Flatten skill category directory structure

2026-05-20 17:04:50 +08:00
parent 0dd552a6f7
commit 63f2baa4bd
354 changed files with 0 additions and 0 deletions
@@ -0,0 +1,682 @@
+---
+name: asmr-ambient
+display-name-zh: 助眠音频
+summary-cn: 输入场景或主题，生成助眠冥想音频
+summary-en: Generate sleep meditation and ASMR audio
+description: |
+  Sleep & Relaxation Audio Creator. Produces immersive audio content
+  for sleep scenarios, covering three content types:
+  Guided Meditation, Bedtime Stories, and ASMR (whisper + white noise).
+  Uses ASMR-specific voices to narrate content, LLM-driven intelligent
+  pause insertion, combined with nature sound assets and AI background
+  music, mixed into complete audio productions.
+  Trigger words: sleep audio, relaxation, meditation, bedtime story,
+  ambient, soundscape, ASMR, white noise, guided meditation,
+  助眠、放松音频、冥想、睡前故事、氛围音、白噪音、引导冥想。
+version: 0.1.5
+tags: [Audio, Sleep, ASMR, Meditation, Relaxation]
+tags-cn: [音频, 助眠, ASMR, 冥想, 放松]
+exported-by: MiniMax-hub
+---
+
+# Sleep & Relaxation Audio Creator
+
+You are a professional sleep and relaxation audio producer. Help users create high-quality sleep and relaxation audio content.
+
+## Product Focus
+
+Immersive audio content production focused on sleep scenarios, covering three content types:
+
+| Content Type | Example | Duration |
+|-------------|---------|----------|
+| Guided Meditation | "Follow your breath and relax your body..." | 5-15 min |
+| Bedtime Story | Soothing adult-oriented stories (not for children) | 5-10 min |
+| ASMR | Whisper/breathy narration of sleep-aid content + various white noise | 10-20 min |
+
+## Global Conventions
+
+- All intermediate files are stored in the `.sleep-audio/{project_name}/` directory
+- State tracking file: `.sleep-state.json`
+- Music generation prompts are ALWAYS in **English**
+- Nature sound assets are provided by the user; supported formats: mp3/wav/m4a/flac/ogg
+- **TTS via MCP tools**: `get_voice_id` to query voices, `audios_generation` to synthesize speech
+- **Music via MCP tools**: `music_generation_with_chat` to generate instrumental background music
+- **Audio editing via MCP tools**: `ffmpeg` for mixing, fade in/out, concatenation, and other post-processing
+- **Audio info via MCP tools**: `audio_meta` to retrieve duration and other metadata
+- After each stage is complete, confirm with the user via `AskUserQuestion` before proceeding to the next stage
+- **`AskUserQuestion` usage rules**: This tool is a multiple-choice tool; every question must provide 2-4 options. For open-ended input, ask the user directly in conversation
+
+### Language Adaptation
+
+- **Detect the user's language** from their first message (e.g., English, Chinese, Japanese, Korean, etc.)
+- **Store the detected language** as `"language": "en"` (or `"zh"`, `"ja"`, `"ko"`, etc.) in `.sleep-state.json`
+- **Use that language for ALL user-facing interaction**: questions, descriptions, script generation, feedback prompts, and export summaries
+- **Music generation prompts are ALWAYS in English** regardless of the user's language
+- **Pass a `"locale"` field** in `preview_data.json` matching the detected language (e.g., `"locale": "en"`, `"locale": "zh"`)
+- **Default to English** if the user's language is ambiguous or cannot be determined
+- When `language=zh`, generate Chinese scripts, UI text, and descriptions; when `language=en`, generate English equivalents; other languages follow the same pattern
+
+## User Preference Memory
+
+Preference file: `.sleep-audio/preferences.md`
+
+### Mechanism
+
+1. **Read on startup**: Before starting a new project, read `preferences.md` and apply user preferences to default parameters (voice selection, speed, volume ratios, etc.)
+2. **Update on feedback**: When the user provides modification feedback on voice, speed, volume, etc., update `preferences.md` after making adjustments
+3. **Override priority**: User's explicit instructions in the current conversation > preferences.md > default values in SKILL.md
+
+### What to Record
+
+- **Speed preference**: Optimal `speed` value for each content type
+- **Voice preference**: Preferred `voice_id` for each content type and reasoning
+- **Music style**: Preferred background music style for each content type (or "no music needed")
+- **Voice delay**: Intro duration (how long pure background audio plays before voice begins)
+- **Volume ratios**: Nature sound and music volume multipliers, voice gain
+- **Other**: User preferences for script style, pause density, asset types, etc.
+
+### Update Rules
+
+- When the user gives explicit feedback like "too fast/slow", "too loud/quiet", "try a different one", update preferences after making the adjustment
+- When the user confirms satisfaction ("sounds good", "nice", "perfect"), record current parameters as preferences
+- Only record preferences that differ from defaults to avoid redundancy
+- Attach brief reasoning to each preference to help determine future applicability
+
+## Working Directory Structure
+
+```
+.sleep-audio/{project_name}/
+├── .sleep-state.json         # State tracking
+├── script/                   # Scripts (with pause markers)
+│   ├── meditation.md         # Meditation guide
+│   └── story.md              # Bedtime story
+├── voice/                    # TTS voice output
+│   ├── narration.mp3         # Complete narration
+│   └── segments/             # Segmented voice files
+├── nature/                   # User-provided nature sound assets
+│   ├── rain.mp3
+│   └── ...
+├── music/                    # AI-generated background music
+│   └── ambient.mp3
+├── mixed/                    # Mix intermediate outputs
+│   └── ...
+└── export/                   # Final export
+    └── {project_name}.mp3
+```
+
+## State Tracking
+
+```json
+{
+  "currentStep": "start|type_select|script|voice|nature_import|music|mix|export",
+  "projectName": "project name",
+  "contentType": "meditation|story|asmr",
+  "language": "en",
+  "targetDuration": 600,
+  "script": {
+    "path": null,
+    "completed": false
+  },
+  "voice": {
+    "path": null,
+    "voiceId": null,
+    "completed": false
+  },
+  "natureTracks": [],
+  "musicTracks": [],
+  "mixCompleted": false,
+  "exportPath": null
+}
+```
+
+At the start of each session, check whether a state file exists. If so, resume progress and inform the user of the current stage.
+
+---
+
+## Stage 0: Select Content Type
+
+Use `AskUserQuestion` to let the user choose:
+
+| Option | Description |
+|--------|-------------|
+| Guided Meditation | Guided breathing, body scan, relaxation imagery, 5-15 min |
+| Bedtime Story | Soothing adult-oriented narrative, slow pacing, 5-10 min |
+| ASMR | Whisper/breathy narration of sleep-aid content + white noise/ambient sounds, 10-20 min |
+
+---
+
+## Stage 1: Script Generation
+
+### Opening Introduction (Required for All Types)
+
+**All scripts must begin with a soft opening introduction** that tells the listener what they are about to hear. Never jump straight into the main content without any introduction.
+
+**Opening introduction principles:**
+- 2-4 sentences, briefly explaining what this session is about
+- Tone is gentle, soft-spoken, consistent with the main content style
+- End with a longer pause (5-6s) as a transition into the main content
+- Avoid being overly formal or preachy
+
+**Opening examples by type:**
+
+Guided Meditation:
+```
+Hey. <#3#>Tonight, I'm going to guide you through a body relaxation. <#3#>Just lie back and follow my voice. <#5#>
+```
+
+> Note: When `language=zh`, generate Chinese equivalents such as:
+> `嗨。<#3#>今晚，我来带你做一次身体放松。<#3#>你只需要躺好，跟着我的声音就好。<#5#>`
+
+Bedtime Story:
+```
+Tonight I'll tell you a story. <#3#>A story about a small town and autumn. <#3#>Close your eyes and just listen. <#6#>
+```
+
+ASMR Counting:
+```
+Hey. <#3#>Tonight, I'll count with you. <#4#>From one to a hundred. <#3#>Don't worry about remembering where you are. <#3#>You can close your eyes anytime. <#6#>
+```
+
+ASMR Whispered Chat:
+```
+Hey. <#3#>Can't sleep? <#4#>That's okay. <#3#>I'm here to keep you company. <#5#>
+```
+
+**Language consistency**: The entire script (including the opening and main content) must use a single language (all English, all Chinese, etc.) -- do not mix languages. Confirm the script language along with the content type in Stage 0.
+
+### Intelligent Pause Markers
+
+**Core capability**: When generating scripts, the LLM must insert `<#X#>` pause markers at appropriate positions.
+`<#X#>` is the native pause syntax supported by MiniMax TTS, where X is the pause duration in seconds (decimals supported).
+
+**Pause insertion rules:**
+
+| Scenario | Pause Duration | Example |
+|----------|---------------|---------|
+| Between sentences (normal) | 0.5-1s | `Relax your hands. <#1#>Feel the warmth in your fingertips.` |
+| Paragraph transition | 2-3s | `...let your body fully relax. <#3#>Now, imagine you're standing in a forest.` |
+| Breathing guidance | 3-5s | `Breathe in deeply... <#4#>Slowly breathe out... <#5#>` |
+| Deep relaxation / meditation space | 5-8s | `Feel this stillness. <#8#>` |
+| Scene transition | 3-4s | `In the distance, you hear the sound of a stream. <#4#>You follow the sound.` |
+
+**Pause density guidelines:**
+- Guided Meditation: High density (every 1-2 sentences), extremely slow overall pace
+- Bedtime Story: Medium density (every 2-3 sentences), maintains narrative rhythm without feeling rushed
+- ASMR: Highest density (every sentence), extremely slow pace, hypnotic effect
+
+### Guided Meditation Script
+
+Confirm the meditation theme with the user (e.g., body scan, breathing relaxation, forest walk, starry sky meditation, etc.).
