in other apps — overkill here.
1. Install Claude Code (macOS / zsh; needs Node.js 18+ and a paid Claude plan — Pro / Max / Team / Enterprise; the free plan won't work):
node --version # Node.js 18 or newer
mkdir -p ~/.npm-global && npm config set prefix '~/.npm-global'
echo 'export PATH="$HOME/.npm-global/bin:$PATH"' >> ~/.zshrc && source ~/.zshrc
npm install -g @anthropic-ai/claude-code
claude --version # verify
2. Create a project folder and launch:
mkdir ~/car-audio-setup && cd ~/car-audio-setup
claude # first run opens the browser to log in
3. Don't over-structure. Start with an empty folder. Drop in your first REW export (.txt/.csv) as-is, keep one markdown notes log (append-only), and let folders (measurements/, configs/, eq-log/) appear when you actually need them.
4. Your first message = "step 0". Describe your system in one message: car + speaker layout (how many ways, sub?), DSP model (Helix / Audison / other), measurement gear (mic, interface), and whether you already have measurements or are starting from the hardware. The skill takes it from there — and first asks your preferred working language (English / your native, if supported: EN · UK · DE · PL) so the conversation and your project files come out in it. Then: intake → signal-chain check → mic calibration → first sweep → …
(Then install the skill itself — next section.)
Requirements
- Claude Code (or claude.ai with skills).
- REW with the API server enabled (
localhost:4735), a calibrated measurement mic. - Your DSP's software and a way to load EQ into it (file import ideal; for 30+ DSPs without import, see REW-EQ-CopyPaste-Assistant).
- Optional: a second AI model as the "Critic" (any vendor — the roles are vendor-agnostic). Without it, a human plays the reviewer; the process still holds.
Install
Easiest — let Claude install it for you. Open Claude Code and just ask, for example:
"Clone https://github.com/ayukhno/autosound-tuning-skill into a temporary folder, then copy the two inner skill folders —
skills/autosound-tuningandskills/review-loop— into my user-level~/.claude/skills/, so each skill'sSKILL.mdends up directly at~/.claude/skills/<name>/SKILL.md. Don't clone the whole repo into the skills folder."
Claude places both skills (they're a pair) where you choose:
- User-level —
~/.claude/skills/(recommended): available in every folder you open Claude Code in. Pick this if you'll tune more than one car or want it always on hand. - Project-level —
<your-project>/.claude/skills/: only inside that one project. Pick this to keep a single car's repo self-contained.
⚠️ Common install mistake: don't
git clonethe repo into~/.claude/skills/autosound-tuning/. That nestsSKILL.mdone level too deep (…/autosound-tuning/skills/autosound-tuning/SKILL.md) and Claude Code reportsUnknown skill: autosound-tuning(or never offers the skill at all). Always copy the innerskills/*folders soSKILL.mdsits directly under the skill folder.
Or do it manually:
git clone https://github.com/ayukhno/autosound-tuning-skill.git
# user-level (available everywhere) — copy the INNER skill folders:
cp -R autosound-tuning-skill/skills/autosound-tuning ~/.claude/skills/
cp -R autosound-tuning-skill/skills/review-loop ~/.claude/skills/
# …or project-level: cp -R the same two folders into your-project/.claude/skills/
# verify — each line must print the path (SKILL.md sits DIRECTLY under the skill folder):
ls ~/.claude/skills/autosound-tuning/SKILL.md
ls ~/.claude/skills/review-loop/SKILL.md
After installing, start a fresh Claude Code session — skills are loaded at startup.
Troubleshooting — Unknown skill / the skill never triggers: almost always the nesting above. If ls ~/.claude/skills/autosound-tuning/SKILL.md fails but …/autosound-tuning/skills/autosound-tuning/SKILL.md exists, you cloned the whole repo into the skill folder. Fix it, then restart Claude Code:
mv ~/.claude/skills/autosound-tuning ~/.claude/skills/autosound-tuning-repo
cp -R ~/.claude/skills/autosound-tuning-repo/skills/autosound-tuning ~/.claude/skills/
cp -R ~/.claude/skills/autosound-tuning-repo/skills/review-loop ~/.claude/skills/
rm -rf ~/.claude/skills/autosound-tuning-repo
Then open Claude Code in your project and say e.g. "tune a new car from scratch" — the skill starts with intake (references/project-intake.md): quickstart, equipment + goals interview, target-curve choice (no default — chosen with you), install verification, and project-file generation. (It first asks your preferred working language — you can run the whole local project in your own language; English here is just the skill's internal method.)
Contributing your experience
The skill learns from every tune — and it gathers that feedback right in the terminal, while you work, not via a form to fill in. At wrap-up (Phase 7) it runs a short closing ritual: asks what actually helped, what was off, and any DSP/car quirk you hit — then, with your explicit consent, offers to share the generalizable lessons.
What the final survey is for, and what it collects: to grow the shared method + the knowledge/ library. It captures method + equipment classes only — the car body/cabin behavior, the DSP/processor and gear class, and which techniques worked. Never personal data, never full measurements. You see exactly what would be shared and opt in (or out) per item.
Confirmed lessons get folded into the skill and the knowledge/ car/DSP profiles (with attribution). Tips that contradict existing findings are kept as variants, not deletions — a different cabin may make them right.
(Prefer GitHub? You can still open a field-feedback issue — same rule: method/equipment classes only.)
License
CC BY-SA 4.0 — use it, adapt it, share it; keep derivatives open and give attribution. It's a method/knowledge work, so share-alike keeps the community's accumulated experience open.