generating-sounds-with-ai
raphaelsalaja/skill · Agent Skill
進行Web Audio API程式碼審查,以確保程式設計音效合成的品質、使用者體驗的決策和最佳實踐引數的遵守。
前往安裝 ↓總覽
| 發布者 | |
|---|---|
| 儲存庫 | raphaelsalaja/skill ↗ |
| 技能資料夾 | generating-sounds-with-ai |
| 星 | 42 |
| 分支 | 5 |
| 附帶檔案 | 1 |
| 授權 | MIT |
| 最近更新 | 2026年3月17日 |
| 連結 | 技能原始碼 ↗ · 儲存庫 ↗ |
Mozgi 檢查
我們在技能檔案中發現的內容。詳細資訊需訂閱。
安裝
技能就是一個含 SKILL.md 的資料夾。選擇你的代理,在專案根目錄的終端機(macOS、Linux 或 Windows 的 Git Bash)執行指令。
把提示貼到 Claude、Codex 或其他助理:它會自行檢查並安裝技能。
Install the “generating-sounds-with-ai” skill from mozgi.io (https://mozgi.io/skills/raphaelsalaja-generating-sounds-with-ai/) for me. 1. Download the archive: https://mozgi.io/skills/raphaelsalaja-generating-sounds-with-ai/download If it fails, take the skill folder from the source: https://github.com/raphaelsalaja/skill/tree/main/skills/generating-sounds-with-ai 2. Read SKILL.md and the other files. Check the skill is safe: no hidden commands, no sending data elsewhere, no scripts doing something other than described. If anything looks suspicious, stop and tell me. 3. Put the “generating-sounds-with-ai” folder into your agent's skills folder: .claude/skills/ for Claude Code, .agents/skills/ for Codex, .cursor/skills/ for Cursor, .gemini/skills/ for Gemini CLI. Inside the project if we are in one, otherwise in the home folder (~/.claude/skills/ etc.). 4. Make sure <skills folder>/generating-sounds-with-ai/SKILL.md exists. 5. Briefly tell me what this skill does and offer to try it on my task. If you cannot do a step yourself, tell me what to click or type.
git clone --depth 1 --branch main https://github.com/raphaelsalaja/skill.git /tmp/skill mkdir -p .claude/skills cp -r /tmp/skill/skills/generating-sounds-with-ai .claude/skills/generating-sounds-with-ai
重新啟動 Claude Code — 技能就會出現在清單中,手動呼叫:/generating-sounds-with-ai。要套用到所有專案,請放在 ~/.claude/skills。
git clone --depth 1 --branch main https://github.com/raphaelsalaja/skill.git /tmp/skill mkdir -p .agents/skills cp -r /tmp/skill/skills/generating-sounds-with-ai .agents/skills/generating-sounds-with-ai
Codex 會從目前資料夾往上到儲存庫根目錄,在 .agents/skills 中尋找技能;用 /skills 或 $generating-sounds-with-ai 呼叫。所有專案:~/.agents/skills。
git clone --depth 1 --branch main https://github.com/raphaelsalaja/skill.git /tmp/skill mkdir -p .cursor/skills cp -r /tmp/skill/skills/generating-sounds-with-ai .cursor/skills/generating-sounds-with-ai
Cursor 啟動時會載入技能;在 Agent 聊天輸入「/」呼叫。Cursor 也會讀取 .claude/skills 與 .agents/skills。所有專案:~/.cursor/skills。
git clone --depth 1 --branch main https://github.com/raphaelsalaja/skill.git /tmp/skill mkdir -p .gemini/skills cp -r /tmp/skill/skills/generating-sounds-with-ai .gemini/skills/generating-sounds-with-ai
任務符合時,Gemini CLI 會提議啟用技能,由你確認。所有專案:~/.gemini/skills。
安裝前請先檢視 SKILL.md 與相關檔案。點擊檔案即可預覽。
SKILL.md 內容
作者原文 · 唯讀Generating Sounds with AI
Review Web Audio API code for sound synthesis best practices.
How It Works
1. Read the specified files (or prompt user for files/pattern)
2. Check against all rules below
3. Output findings in file:line format
Rule Categories
| Priority | Category | Prefix |
|----------|----------|--------|
| 1 | Context Management | context- |
| 2 | Decay & Envelope | envelope- |
| 3 | Sound Design | design- |
| 4 | Parameters | param- |
Rules
Context Management Rules
context-reuse-single
Reuse a single AudioContext instance; do not create new ones per sound.
Fail:
function playSound() {
const ctx = new AudioContext();
// Creates new context every call
}
Pass:
let audioContext: AudioContext | null = null;
function getAudioContext(): AudioContext {
if (!audioContext) {
audioContext = new AudioContext();
}
return audioContext;
}
context-resume-suspended
Check and resume suspended AudioContext before playing.
Fail:
function playSound() {
const ctx = getAudioContext();
// Plays immediately without checking state
}
Pass:
function playSound() {
const ctx = getAudioContext();
if (ctx.state === "suspended") {
ctx.resume();
}
}
context-cleanup-nodes
Disconnect and clean up audio nodes after playback.
Fail:
source.start();
// Nodes remain connected after sound ends
Pass:
source.start();
source.onended = () => {
source.disconnect();
gain.disconnect();
};
Envelope Rules
envelope-exponential-decay
Use exponential ramps for natural decay, not linear.
Fail:
gain.gain.linearRampToValueAtTime(0, t + 0.05);
Pass:
gain.gain.exponentialRampToValueAtTime(0.001, t + 0.05);
envelope-no-zero-target
Exponential ramps cannot target 0; use 0.001 or similar small value.