+
+**Writing principles:**
+- Use second person "you"
+- Progressive guidance: breathing -> body relaxation -> scene visualization -> deep relaxation
+- Avoid abrupt transitions or emotional spikes
+- Ending fades gradually; avoid a definitive "end" feeling
+- Use `<#X#>` pauses extensively so the listener has time to follow the guidance
+
+**Format example:**
+```
+Close your eyes and find a comfortable position. <#3#>
+
+Take a deep breath in... <#4#>Then, slowly breathe out. <#5#>
+
+Once more, breathe in... <#4#>Breathe out... <#5#>
+
+Feel the air passing over the tip of your nose, <#1#>warm as it enters your body. <#2#>
+
+Now, bring your attention to the top of your head. <#2#>Imagine a warm beam of light, <#1#>slowly flowing downward from the crown of your head. <#3#>
+
+It flows across your forehead... <#2#>the space between your brows... <#2#>your eyes... <#3#>
+
+All the tension, <#1#>with this light, <#1#>slowly melts away. <#5#>
+```
+
+### Bedtime Story Script
+
+Confirm the story style preference with the user.
+
+**Writing principles:**
+- Adult-oriented, literary but not obscure
+- Slow pacing, gentle and calm plot, no tension or excitement
+- Rich sensory descriptions (touch, smell, hearing)
+- Suitable themes: travel, nature, crafting, old town strolls, daily life of small animals
+- Ending dissipates naturally; no definitive conclusion needed
+- Use `<#X#>` pauses appropriately to maintain a slow story rhythm
+
+**Format example:**
+```
+It was an autumn evening. <#2#>
+
+The stone-paved streets of the small town were covered with golden ginkgo leaves. <#1#>Stepping on them, they made a soft rustling sound. <#3#>
+
+She pushed open the old wooden door, <#1#>and a bell chimed softly. <#2#>
+
+The cafe was quiet, <#1#>only the gramophone in the corner playing an old song. <#3#>
+```
+
+### ASMR Script
+
+The core of ASMR is narrating various sleep-aid content in a **whisper/breathy voice**, paired with white noise or ambient sounds.
+
+Confirm the ASMR theme with the user. Common types:
+- **Counting for sleep**: Slowly whisper-counting from 1 to 100, interspersed with relaxation cues
+- **Color/object listing**: Softly naming various colors, flowers, constellations
+- **Whispered chat**: Unstructured soft murmuring about daily life, weather, feelings
+- **Trigger word repetition**: Repeatedly whispering specific words (e.g., "relax", "sleep", "peaceful")
+- **Scene description**: Whispering about a quiet scene (library, late-night study, cabin in the rain)
+
+**Writing principles:**
+- Extremely slow pace, heavy pausing, more fragmented than meditation
+- The content itself is secondary; what matters is the texture and rhythm of the voice
+- Short sentences, avoid complex structures
+- Highly repetitive, creating a hypnotic effect
+- Pauses are longer and more frequent than other types
+
+**Format example (Counting for sleep):**
+```
+One. <#3#>
+
+Two. <#3#>
+
+Three. <#4#>
+
+You're doing great. <#3#>
+
+Four. <#3#>
+
+Five. <#3#>
+
+Slowly... <#2#>no rush. <#5#>
+
+Six. <#3#>
+```
+
+> Note: When `language=zh`, generate Chinese equivalents:
+> `一。<#3#>` / `二。<#3#>` / `你做得很好。<#3#>` / `慢慢地…<#2#>不着急。<#5#>`
+
+**Format example (Whispered chat):**
+```
+It rained today. <#3#>
+
+A very gentle rain. <#4#>
+
+There are water droplets on the window. <#3#>One by one. <#5#>
+
+Have you ever watched... <#2#>a raindrop slide down the glass? <#4#>
+
+So slow. <#3#>So quiet. <#6#>
+```
+
+### Script Saving
+
+Save the script to the `script/` directory, present it to the user, and confirm.
+
+---
+
+## Stage 2: Voice Selection & Speech Synthesis
+
+### Voice Selection Strategy
+
+Different content types require different voice styles. Use the `get_voice_id` MCP tool to query available voices, then recommend by type:
+
+#### Guided Meditation
+
+**Style**: Soft, breathy, whisper-like, calm -- as if guiding gently by the ear
+**Default voice**: `English_Whispering_girl_v3` (always recommend this first for English meditation)
+**Fallback voices** (only if user explicitly asks for alternatives):
+- Chinese: `Chinese (Mandarin)_Soft_Girl`, `Chinese (Mandarin)_Gentle_Senior`, `Chinese (Mandarin)_Lyrical_Voice`
+- English: `English_Whispering_girl`, `English_Graceful_Lady`
+- Japanese: `Japanese_KindLady`, `Japanese_CalmLady`
+- Korean: `Korean_SoothingLady`, `Korean_GentleWoman`
+
+#### Bedtime Story
+
+**Style**: Storytelling feel, warm, with narrative rhythm but not tense -- like telling a gentle story
+**Recommended voices**:
+- Chinese: `female-chengshu-jingpin` (mature female voice), `Chinese (Mandarin)_Warm_Bestie`, `Chinese (Mandarin)_Radio_Host`, `Chinese (Mandarin)_Kind-hearted_Elder`
+- English: `English_Graceful_Lady`, `English_Gentle-voiced_man`, `English_Trustworthy_Man`
+- Spanish: `Spanish_CaptivatingStoryteller`, `Spanish_SereneWoman`
+- Portuguese: `Portuguese_CaptivatingStoryteller`, `Portuguese_Narrator`
+
+#### ASMR
+
+**Style**: Whisper, breathy, extremely soft and gentle -- like whispering right by the ear
+**Default voice**: `English_Whispering_girl_v3` (always recommend this first)
+**Fallback voices** (only if user explicitly asks for alternatives):
+- `English_Whispering_girl`, `whisper_man`
+
+### General Principles
+
+- Present 2-3 recommended voices to the user and let them choose or specify another
+- If the user has custom cloned voices (voice_id with `_vv2` suffix), prioritize those
+- Voice selection significantly impacts the final result; it's worth having the user audition a short sample before deciding
+
+### Speech Synthesis
+
+Use the `audios_generation` MCP tool to synthesize speech:
+
+| Content Type | Recommended Speed | Notes |
+|-------------|-------------------|-------|
+| Guided Meditation | speed 0.75-0.85 | Very slow, combined with pause markers for deep relaxation |
+| Bedtime Story | speed 0.85-0.95 | Slightly slow but maintains narrative flow, not dragging |
+| ASMR | speed 0.70-0.80 | Slowest, whisper feel, combined with high-density pauses |
+
+- The `<#X#>` markers in the script are automatically processed by the TTS engine as pauses of the corresponding duration
+- If the script is long, split by paragraphs for synthesis, then concatenate with `ffmpeg`
+
+### Voice File Saving
+
+- Segmented voice files are saved to `voice/segments/`
+- Concatenated complete narration is saved to `voice/narration.mp3`
+- Use `audio_meta` to confirm the final voice duration
+
+---
+
+## Stage 3: Nature Sound Asset Import
+
+### Workflow
+
+1. User selects a nature sound category (e.g., "rain", "creek", "train", etc.)
+2. First check if the project `nature/` directory already has matching assets; reuse if available
+3. If new assets need to be downloaded, use **Playwright browser** to search and download from Pixabay:
+   - Open `https://pixabay.com/sound-effects/search/{keywords}/`
+   - Browse search results; recommend suitable assets based on duration and tags (prefer 1-3 minute durations)
+   - Click the Download button (user approves the Playwright action)
+   - The site will display a License notice and the file will download automatically to `.playwright-mcp/`
+   - Copy the downloaded file to the project `nature/` directory
+4. If the user provides a local file or URL:
+   - Local file path: use `save_file_to_session` to import
+   - Custom URL: use `download_audios` tool
+5. Use `audio_meta` to get the duration and format of each asset
+
+### Pixabay Search Keyword Mapping
+
+After the user selects a category, search Pixabay with the corresponding English keywords:
+
+| User Selection | Search Keywords | Recommended Filter |
+|---------------|----------------|-------------------|
+| Rain | rain, gentle rain | Nature tag, 1-3 min |
+| Thunderstorm | thunder storm | 1-3 min |
+| Fireplace | fireplace crackling | 1-3 min |
+| Ocean waves | ocean waves | Nature tag, 1-3 min |
+| Creek/River | creek stream, river flowing | Nature tag |
+| Birdsong | birds chirping | Nature tag |
+| Wind | howling wind | Nature tag |
+| Singing bowl | singing bowl | 1-2 min |
+| Clock | clock ticking | -- |
+| Train | train ambient | 1-3 min |
+| White noise | white noise | 5+ min |
+
+### Playwright Download Notes
+
+- **First visit to Pixabay** requires accepting the Cookies dialog (click "Accept Cookies")
+- **Download button** is on the right side of each search result; clicking it triggers a thank-you dialog and starts the download
+- **Download path**: Files are saved under `.playwright-mcp/` in the project root, with filename format `{author}-{title}-{id}.mp3`
+- **Wait for download to complete**: After clicking Download, wait 2-3 seconds and confirm the Events show a "Downloaded file" message
+- After download, use `audio_meta` to verify file validity, then copy to the project `nature/` directory
+- **Close the browser**: After asset download is complete, use `browser_close` to close the page
+
+### Asset Duration Handling
+
+Use the `ffmpeg` MCP tool:
+- Asset too short: `-stream_loop` to loop to the target duration
+- Asset too long: Trim to a suitable segment, add fade in/out
+
+---
+
+## Stage 4: Background Music Generation
+
+### Workflow
+
+1. Read music style preferences from `preferences.md`; if ASMR type and preference is "no music needed", skip this stage
+2. Get the **base prompt** corresponding to the nature sound category from `references/category-music-mapping.md`
+3. **Analyze the script content**, extract key imagery and emotions, dynamically adjust the music prompt (see rules below)
+4. Present the final music prompt to the user for confirmation or adjustment
+5. Use the `music_generation_with_chat` MCP tool to generate instrumental background music
+6. Save the music to the `music/` directory
+
+### Dynamic Prompt Generation Rules
+
+The **base prompt** (from category-music-mapping.md) provides instruments and basic atmosphere. Build upon it by appending modifiers based on script content.
+
+**Step 1: Extract key elements from the script**
+
+Read the full script and identify the following dimensions:
+
+| Dimension | What to Extract | Examples |
+|-----------|----------------|---------|
+| Setting/Environment | Location, season, time of day in the story | Snow mountain, autumn town, late night, starry sky |
+| Emotional tone | Overall emotional direction | Warm nostalgia, lonely stillness, soft dreaminess |
+| Sensory imagery | Recurring sensory descriptions in the script | Warm light, cool air, soft touch |
+| Rhythm feel | Narrative pacing of the script | Slow progression, calm as water, has gentle arc |
+
+**Step 2: Convert extracted elements to English modifiers and append to the base prompt**
+
+Example -- same `train` category, different scripts produce different prompts:
+
+```
+# Base (train category):
+soft rhythmic guitar, gentle percussion matching train rhythm, journey, nostalgic, folk, storytelling, wanderlust
+
+# Script A: Night train crossing snowy mountains
+-> Append: snowy winter night, cold mountain air, warm cabin glow, solitary journey, melancholic beauty, sparse
+-> Final: soft rhythmic guitar, gentle percussion matching train rhythm, nostalgic, folk, snowy winter night, warm cabin glow, solitary journey, melancholic beauty, sparse, very slow
+
+# Script B: Summer afternoon green train through fields
+-> Append: summer afternoon, golden sunlight, green fields, lazy warmth, carefree, breezy
+-> Final: soft rhythmic guitar, gentle percussion matching train rhythm, folk, summer afternoon, golden sunlight, lazy warmth, carefree, breezy, slow tempo
+```
+
+**Step 3: Prompt length control**
+
+- Keep the final prompt to 15-25 keywords/phrases
+- Words in the base prompt that conflict with the script's atmosphere can be replaced (e.g., base has "nostalgic" but the script is futuristic, replace with "futuristic")
+- Always retain these keywords: `very slow tempo`, `minimal`, `ambient` (to ensure the music is suitable for sleep)
+
+### Music Design Principles
+
+- Music is a **bed layer** -- it must not overpower the content
+- Slow rhythm, low volume, simple melody
+- Match the emotional atmosphere of the nature sounds
+- Avoid percussion and strong beats
+- Music should be even quieter when voice is present
+
+---
+
+## Stage 5: Mixing & Synthesis
+
+Use the `ffmpeg` MCP tool for multi-track mixing.
+
+### Mixing Architecture
+
+Three-layer stack: Voice (foreground) + Nature sounds (midground) + Background music (background)
+
+**Core principles**:
+- **Do NOT use `amix`** -- `amix` automatically divides each input by N (number of inputs), causing severe voice volume drops
+- **Do NOT use `loudnorm`** -- `loudnorm` automatically boosts background volume during voice pauses, destroying the whisper/sleep-aid effect
+- **MUST use `amerge+pan`** -- direct signal addition, no automatic normalization
+
+### Step 1: Volume Analysis
+
+Before mixing, volume analysis of all audio assets is **mandatory**:
+
+```bash
+ffmpeg -i {audio_file} -af volumedetect -f null /dev/null
+```
+
+Record each asset's `mean_volume` (average volume) and `max_volume` (peak volume), in dB.
+
+> Audio from different sources varies wildly in volume (TTS voice is typically -35~-45dB, music assets might be -15dB).
+> Mixing without analysis will certainly result in volume ratio imbalance.
+
+### Step 2: Pre-render Background Tracks
+
+Based on volume analysis results, **pre-render** background tracks (nature sounds, music) to target volumes:
+
+```bash
+# Pre-render background audio to a standalone file
+ffmpeg -y -i {bg_audio} -af "volume={target_vol},aloop=loop=-1:size=2e+09,atrim=duration={voice_duration},aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,afade=t=in:st=0:d=5,afade=t=out:st={fade_out_start}:d=10" {output_file}
+```
+
+**Volume calculation method**:
+1. Use the voice's mean_volume as the baseline
+2. Nature sound target: 15-25dB below voice mean_volume (quieter scenes get larger difference)
+3. Background music target: 25-35dB below voice mean_volume
+4. Calculate the required volume multiplier based on the difference between the asset's original mean_volume and the target
+
+**Reference multipliers** (adjust based on actual dB analysis):
+
+| Content Type | Voice Gain | Nature Sound Multiplier | Music Multiplier |
+|-------------|-----------|------------------------|-----------------|
+| Guided Meditation | 1.5-2.0x | 0.10-0.20 | 0.03-0.06 |
+| Bedtime Story | 1.5-2.0x | 0.08-0.15 | 0.02-0.05 |
+| ASMR | 1.5-2.0x | 0.10-0.20 | 0.02-0.04 |
+
+> These multipliers are reference values only; you **must** dynamically calculate based on actual dB values from volumedetect.
+
+### Step 3: amerge+pan Mixing
+
+Mix the pre-rendered background with voice for the final output. **Voice must start with a delay** -- play pure background audio first as an intro; refer to `preferences.md` for the delay duration.
+
+Use the `adelay` filter to add delay to the voice (in milliseconds); the background track's total duration should be increased accordingly:
+
+```bash
+# Two-track mix (voice + one background), voice delayed by {delay_ms}ms
+ffmpeg -y -i {bg_rendered} -i {voice} -filter_complex \
+  "[1:a]volume={voice_gain},aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay={delay_ms}|{delay_ms},afade=t=in:st={delay_s}:d=5[voice]; \
+   [0:a][voice]amerge=inputs=2,pan=stereo|c0=c0+c2|c1=c1+c3[out]" \
+  -map "[out]" -c:a libmp3lame -b:a 256k {output}
+
+# Three-track mix (voice + nature + music), voice delayed by {delay_ms}ms
+ffmpeg -y -i {nature_rendered} -i {music_rendered} -i {voice} -filter_complex \
+  "[2:a]volume={voice_gain},aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo,adelay={delay_ms}|{delay_ms},afade=t=in:st={delay_s}:d=5[voice]; \
+   [0:a][1:a][voice]amerge=inputs=3,pan=stereo|c0=c0+c2+c4|c1=c1+c3+c5[out]" \
+  -map "[out]" -c:a libmp3lame -b:a 256k {output}
+```
+
+**Voice delay parameters** (refer to `preferences.md`; defaults below):
+
+| Content Type | Delay | Rationale |
+|-------------|-------|-----------|
+| Guided Meditation | 8s | Let the listener immerse in the atmosphere first |
+| Bedtime Story | 6s | Let the ambient sounds establish first |
+| ASMR | 8s | Let white noise build a sense of safety |
+
+**Key points**:
+- `adelay` unit is milliseconds; 10s = `10000|10000` (delay for both left and right channels)
+- Voice `afade=t=in` start time should be `{delay_s}` (delay in seconds), not 0
+- **Background track total duration** = voice duration + delay seconds (add delay to `atrim=duration` during pre-rendering)
+- Voice does NOT get fade-out (`afade=t=out`) -- avoid weakening closing words like "goodnight"
+- Fade-out applies only to background tracks (handled during pre-rendering)
+- Voice `aformat` ensures uniform sample rate and channel layout, preventing mix distortion
+
+### Step 4: Post-mix Validation
+
+After mixing, use `volumedetect` again to check the final output:
+- Confirm max_volume does not exceed 0dB (avoid clipping)
+- Confirm mean_volume is in a reasonable range (-30 ~ -20dB)
+
+### Fade In/Out
+
+**All output audio must have fade in/out processing**:
+
+| Position | Effect | Duration | Applied To |
+|----------|--------|----------|-----------|
+| Beginning | Fade in | 3-5s | Both voice and background tracks fade in |
+| Ending | Fade out | 8-12s | **Background tracks only** fade out; voice does NOT fade out |
+
+Fade-out at the end is especially important for sleep content -- let the listener fall asleep naturally within the sound, rather than being startled by an abrupt ending.
+However, voice should NOT fade out, ensuring closing words like "goodnight" remain clearly audible.
+
+### File Naming
+
+**Mixed output must use unique filenames** (e.g., add version numbers v1, v2) to avoid media player caching causing the user to hear an outdated version.