Fail:
gain.gain.exponentialRampToValueAtTime(0, t + 0.05);
Pass:
gain.gain.exponentialRampToValueAtTime(0.001, t + 0.05);
envelope-set-initial-value
Set initial value before ramping to avoid glitches.
Fail:
gain.gain.exponentialRampToValueAtTime(0.001, t + 0.05);
// No setValueAtTime before ramp
Pass:
gain.gain.setValueAtTime(0.3, t);
gain.gain.exponentialRampToValueAtTime(0.001, t + 0.05);
Sound Design Rules
design-noise-for-percussion
Use filtered noise for clicks/taps, not oscillators.
Fail:
// Click sound using sine oscillator
const osc = ctx.createOscillator();
osc.type = "sine";
// Results in tonal "beep" not "click"
Pass:
// Click sound using noise burst
const buffer = ctx.createBuffer(1, ctx.sampleRate * 0.008, ctx.sampleRate);
const data = buffer.getChannelData(0);
for (let i = 0; i < data.length; i++) {
data[i] = (Math.random() * 2 - 1) * Math.exp(-i / 50);
}
design-oscillator-for-tonal
Use oscillators with pitch movement for tonal sounds (pops, confirmations).
Fail:
// Confirmation sound using static frequency
osc.frequency.value = 400;
Pass:
// Confirmation sound with pitch sweep
osc.frequency.setValueAtTime(400, t);
osc.frequency.exponentialRampToValueAtTime(600, t + 0.04);
design-filter-for-character
Apply bandpass filter to shape percussive sounds.
Fail:
// Raw noise without filtering
source.connect(gain).connect(ctx.destination);
Pass:
const filter = ctx.createBiquadFilter();
filter.type = "bandpass";
filter.frequency.value = 4000;
filter.Q.value = 3;
source.connect(filter).connect(gain).connect(ctx.destination);
Parameter Rules
param-click-duration
Click/tap sounds should be 5-15ms duration.
Fail:
const buffer = ctx.createBuffer(1, ctx.sampleRate * 0.1, ctx.sampleRate);
// 100ms is too long for a click
Pass:
const buffer = ctx.createBuffer(1, ctx.sampleRate * 0.008, ctx.sampleRate);
// 8ms is appropriate for a click
param-filter-frequency-range
Bandpass filter for clicks should be 3000-6000Hz.
Fail:
filter.frequency.value = 500; // Too low, sounds muffled
Pass:
filter.frequency.value = 4000; // Crisp, present
param-reasonable-gain
Gain values should not exceed 1.0 to prevent clipping.
Fail:
gain.gain.setValueAtTime(1.5, t);
Pass:
gain.gain.setValueAtTime(0.3, t);
param-q-value-range
Filter Q for clicks should be 2-5 for focused but not harsh sound.
Fail:
filter.Q.value = 15; // Too resonant, harsh
Pass:
filter.Q.value = 3; // Focused but natural
Output Format
When reviewing files, output findings as:
file:line - [rule-id] description of issue
Example:
lib/sounds.ts:23 - [envelope-exponential-decay] Using linearRampToValueAtTime instead of exponential
lib/sounds.ts:45 - [context-reuse-single] Creating new AudioContext on each call
Summary Table
After findings, output a summary:
| Rule | Count | Severity |
|------|-------|----------|
| context-reuse-single | 1 | HIGH |
| envelope-exponential-decay | 3 | MEDIUM |
| param-click-duration | 1 | LOW |
Parameter Translation Table
When user describes issues, translate to parameter changes:
| User Says | Parameter Change |
|-----------|------------------|
| "too harsh" | Lower filter frequency, reduce Q |
| "too muffled" | Higher filter frequency |
| "too long" | Shorter duration, faster decay |
| "cuts off abruptly" | Use exponential decay |
| "more mechanical" | Higher Q, faster decay |
| "softer" | Lower gain, triangle wave |
References
常見問題
generating-sounds-with-ai 技能有什麼用途?
進行Web Audio API程式碼審查,以確保程式設計音效合成的品質、使用者體驗的決策和最佳實踐引數的遵守。
如何安裝 generating-sounds-with-ai?
打開本頁的「安裝」區塊並選擇你的代理 — Claude Code、Codex、Cursor 或 Gemini CLI:那裡有把技能資料夾複製到正確位置的現成指令。也可以複製「透過助理」分頁中的提示,讓助理自行檢查並安裝技能。
generating-sounds-with-ai 適用哪些 AI 模型?
技能是純 Markdown 指示(Agent Skills 格式),因此 generating-sounds-with-ai 適用任何讀取 SKILL.md 的代理:Claude Code、Codex、Cursor、Gemini CLI、GitHub Copilot、OpenCode 等。只有在任務符合技能描述時,模型才會載入指示。
generating-sounds-with-ai 是免費的嗎?
此技能由 raphaelsalaja 公開發布,採用 MIT 授權。技能本身免費 — 你只需向 AI 供應商支付實際使用的 token 費用。
安裝技能安全嗎?
技能是代理會遵循的指示,安裝前請閱讀 SKILL.md 及相關檔案。技能資料夾中的腳本只有在代理執行時才會運行 — 留意它的動作,不要授予多餘權限。
技能和 MCP 伺服器或外掛有何不同?
MCP 伺服器或外掛為模型提供新工具 — 會執行並回傳結果的程式碼。技能提供的是知識與流程:如何處理任務、要遵循哪些步驟、好的成果長什麼樣子。它不需要伺服器或 API 金鑰。