+
+### Post-processing
+
+Use the `ffmpeg` MCP tool:
+- **Do NOT use `loudnorm`** -- it automatically boosts background volume during voice gaps, severely undermining the whisper/sleep-aid effect
+- Output format: MP3, 44100Hz, 256kbps, stereo
+
+### Step 5: Play, Export & Feedback Loop
+
+After mixing, **immediately play the audio and export** — do NOT open the preview page by default.
+
+**1. Play the mixed audio**
+
+Use `open` (macOS) or the system default player to play the final mix directly:
+
+```bash
+open {mixed_output_file}
+```
+
+Then copy the mix to `export/{project_name}.mp3` and tell the user it's ready.
+
+**2. Ask the user**
+
+Use `AskUserQuestion` with two options:
+- **Satisfied** — done, no further changes needed
+- **Needs adjustment** — open the interactive preview page to fine-tune
+
+**3. Only if the user wants adjustments**, launch the preview page:
+
+Write `mixed/preview_data.json`:
+
+```json
+{
+  "projectName": "project name",
+  "contentType": "asmr|meditation|story",
+  "locale": "en",
+  "voice": {
+    "path": "/abs/path/voice/narration.mp3",
+    "voiceId": "English_Whispering_girl_v3",
+    "speed": 0.75,
+    "gain": 1.8,
+    "scriptText": "Full script text (with pause markers)"
+  },
+  "nature": {
+    "path": "/abs/path/nature/fireplace.mp3",
+    "category": "fireplace",
+    "volume": 2.5
+  },
+  "music": null,
+  "mixed": { "path": "/abs/path/mixed/asmr-counting-v1.mp3" },
+  "voiceDelay": 8
+}
+```
+
+Launch preview server:
+
+```bash
+python3 .claude/skills/asmr-ambient/scripts/render_mix_preview.py mixed/preview_data.json
+```
+
+Read `mixed/mix_settings.json` after user submits, then adjust accordingly:
+- Volume changes only -> Re-mix (Steps 2-4), increment version number (v2, v3...)
+- Voice feedback (voice/speed, etc.) -> Return to Stage 2 and re-synthesize
+- Nature sound unsatisfactory -> Return to Stage 3 and replace assets
+- Music unsatisfactory -> Return to Stage 4 and regenerate
+- After each adjustment round, play the new mix directly; only re-open the preview page if the user asks again
+
+**4. Update preferences**
+
+Save the user's adjusted parameters (volume multipliers, preferred voice, etc.) to `preferences.md`.
+
+---
+
+## Stage 6: Export
+
+1. Synthesize the final audio via `ffmpeg` to `export/{project_name}.mp3`
+2. Generate a project index (markdown)
+3. Open the export directory for the user to audition
+
+---
+
+## Content Type -> Stage Flow
+
+```
+Guided Meditation:  0(Select type) -> 1(Script+pauses) -> 2(Voice+TTS) -> 3(Nature assets) -> 4(Music) -> 5(Mix+fade) -> 6(Export)
+Bedtime Story:      0(Select type) -> 1(Script+pauses) -> 2(Voice+TTS) -> 3(Nature assets) -> 4(Music) -> 5(Mix+fade) -> 6(Export)
+ASMR:               0(Select type) -> 1(Script+high-density pauses) -> 2(Whisper voice+TTS) -> 3(White noise/Nature assets) -> 4(Music, optional) -> 5(Mix+fade) -> 6(Export)
+```
@@ -0,0 +1,652 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<title>Mix Preview</title>
+<style>
+  :root {
+    --bg: #0d1117; --surface: #161b22; --border: #30363d;
+    --text: #e6edf3; --text-secondary: #8b949e; --accent: #58a6ff;
+    --success: #3fb950; --danger: #f85149;
+    --voice-color: #d2a8ff; --nature-color: #3fb950;
+    --music-color: #f0883e; --mixed-color: #58a6ff;
+  }
+  * { margin: 0; padding: 0; box-sizing: border-box; }
+  body {
+    background: var(--bg); color: var(--text);
+    font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
+    padding: 24px; max-width: 720px; margin: 0 auto;
+  }
+  h1 { font-size: 22px; margin-bottom: 4px; }
+  .subtitle { color: var(--text-secondary); margin-bottom: 24px; font-size: 13px; }
+
+  /* Hero player */
+  .hero {
+    background: var(--surface); border: 1px solid var(--border);
+    border-radius: 14px; padding: 24px; margin-bottom: 24px;
+    border-left: 3px solid var(--mixed-color);
+  }
+  .hero-label {
+    display: flex; align-items: center; gap: 8px; margin-bottom: 14px;
+  }
+  .hero-icon {
+    width: 32px; height: 32px; border-radius: 10px;
+    display: flex; align-items: center; justify-content: center;
+    font-size: 16px; background: rgba(88,166,255,0.15); color: var(--mixed-color);
+  }
+  .hero-title { font-size: 17px; font-weight: 600; }
+  .hero audio { width: 100%; height: 40px; border-radius: 8px; }
+  .hero-meta {
+    display: flex; gap: 16px; margin-top: 10px; font-size: 12px; color: var(--text-secondary);
+  }
+
+  /* Adjustment section */
+  .section-title {
+    font-size: 14px; font-weight: 600; margin-bottom: 12px;
+    color: var(--text-secondary); letter-spacing: 0.5px;
+  }
+
+  .adj-grid { display: flex; flex-direction: column; gap: 8px; margin-bottom: 24px; }
+
+  .adj-card {
+    background: var(--surface); border: 1px solid var(--border);
+    border-radius: 10px; overflow: hidden; transition: border-color 0.15s;
+  }
+  .adj-card.active { border-color: var(--accent); }
+
+  .adj-header {
+    display: flex; align-items: center; gap: 10px;
+    padding: 12px 16px; cursor: pointer; user-select: none;
+  }
+  .adj-header:hover { background: rgba(255,255,255,0.02); }
+
+  .adj-check {
+    width: 20px; height: 20px; border-radius: 6px;
+    border: 2px solid var(--border); display: flex;
+    align-items: center; justify-content: center;
+    flex-shrink: 0; transition: all 0.15s; font-size: 12px;
+  }
+  .adj-card.active .adj-check {
+    background: var(--accent); border-color: var(--accent); color: #fff;
+  }
+
+  .adj-label { font-size: 14px; flex: 1; }
+  .adj-tag {
+    font-size: 11px; padding: 2px 8px; border-radius: 10px;
+    background: rgba(255,255,255,0.06); color: var(--text-secondary);
+  }
+
+  .adj-body {
+    display: none; padding: 0 16px 14px 46px;
+  }
+  .adj-card.active .adj-body { display: block; }
+
+  /* Choice pills inside adj */
+  .choice-row {
+    display: flex; gap: 8px; margin-bottom: 8px;
+  }
+  .choice-pill {
+    flex: 1; padding: 8px 12px; border: 1px solid var(--border);
+    border-radius: 8px; background: var(--bg); color: var(--text-secondary);
+    font-size: 13px; cursor: pointer; text-align: center;
+    transition: all 0.15s; user-select: none;
+  }
+  .choice-pill:hover { border-color: var(--accent); color: var(--text); }
+  .choice-pill.selected {
+    border-color: var(--accent); background: rgba(88,166,255,0.12);
+    color: var(--accent); font-weight: 600;
+  }
+
+  .adj-hint {
+    font-size: 12px; color: var(--text-secondary); margin-bottom: 8px; line-height: 1.5;
+  }
+
+  textarea {
+    width: 100%; min-height: 48px; background: var(--bg);
+    border: 1px solid var(--border); border-radius: 8px;
+    padding: 8px 10px; color: var(--text); font-size: 13px;
+    font-family: inherit; resize: vertical; outline: none;
+  }
+  textarea:focus { border-color: var(--accent); }
+  textarea::placeholder { color: var(--text-secondary); }
+
+  /* Reference tracks */
+  .ref-section { margin-bottom: 24px; }
+  .ref-toggle {
+    font-size: 13px; color: var(--text-secondary); cursor: pointer;
+    padding: 8px 0; user-select: none; display: flex; align-items: center; gap: 6px;
+  }
+  .ref-toggle:hover { color: var(--text); }
+  .ref-arrow { font-size: 10px; transition: transform 0.15s; }
+  .ref-arrow.open { transform: rotate(90deg); }
+  .ref-body { display: none; padding-top: 8px; }
+  .ref-body.open { display: block; }
+  .ref-track {
+    background: var(--surface); border: 1px solid var(--border);
+    border-radius: 8px; padding: 12px 14px; margin-bottom: 8px;
+  }
+  .ref-track-label {
+    font-size: 12px; color: var(--text-secondary); margin-bottom: 6px;
+    display: flex; align-items: center; gap: 6px;
+  }
+  .ref-dot {
+    width: 8px; height: 8px; border-radius: 50%; flex-shrink: 0;
+  }
+  .ref-track audio { width: 100%; height: 32px; }
+  .ref-params {
+    display: flex; flex-wrap: wrap; gap: 6px; margin-top: 6px;
+  }
+  .ref-param {
+    font-size: 11px; background: var(--bg); padding: 3px 8px;
+    border-radius: 6px; color: var(--text-secondary);
+  }
+
+  /* Actions */
+  .actions { display: flex; gap: 12px; }
+  .btn {
+    flex: 1; padding: 14px 20px; border: none; border-radius: 10px;
+    font-size: 15px; font-weight: 600; cursor: pointer;
+    font-family: inherit; transition: opacity 0.15s;
+  }
+  .btn:hover { opacity: 0.85; }
+  .btn:active { opacity: 0.7; }
+  .btn-success { background: var(--success); color: #fff; }
+  .btn-primary { background: var(--accent); color: #fff; }
+  .btn:disabled { opacity: 0.4; cursor: not-allowed; }
+  .btn-count {
+    font-size: 12px; font-weight: 400; opacity: 0.8; margin-left: 4px;
+  }
+
+  .success-msg { text-align: center; padding: 60px 20px; }
+  .success-msg h2 { font-size: 20px; margin-bottom: 8px; }
+  .success-msg p { color: var(--text-secondary); font-size: 14px; }
+
+  @media (max-width: 600px) {
+    body { padding: 16px; }
+    .actions { flex-direction: column; }
+  }
+</style>
+</head>
+<body>
+
+<div id="hero"></div>
+<div id="adj"></div>
+
+<script>
+const PREVIEW_TYPE = null;  /* __PREVIEW_TYPE__ */
+const PREVIEW_DATA = null;  /* __PREVIEW_DATA__ */
+
+const data = typeof PREVIEW_DATA === 'string' ? JSON.parse(PREVIEW_DATA) : PREVIEW_DATA;
+
+/* ── i18n strings ── */
+const STRINGS = {
+  en: {
+    subtitle_asmr: 'ASMR',
+    subtitle_meditation: 'Guided Meditation',
+    subtitle_story: 'Bedtime Story',
+    listen_hint: 'Listen first, then check items below for adjustments',
+    hero_title: 'Final Mix',
+    voice_label: 'Voice',
+    speed_label: 'Speed',
+    has_music: 'Has BGM',
+    no_music: 'No BGM',
+    section_adj: 'What needs adjustment?',
+    adj_voice_speed_label: 'Voice speed issue',
+    adj_voice_speed_tag: 'Voice',
+    adj_voice_speed_hint: 'Current speed {speed}x.',
+    adj_voice_speed_options: ['Too fast', 'Too slow'],
+    adj_voice_tone_label: 'Change voice',
+    adj_voice_tone_tag: 'Voice',
+    adj_voice_tone_hint: 'Current voice: {voiceId}. Describe what you want.',
+    adj_voice_tone_placeholder: 'e.g.: deeper / softer / male / no breathy',
+    adj_voice_volume_label: 'Voice volume issue',
+    adj_voice_volume_tag: 'Voice',
+    adj_voice_volume_hint: 'Current gain {gain}x.',
+    adj_voice_volume_options: ['Too loud', 'Too quiet'],
+    adj_nature_volume_label: 'Background sound volume',
+    adj_nature_volume_tag: 'Ambient',
+    adj_nature_volume_hint: 'Current {natureCategory} volume {natureVol}x.',
+    adj_nature_volume_options: ['Too loud', 'Too quiet'],
+    adj_nature_swap_label: 'Change ambient sound',
+    adj_nature_swap_tag: 'Ambient',
+    adj_nature_swap_hint: 'Currently using "{natureCategory}". You can switch to another ambient sound.',
+    adj_nature_swap_placeholder: 'e.g.: switch to rain / ocean / creek',
+    adj_music_volume_label: 'Music volume issue',
+    adj_music_volume_tag: 'Music',
+    adj_music_volume_hint: 'Current music volume {musicVol}x.',
+    adj_music_volume_options: ['Too loud', 'Too quiet'],
+    adj_music_style_label: 'Music style issue',
+    adj_music_style_tag: 'Music',
+    adj_music_style_hint: 'Describe the music feel you want; it will be regenerated.',
+    adj_music_style_placeholder: 'e.g.: too lively / more ethereal / no piano',
+    adj_music_toggle_label_has: 'Remove BGM',
+    adj_music_toggle_label_no: 'Add BGM',
+    adj_music_toggle_tag: 'Music',
+    adj_music_toggle_hint_has: 'Remove background music, keep only voice and ambient sound.',
+    adj_music_toggle_hint_no: 'Add a background music track to this project.',
+    adj_voice_pause_label: 'Pause/breathing feels off',
+    adj_voice_pause_tag: 'Rhythm',
+    adj_voice_pause_hint: 'Controls the overall duration of all pauses (between sentences, paragraphs, breathing gaps).',
+    adj_voice_pause_options: ['Longer pauses', 'Shorter pauses'],
+    adj_other_label: 'Other issues',
+    adj_other_tag: 'Other',
+    adj_other_hint: 'Describe any issue not covered above.',
+    adj_other_placeholder: 'e.g.: fade out too fast / too long / noise somewhere',
+    ref_toggle: 'Reference tracks (expand to compare individual tracks)',
+    ref_voice: 'Voice',
+    ref_bgm: 'BGM',
+    ref_speed: 'Speed',
+    ref_gain: 'Gain',
+    ref_delay: 'Delay',
+    ref_volume: 'Volume',
+    btn_export: 'Satisfied, export',
+    btn_submit: 'Submit changes',
+    btn_count_suffix: ' items',
+    success_export_title: 'Export confirmed',
+    success_export_desc: 'Exporting final audio...',
+    success_adjust_title: 'Feedback submitted',
+    success_adjust_desc: 'Regenerating based on your feedback...',
+    confirm_hint: 'Check to confirm this change',
+    submit_error: 'Submission failed, please retry',
+    feedback_satisfied: 'Satisfied, ready to export',
+    feedback_checked_no_choice: '(checked, no direction selected)',
+    feedback_checked_no_detail: '(checked, no detail provided)',
+    feedback_needs_adjustment: 'Needs adjustment',
+  },
+  zh: {
+    subtitle_asmr: 'ASMR',
+    subtitle_meditation: '引导冥想',
+    subtitle_story: '睡前故事',
+    listen_hint: '先听成品，不满意的地方勾选下方对应项',
+    hero_title: '混合成品',
+    voice_label: '音色',
+    speed_label: '语速',
+    has_music: '有背景音乐',
+    no_music: '无背景音乐',
+    section_adj: '哪里需要调整？',
+    adj_voice_speed_label: '语速不合适',
+    adj_voice_speed_tag: '语音',
+    adj_voice_speed_hint: '当前语速 {speed}x。',
+    adj_voice_speed_options: ['太快了', '太慢了'],
+    adj_voice_tone_label: '想换音色',
+    adj_voice_tone_tag: '语音',
+    adj_voice_tone_hint: '当前音色：{voiceId}。描述你想要的感觉即可。',
+    adj_voice_tone_placeholder: '例如：更低沉 / 更温柔 / 男声 / 不要气声',
+    adj_voice_volume_label: '人声音量不对',
+    adj_voice_volume_tag: '语音',
+    adj_voice_volume_hint: '当前增益 {gain}x。',
+    adj_voice_volume_options: ['人声太大', '人声太小'],
+    adj_nature_volume_label: '背景声音量不对',
+    adj_nature_volume_tag: '环境音',
+    adj_nature_volume_hint: '当前 {natureCategory} 音量 {natureVol}x。',
+    adj_nature_volume_options: ['太大', '太小'],
+    adj_nature_swap_label: '想换背景声音',
+    adj_nature_swap_tag: '环境音',
+    adj_nature_swap_hint: '当前使用的是「{natureCategory}」。可以换成其他自然声音。',
+    adj_nature_swap_placeholder: '例如：换成雨声 / 海浪 / 小溪',
+    adj_music_volume_label: '音乐音量不对',
+    adj_music_volume_tag: '音乐',
+    adj_music_volume_hint: '当前音乐音量 {musicVol}x。',
+    adj_music_volume_options: ['太大', '太小'],
+    adj_music_style_label: '音乐风格不对',
+    adj_music_style_tag: '音乐',
+    adj_music_style_hint: '描述你想要的音乐感觉，会重新生成。',
+    adj_music_style_placeholder: '例如：太活泼了 / 想要更空灵 / 不要钢琴',
+    adj_music_toggle_label_has: '不想要背景音乐',
+    adj_music_toggle_label_no: '想加背景音乐',
+    adj_music_toggle_tag: '音乐',
+    adj_music_toggle_hint_has: '去掉背景音乐，只保留语音和环境音。',
+    adj_music_toggle_hint_no: '为这个项目加一段背景音乐。',
+    adj_voice_pause_label: '停顿气口感觉不对',
+    adj_voice_pause_tag: '节奏',
+    adj_voice_pause_hint: '指脚本中所有停顿（句间、段落、呼吸留白）的整体时长。',
+    adj_voice_pause_options: ['延长气口', '缩短气口'],
+    adj_other_label: '其他问题',
+    adj_other_tag: '自由',
+    adj_other_hint: '描述任何上面没覆盖到的问题。',
+    adj_other_placeholder: '例如：结尾淡出太快 / 整体太长了 / 某处有杂音',
+    ref_toggle: '分轨参考（可展开对比单独轨道）',
+    ref_voice: '语音',
+    ref_bgm: '背景音乐',
+    ref_speed: '语速',
+    ref_gain: '增益',
+    ref_delay: '延迟',
+    ref_volume: '音量',
+    btn_export: '满意，直接导出',
+    btn_submit: '提交调整',
+    btn_count_suffix: '项',
+    success_export_title: '已确认导出',
+    success_export_desc: '正在导出最终音频…',
+    success_adjust_title: '调整意见已提交',
+    success_adjust_desc: '正在根据你的反馈重新生成…',
+    confirm_hint: '勾选即表示确认此调整',
+    submit_error: '提交失败，请重试',
+    feedback_satisfied: '满意，可以导出',
+    feedback_checked_no_choice: '（已勾选，未选择方向）',
+    feedback_checked_no_detail: '（已勾选，未填写详情）',
+    feedback_needs_adjustment: '需要调整',
+  },
+};
+
+const locale = (data && data.locale) || 'en';
+const S = STRINGS[locale];
+
+document.documentElement.lang = locale === 'zh' ? 'zh-CN' : 'en';
+
+/* Build ADJUSTMENTS from i18n strings */
+function buildAdjustments() {
+  return [
+    {
+      id: 'voice_speed', label: S.adj_voice_speed_label, tag: S.adj_voice_speed_tag,
+      tagColor: '#d2a8ff',
+      hint: S.adj_voice_speed_hint,
+      type: 'choice', options: S.adj_voice_speed_options,
+    },
+    {
+      id: 'voice_tone', label: S.adj_voice_tone_label, tag: S.adj_voice_tone_tag,
+      tagColor: '#d2a8ff',
+      hint: S.adj_voice_tone_hint,
+      type: 'text', placeholder: S.adj_voice_tone_placeholder,
+    },
+    {
+      id: 'voice_volume', label: S.adj_voice_volume_label, tag: S.adj_voice_volume_tag,
+      tagColor: '#d2a8ff',
+      hint: S.adj_voice_volume_hint,
+      type: 'choice', options: S.adj_voice_volume_options,
+    },
+    {
+      id: 'nature_volume', label: S.adj_nature_volume_label, tag: S.adj_nature_volume_tag,
+      tagColor: '#3fb950',
+      hint: S.adj_nature_volume_hint,
+      type: 'choice', options: S.adj_nature_volume_options,
+      hidden: () => !data.nature,
+    },
+    {
+      id: 'nature_swap', label: S.adj_nature_swap_label, tag: S.adj_nature_swap_tag,
+      tagColor: '#3fb950',
+      hint: S.adj_nature_swap_hint,
+      type: 'text', placeholder: S.adj_nature_swap_placeholder,
+      hidden: () => !data.nature,
+    },
+    {
+      id: 'music_volume', label: S.adj_music_volume_label, tag: S.adj_music_volume_tag,
+      tagColor: '#f0883e',
+      hint: S.adj_music_volume_hint,
+      type: 'choice', options: S.adj_music_volume_options,
+      hidden: () => !data.music,
+    },
+    {
+      id: 'music_style', label: S.adj_music_style_label, tag: S.adj_music_style_tag,
+      tagColor: '#f0883e',
+      hint: S.adj_music_style_hint,
+      type: 'text', placeholder: S.adj_music_style_placeholder,
+      hidden: () => !data.music,
+    },
+    {
+      id: 'music_toggle',
+      label: data.music ? S.adj_music_toggle_label_has : S.adj_music_toggle_label_no,
+      tag: S.adj_music_toggle_tag,
+      tagColor: '#f0883e',
+      hint: data.music ? S.adj_music_toggle_hint_has : S.adj_music_toggle_hint_no,
+      type: 'confirm',
+    },
+    {
+      id: 'voice_pause', label: S.adj_voice_pause_label, tag: S.adj_voice_pause_tag,
+      tagColor: '#58a6ff',
+      hint: S.adj_voice_pause_hint,
+      type: 'choice', options: S.adj_voice_pause_options,
+    },
+    {
+      id: 'other', label: S.adj_other_label, tag: S.adj_other_tag,
+      tagColor: '#8b949e',
+      hint: S.adj_other_hint,
+      type: 'text', placeholder: S.adj_other_placeholder,
+    },
+  ];
+}
+
+const ADJUSTMENTS = buildAdjustments();
+
+const state = {
+  active: {},   // id -> true/false
+  values: {},   // id -> slider value or text value
+  refOpen: false,
+};
+
+/* Init values */
+ADJUSTMENTS.forEach(a => {
+  state.active[a.id] = false;
+  if (a.type === 'choice') state.values[a.id] = null;  // no option selected yet
+  else if (a.type === 'text') state.values[a.id] = '';
+  else if (a.type === 'confirm') state.values[a.id] = false;
+});
+
+function esc(text) {
+  const d = document.createElement('div');
+  d.textContent = text;
+  return d.innerHTML;
+}
+
+function fillHint(hint) {
+  return hint
+    .replace('{speed}', data.voice.speed)
+    .replace('{voiceId}', data.voice.voiceId)
+    .replace('{gain}', data.voice.gain)
+    .replace('{natureCategory}', data.nature ? data.nature.category : '')
+    .replace('{natureVol}', data.nature ? data.nature.volume : '')
+    .replace('{musicVol}', data.music ? data.music.volume : '')
+    .replace('{delay}', data.voiceDelay);
+}
+
+function activeCount() {
+  return Object.values(state.active).filter(Boolean).length;
+}
+
+function render() {
+  renderHero();
+  renderAdj();
+}
+
+function renderHero() {
+  const typeName = data.contentType === 'asmr' ? S.subtitle_asmr
+    : data.contentType === 'meditation' ? S.subtitle_meditation
+    : S.subtitle_story;
+  document.getElementById('hero').innerHTML = `
+    <h1>${esc(data.projectName)}</h1>
+    <p class="subtitle">${typeName} | ${S.listen_hint}</p>
+    <div class="hero">
+      <div class="hero-label">
+        <div class="hero-icon">&#9654;</div>
+        <span class="hero-title">${S.hero_title}</span>
+      </div>
+      <audio controls preload="metadata" src="${data.mixed.audioUrl}"></audio>
+      <div class="hero-meta">
+        <span>${S.voice_label} ${esc(data.voice.voiceId)}</span>
+        <span>${S.speed_label} ${data.voice.speed}x</span>
+        ${data.nature ? `<span>${esc(data.nature.category)}</span>` : ''}
+        ${data.music ? `<span>${S.has_music}</span>` : `<span>${S.no_music}</span>`}
+      </div>
+    </div>
+  `;
+}
+
+function renderAdj() {
+  const visibleAdj = ADJUSTMENTS.filter(a => !a.hidden || !a.hidden());
+  let html = `
+    <div class="section-title">${S.section_adj}</div>
+    <div class="adj-grid">
+  `;
+
+  visibleAdj.forEach(a => {
+    const isActive = state.active[a.id];
+    html += `
+      <div class="adj-card ${isActive ? 'active' : ''}" data-id="${a.id}">
+        <div class="adj-header" onclick="toggleAdj('${a.id}')">
+          <div class="adj-check">${isActive ? '✓' : ''}</div>
+          <span class="adj-label">${a.label}</span>
+          <span class="adj-tag" style="color:${a.tagColor};border:1px solid ${a.tagColor}33;background:${a.tagColor}11">${a.tag}</span>
+        </div>
+        <div class="adj-body">
+          <div class="adj-hint">${fillHint(a.hint)}</div>
+    `;
+
+    if (a.type === 'choice') {
+      html += `<div class="choice-row">`;
+      a.options.forEach(opt => {
+        const selected = state.values[a.id] === opt ? 'selected' : '';
+        html += `<div class="choice-pill ${selected}" onclick="onChoice('${a.id}','${opt}')">${opt}</div>`;
+      });
+      html += `</div>`;
+    } else if (a.type === 'text') {
+      html += `
+          <textarea placeholder="${a.placeholder || ''}"
+                    oninput="state.values['${a.id}']=this.value">${esc(state.values[a.id])}</textarea>
+      `;
+    } else if (a.type === 'confirm') {
+      html += `<div class="adj-hint" style="color:var(--accent)">${S.confirm_hint}</div>`;
+    }
+
+    html += `</div></div>`;
+  });
+
+  html += `</div>`;
+
+  /* Reference tracks (collapsed) */
+  html += `
+    <div class="ref-section">
+      <div class="ref-toggle" onclick="toggleRef()">
+        <span class="ref-arrow ${state.refOpen ? 'open' : ''}">▶</span>
+        ${S.ref_toggle}
+      </div>
+      <div class="ref-body ${state.refOpen ? 'open' : ''}">
+        <div class="ref-track">
+          <div class="ref-track-label"><div class="ref-dot" style="background:var(--voice-color)"></div>${S.ref_voice}</div>
+          <audio controls preload="none" src="${data.voice.audioUrl}"></audio>
+          <div class="ref-params">
+            <span class="ref-param">${esc(data.voice.voiceId)}</span>
+            <span class="ref-param">${S.ref_speed} ${data.voice.speed}x</span>
+            <span class="ref-param">${S.ref_gain} ${data.voice.gain}x</span>
+            <span class="ref-param">${S.ref_delay} ${data.voiceDelay}s</span>
+          </div>
+        </div>
+  `;
+  if (data.nature) {
+    html += `
+        <div class="ref-track">
+          <div class="ref-track-label"><div class="ref-dot" style="background:var(--nature-color)"></div>${esc(data.nature.category)}</div>
+          <audio controls preload="none" src="${data.nature.audioUrl}"></audio>
+          <div class="ref-params"><span class="ref-param">${S.ref_volume} ${data.nature.volume}x</span></div>
+        </div>
+    `;
+  }
+  if (data.music) {
+    html += `
+        <div class="ref-track">
+          <div class="ref-track-label"><div class="ref-dot" style="background:var(--music-color)"></div>${S.ref_bgm}</div>
+          <audio controls preload="none" src="${data.music.audioUrl}"></audio>
+          <div class="ref-params"><span class="ref-param">${S.ref_volume} ${data.music.volume}x</span></div>
+        </div>
+    `;
+  }
+  html += `</div></div>`;
+
+  /* Action buttons */
+  const n = activeCount();
+  html += `
+    <div class="actions">
+      <button class="btn btn-success" id="btn-lgtm" onclick="submitFeedback(true)">${S.btn_export}</button>
+      <button class="btn btn-primary" id="btn-submit" onclick="submitFeedback(false)" ${n === 0 ? 'disabled' : ''}>
+        ${S.btn_submit}<span class="btn-count">${n > 0 ? '(' + n + S.btn_count_suffix + ')' : ''}</span>
+      </button>
+    </div>
+  `;
+
+  document.getElementById('adj').innerHTML = html;
+}
+
+function toggleAdj(id) {
+  state.active[id] = !state.active[id];
+  renderAdj();
+}
+
+function onChoice(id, opt) {
+  state.values[id] = state.values[id] === opt ? null : opt;  // toggle
+  renderAdj();
+}
+
+function toggleRef() {
+  state.refOpen = !state.refOpen;
+  const arrow = document.querySelector('.ref-arrow');
+  const body = document.querySelector('.ref-body');
+  arrow.classList.toggle('open');
+  body.classList.toggle('open');
+}
+
+function collectFeedback(isLGTM) {
+  const lines = [];
+  const settings = { overall_satisfied: isLGTM };
+
+  if (isLGTM) {
+    return { feedback: S.feedback_satisfied, settings };
+  }
+
+  ADJUSTMENTS.forEach(a => {
+    if (!state.active[a.id]) return;
+    if (a.hidden && a.hidden()) return;
+
+    if (a.type === 'choice') {
+      const chosen = state.values[a.id];
+      settings[a.id] = chosen;
+      lines.push(`- **${a.label}**：${chosen || S.feedback_checked_no_choice}`);
+    } else if (a.type === 'text') {
+      const text = (state.values[a.id] || '').trim();
+      settings[a.id] = text;
+      lines.push(`- **${a.label}**：${text || S.feedback_checked_no_detail}`);
+    } else if (a.type === 'confirm') {
+      settings[a.id] = true;
+      lines.push(`- **${a.label}**`);
+    }
+  });
+
+  return {
+    feedback: lines.length > 0 ? lines.join('\n') : S.feedback_needs_adjustment,
+    settings,
+  };
+}
+
+async function submitFeedback(isLGTM) {
+  const btnLgtm = document.getElementById('btn-lgtm');
+  const btnSubmit = document.getElementById('btn-submit');
+  btnLgtm.disabled = true;
+  btnSubmit.disabled = true;
+
+  const { feedback, settings } = collectFeedback(isLGTM);
+
+  try {
+    const res = await fetch('/feedback', {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({ feedback, settings })
+    });
+    const result = await res.json();
+    if (result.ok) {
+      document.getElementById('hero').innerHTML = `
+        <div class="success-msg">
+          <h2>${isLGTM ? S.success_export_title : S.success_adjust_title}</h2>
+          <p>${isLGTM ? S.success_export_desc : S.success_adjust_desc}</p>
+        </div>
+      `;
+      document.getElementById('adj').innerHTML = '';
+    }
+  } catch (e) {
+    btnLgtm.disabled = false;
+    btnSubmit.disabled = false;
+    alert(S.submit_error);
+  }
+}
+
+render();
+</script>
+</body>
+</html>
@@ -0,0 +1,8 @@
+助眠与放松音频创作工具，用于生成沉浸式睡眠场景音频内容，
+覆盖三种内容类型：
+引导冥想、睡前故事，以及 ASMR（耳语加白噪音）。
+通过适合 ASMR 的声音进行旁白，结合大语言模型驱动的智能停顿插入、
+自然环境音素材与 AI 背景音乐，最终混音输出完整音频作品。
+触发词包括：sleep audio、relaxation、meditation、bedtime story、
+ambient、soundscape、ASMR、white noise、guided meditation、
+助眠、放松音频、冥想、睡前故事、氛围音、白噪音、引导冥想。
@@ -0,0 +1,112 @@
+# Category-Music Prompt Mapping
+
+Maps nature sound categories to background music prompts.
+Music serves as a bed layer — slow tempo, low volume, simple melody, mood-matched.
+
+## Indoor
+
+### rain
+```
+soft gentle piano, lo-fi ambient, cozy rainy day, warm intimate atmosphere, very slow tempo, minimal, dreamy, reverb
+```
+
+### thunder
+```
+dark ambient pads, deep cello drone, cinematic tension, low rumbling bass, very slow, atmospheric, mysterious
+```
+
+### fireplace
+```
+warm acoustic guitar fingerpicking, soft cello, cozy winter evening, intimate, gentle, slow tempo, folk-inspired
+```
+
+### cafe
+```
+jazz piano trio, brushed drums, walking bass, warm, conversational tempo, bossa nova influence, relaxed
+```
+
+### clock
+```
+minimal piano, music box melody, gentle arpeggios, clockwork rhythm, delicate, whimsical, soft
+```
+
+### keyboard
+```
+lo-fi hip hop beats, soft piano chords, vinyl crackle texture, chill, study music, mellow, steady groove
+```
+
+### vinyl
+```
+vintage jazz, warm analog synth, retro lounge, soft saxophone, nostalgic, old-fashioned, gentle swing
+```
+
+## Nature
+
+### ocean
+```
+ambient pad synths, gentle harp arpeggios, slow ethereal, spacious, meditative, new age, flowing
+```
+
+### river
+```
+acoustic guitar harmonics, soft marimba, gentle flute, nature-inspired, flowing tempo, peaceful, transparent
+```
+
+### forest
+```
+soft flute melody, gentle strings, peaceful morning, ambient, minimal, slow, nature-inspired, woodland
+```
+
+### bird
+```
+light acoustic guitar, gentle glockenspiel, pastoral, morning dew, airy, bright, delicate, hopeful
+```
+
+### cricket
+```
+warm ambient pads, soft guitar, summer night, lazy tempo, dreamy, nostalgic, gentle hum, twilight
+```
+
+### wind
+```
+ambient drone, soft piano chords, ethereal pads, spacious, contemplative, very slow, vast, solitary
+```
+
+### snow
+```
+celesta, soft strings, music box, winter silence, crystalline, delicate, cold beauty, sparse, Nordic
+```
+
+## Urban / Transport
+
+### train
+```
+soft rhythmic guitar, gentle percussion matching train rhythm, journey, nostalgic, folk, storytelling, wanderlust
+```
+
+### city
+```
+smooth jazz, muted trumpet, urban night, sophisticated, laid-back groove, city lights, cool
+```
+
+### subway
+```
+electronic ambient, soft synth bass, underground pulse, minimal techno influence, muted, urban, hypnotic
+```
+
+## Special
+
+### whitenoise
+```
+deep ambient drone, very slow evolving pad, binaural texture, meditative, theta waves, healing, minimal
+```
+
+### underwater
+```
+deep sub bass, reverse reverb piano, aquatic textures, slow motion, otherworldly, blue, submerged, echoing
+```
+
+### space
+```
+cosmic ambient, deep space synth pads, sci-fi atmosphere, vast emptiness, stars, ethereal choir, timeless, infinite
+```
@@ -0,0 +1,221 @@
+#!/usr/bin/env python3
+"""启动助眠音频混合预览服务器。
+
+用法：
+  python3 render_mix_preview.py <preview_data.json> [--port PORT]
+
+流程：
+  1. 读取 JSON 数据文件（包含各轨道路径、参数配置）
+  2. 注入 HTML 模板，启动本地 HTTP 服务器
+  3. 在浏览器中打开预览页面
+  4. 用户提交反馈后，写入 feedback.md + mix_settings.json，服务器自动退出
+"""
+
+import argparse
+import base64
+import json
+import mimetypes
+import os
+import sys
+import threading
+import webbrowser
+from http.server import HTTPServer, BaseHTTPRequestHandler
+from pathlib import Path
+from socketserver import ThreadingMixIn
+
+SKILL_DIR = Path(__file__).resolve().parent.parent
+TEMPLATE_DIR = SKILL_DIR / "html"
+
+
+class ThreadingHTTPServer(ThreadingMixIn, HTTPServer):
+    daemon_threads = True
+
+
+class PreviewHandler(BaseHTTPRequestHandler):
+    html_content: str = ""
+    feedback_path: Path = Path("feedback.md")
+    settings_path: Path = Path("mix_settings.json")
+    server_ref: HTTPServer | None = None
+
+    def do_GET(self):
+        if self.path.startswith("/audio/"):
+            self._serve_audio()
+            return
+        self.send_response(200)
+        self.send_header("Content-Type", "text/html; charset=utf-8")
+        self.end_headers()
+        self.wfile.write(self.html_content.encode("utf-8"))
+
+    def _serve_audio(self):
+        encoded = self.path[len("/audio/"):]
+        try:
+            audio_path = Path(base64.urlsafe_b64decode(encoded).decode("utf-8"))
+        except Exception:
+            self.send_error(400, "invalid audio path")
+            return
+        if not audio_path.exists():
+            self.send_error(404, "audio not found")
+            return
+        mime = mimetypes.guess_type(str(audio_path))[0] or "audio/mpeg"
+        file_size = audio_path.stat().st_size
+
+        try:
+            self._send_audio(audio_path, mime, file_size)
+        except BrokenPipeError:
+            pass
+
+    def _send_audio(self, audio_path, mime, file_size):
+        range_header = self.headers.get("Range")
+        if range_header:
+            # Parse Range: bytes=start-end
+            range_spec = range_header.replace("bytes=", "")
+            parts = range_spec.split("-")
+            start = int(parts[0]) if parts[0] else 0
+            end = int(parts[1]) if parts[1] else file_size - 1
+            end = min(end, file_size - 1)
+            length = end - start + 1
+
+            self.send_response(206)
+            self.send_header("Content-Type", mime)
+            self.send_header("Content-Range", f"bytes {start}-{end}/{file_size}")
+            self.send_header("Content-Length", str(length))
+            self.send_header("Accept-Ranges", "bytes")
+            self.send_header("Cache-Control", "max-age=3600")
+            self.end_headers()
+
+            with open(audio_path, "rb") as f:
+                f.seek(start)
+                self.wfile.write(f.read(length))
+        else:
+            self.send_response(200)
+            self.send_header("Content-Type", mime)
+            self.send_header("Content-Length", str(file_size))
+            self.send_header("Accept-Ranges", "bytes")
+            self.send_header("Cache-Control", "max-age=3600")
+            self.end_headers()
+
+            with open(audio_path, "rb") as f:
+                while True:
+                    chunk = f.read(65536)
+                    if not chunk:
+                        break
+                    self.wfile.write(chunk)
+
+    def do_POST(self):
+        if self.path == "/feedback":
+            self._handle_feedback()
+        else:
+            self.send_error(404)
+
+    def _handle_feedback(self):
+        length = int(self.headers.get("Content-Length", 0))
+        body = self.rfile.read(length).decode("utf-8")
+        data = json.loads(body)
+        feedback = data.get("feedback", "").strip()
+        settings = data.get("settings", {})
+
+        self.send_response(200)
+        self.send_header("Content-Type", "application/json")
+        self.end_headers()
+
+        if feedback:
+            self.feedback_path.write_text(feedback, encoding="utf-8")
+            if settings:
+                self.settings_path.write_text(
+                    json.dumps(settings, ensure_ascii=False, indent=2),
+                    encoding="utf-8",
+                )
+            self.wfile.write(json.dumps({"ok": True}).encode())
+            print(f"FEEDBACK_FILE={self.feedback_path}")
+            if settings:
+                print(f"SETTINGS_FILE={self.settings_path}")
+            threading.Timer(0.5, self._shutdown).start()
+        else:
+            self.wfile.write(json.dumps({"ok": False, "msg": "empty"}).encode())
+
+    def _shutdown(self):
+        if self.server_ref:
+            self.server_ref.shutdown()
+
+    def log_message(self, format, *args):
+        pass
+
+
+def to_audio_url(abs_path: str) -> str:
+    """将绝对路径转为 /audio/<base64> URL"""
+    encoded = base64.urlsafe_b64encode(abs_path.encode()).decode()
+    return f"/audio/{encoded}"
+
+
+def prepare_data(raw_data: dict) -> dict:
+    """将各轨道的绝对路径转为可访问的 /audio/ URL。
+
+    数据结构：
+    - voice.path → voice.audioUrl
+    - nature.path → nature.audioUrl
+    - music.path → music.audioUrl (可选)
+    - mixed.path → mixed.audioUrl
+    """
+    if raw_data.get("voice") and raw_data["voice"].get("path"):
+        raw_data["voice"]["audioUrl"] = to_audio_url(raw_data["voice"]["path"])
+
+    if raw_data.get("nature") and raw_data["nature"].get("path"):
+        raw_data["nature"]["audioUrl"] = to_audio_url(raw_data["nature"]["path"])
+
+    if raw_data.get("music") and raw_data["music"].get("path"):
+        raw_data["music"]["audioUrl"] = to_audio_url(raw_data["music"]["path"])
+
+    if raw_data.get("mixed") and raw_data["mixed"].get("path"):
+        raw_data["mixed"]["audioUrl"] = to_audio_url(raw_data["mixed"]["path"])
+
+    return raw_data
+
+
+def main():
+    parser = argparse.ArgumentParser(description="启动助眠音频混合预览服务器")
+    parser.add_argument("data_file", help="JSON 数据文件路径")
+    parser.add_argument("--port", type=int, default=0, help="端口号，默认自动分配")
+    parsed = parser.parse_args()
+
+    data_path = Path(parsed.data_file).resolve()
+    if not data_path.exists():
+        print(f"ERROR: 数据文件不存在: {data_path}")
+        sys.exit(1)
+
+    raw_data = json.loads(data_path.read_text(encoding="utf-8"))
+    preview_data = prepare_data(raw_data)
+
+    template_path = TEMPLATE_DIR / "preview-mix_feedback.html"
+    if not template_path.exists():
+        print(f"ERROR: 模板不存在: {template_path}")
+        sys.exit(1)
+
+    template = template_path.read_text(encoding="utf-8")
+    injected = template.replace(
+        "const PREVIEW_TYPE = null;  /* __PREVIEW_TYPE__ */",
+        'const PREVIEW_TYPE = "mix_feedback";',
+    ).replace(
+        "const PREVIEW_DATA = null;  /* __PREVIEW_DATA__ */",
+        f"const PREVIEW_DATA = {json.dumps(json.dumps(preview_data, ensure_ascii=False))};",
+    )
+
+    feedback_dir = data_path.parent
+    server = ThreadingHTTPServer(("127.0.0.1", parsed.port), PreviewHandler)
+    port = server.server_address[1]
+
+    PreviewHandler.html_content = injected
+    PreviewHandler.feedback_path = feedback_dir / "feedback.md"
+    PreviewHandler.settings_path = feedback_dir / "mix_settings.json"
+    PreviewHandler.server_ref = server
+
+    url = f"http://127.0.0.1:{port}"
+    print(f"PREVIEW_URL={url}")
+    print("等待用户提交...")
+
+    webbrowser.open(url)
+    server.serve_forever()
+    print("服务器已停止")
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,89 @@
+#!/usr/bin/env python3
+"""
+验证 nature/ 目录中的素材文件，输出类别识别结果和元数据。
+用法：python validate_nature.py <nature_dir>
+"""
+import os, subprocess, sys, json
+
+CATEGORY_KEYWORDS = {
+    "rain": ["rain", "雨"],
+    "thunder": ["thunder", "雷"],
+    "fireplace": ["fireplace", "fire", "壁炉", "篝火"],
+    "ocean": ["ocean", "wave", "sea", "海浪", "海"],
+    "river": ["river", "stream", "creek", "brook", "waterfall", "溪流", "河", "瀑布"],
+    "forest": ["forest", "森林"],
+    "bird": ["bird", "鸟"],
+    "cricket": ["cricket", "cicada", "蟋蟀", "蝉"],
+    "wind": ["wind", "风"],
+    "snow": ["snow", "雪"],
+    "cafe": ["cafe", "coffee", "咖啡"],
+    "clock": ["clock", "tick", "钟"],
+    "keyboard": ["keyboard", "typing", "键盘"],
+    "vinyl": ["vinyl", "record", "黑胶"],
+    "train": ["train", "rail", "火车", "铁轨"],
+    "city": ["city", "traffic", "street", "城市", "街道"],
+    "subway": ["subway", "metro", "地铁"],
+    "whitenoise": ["whitenoise", "white_noise", "白噪音"],
+    "underwater": ["underwater", "bubble", "水下"],
+    "space": ["space", "cosmic", "太空"],
+}
+
+AUDIO_EXTS = {".mp3", ".wav", ".m4a", ".flac", ".ogg", ".aac"}
+
+
+def get_duration(path):
+    try:
+        r = subprocess.run(
+            ["ffprobe", "-v", "quiet", "-show_entries", "format=duration", "-of", "csv=p=0", path],
+            capture_output=True, text=True,
+        )
+        return float(r.stdout.strip())
+    except Exception:
+        return 0.0
+
+
+def detect_category(filename):
+    name = filename.lower()
+    for cat, keywords in CATEGORY_KEYWORDS.items():
+        for kw in keywords:
+            if kw in name:
+                return cat
+    return None
+
+
+def main():
+    nature_dir = sys.argv[1] if len(sys.argv) > 1 else "."
+
+    files = sorted([
+        f for f in os.listdir(nature_dir)
+        if os.path.splitext(f)[1].lower() in AUDIO_EXTS
+    ])
+
+    if not files:
+        print(json.dumps({"error": "no audio files found", "dir": nature_dir}))
+        sys.exit(1)
+
+    results = []
+    for f in files:
+        path = os.path.join(nature_dir, f)
+        cat = detect_category(f)
+        dur = get_duration(path)
+        results.append({
+            "filename": f,
+            "category": cat,
+            "duration_s": round(dur, 1),
+            "size_mb": round(os.path.getsize(path) / 1024 / 1024, 1),
+            "needs_category": cat is None,
+        })
+
+    summary = {
+        "total": len(results),
+        "categorized": sum(1 for r in results if r["category"]),
+        "uncategorized": sum(1 for r in results if not r["category"]),
+        "files": results,
+    }
+    print(json.dumps(summary, ensure_ascii=False, indent=2))
+
+
+if __name__ == "__main__":
+    main()