@renoise/plugin 0.2.1

package/skills/tiktok-content-maker/examples/dress-demo.md

@@ -0,0 +1,86 @@
+# Full Example: Keep Resistance Loop Bands — 15s E-commerce Video
+
+## Input
+
+- Product image: Keep brand resistance loop bands, 3-pack (pink/blue/green), pastel macaron colors
+- Model reference: Athletic female, blonde ponytail, sports bra + fitted shorts (for analysis only, NOT uploaded to Renoise)
+
+## Product Analysis Result
+
+```json
+{
+  "product": {
+    "type": "Resistance loop bands",
+    "color": "Pink 10lb, Blue 15lb, Mint green 20lb",
+    "material": "TPE elastic, matte finish, soft and skin-friendly",
+    "highlights": "Three bands with progressive resistance for training, foldable and portable, pastel macaron color scheme",
+    "brand_tone": "Youthful athletic, trendy fitness"
+  },
+  "model": {
+    "gender": "female",
+    "age_range": "25-30",
+    "hair": "Blonde high ponytail",
+    "outfit": "Black sports bra + black fitted shorts",
+    "vibe": "Confident, energetic, professional fitness feel"
+  },
+  "scene_suggestions": [
+    "Bright modern living room morning workout",
+    "Hotel room travel fitness",
+    "Bedroom bedtime stretching"
+  ],
+  "selling_points": [
+    "Three resistance levels for progressive training, suitable for beginners to advanced",
+    "Compact and portable, folds to fit in a bag for anytime/anywhere workouts",
+    "Pastel macaron color scheme, won't roll up during use"
+  ]
+}
+```
+
+## Generated Script
+
+### 1. Video Prompt (English, with dialogue)
+
+> The product is a set of three Keep brand elastic resistance loop bands — flat, wide, smooth matte TPE material with a soft rubbery texture, each band approximately 5cm wide and forming a closed loop. Colors: pastel pink (lightest resistance), sky blue (medium), mint green (heaviest). Each band has a small white "Keep" logo printed on the surface. The bands must match the reference image exactly in color, width, shape, material finish, and logo placement throughout every frame of the video. A fit young woman in her mid-twenties with blonde hair in a high ponytail, light tan skin, athletic build, wearing a black sports bra and black fitted shorts, holds the three pastel-colored Keep resistance bands fanned out in her hand — camera starts extreme close-up on the bands showing their flat wide shape and matte surface then whip pans up to her face as she says "Stop scrolling — I threw out all my gym equipment for these three bands." Morning sunlight from a large window catches the smooth TPE finish. Camera does a fast snap dolly in on her hands as she stretches the blue band taut, the flat wide band maintaining its shape and thickness as it stretches, she says "Ten, fifteen, twenty pounds — I started pink, now I am on green, and they never roll up on you." She has the mint green band already looped around both ankles, camera pulls back to medium shot as she performs side leg raises, the wide flat band visible around her ankles keeping its shape, she says "I do legs in my living room, arms on work trips — they fold smaller than my phone" while transitioning into a squat pulse with the pink band above her knees. Without stopping she grabs all three bands, folds them into a tiny square and tucks them into a small gym bag pocket, camera pushes in tight, then she looks straight into the camera with a knowing grin and says "Honestly the best forty bucks I have spent this year," the pastel colors pop against her black outfit, warm golden backlight creates a soft halo, frame holds steady.
+
+### 2. Dialogue Script (English)
+
+```
+[0-3s]   "Stop scrolling — I threw out all my gym equipment for these three bands."
+[3-8s]   "Ten, fifteen, twenty pounds — I started pink, now I'm on green, and they never roll up on you."
+[8-12s]  "I do legs in my living room, arms on work trips — they fold smaller than my phone."
+[12-15s] "Honestly the best forty bucks I've spent this year."
+```
+
+### 3. BGM / Sound Design Suggestions
+
+- **BGM**: High-energy trap-pop beat, BPM 125-135, bass drop synced with the 0s hook
+- **Sound effects**:
+  - [0s] Band snap/recoil sound — paired with opening impact
+  - [3s] Fast whoosh transition
+  - [8s] Workout rhythm drum accent
+  - [12s] Bass swell + freeze-frame hit
+
+## Renoise Submission
+
+```bash
+# 1. Upload product image (product image only, NOT the model image)
+node renoise-cli.mjs material upload <product-image-path>
+# → returns material id, e.g. 194
+
+# 2. Submit task (with product image material, all-in-one)
+node renoise-cli.mjs task generate \
+  --prompt "<Video Prompt above>" \
+  --model renoise-2.0 --duration 15 --ratio 9:16 \
+  --tags ecom,keep,resistance-band \
+  --materials "194:image1"
+```
+
+## Multi-Scene Batch Generation
+
+Reuse the same material ID for the same product, swap scenes to generate multiple videos:
+
+| Scene | Hook Dialogue | Scene Keywords |
+|-------|--------------|----------------|
+| Outdoor park morning workout | "This tiny thing replaced my entire gym bag." | sunlit park lawn, golden hour, dewy grass |
+| Hotel room travel fitness | "Business trip day three and I still have not skipped a workout." | hotel room, city skyline, suitcase |
+| Bedroom bedtime stretching | "My nighttime routine that actually changed my body." | cozy bedroom, string lights, yoga mat |

package/skills/tiktok-content-maker/references/ecom-prompt-guide.md

@@ -0,0 +1,266 @@
+# E-commerce Short Video Prompt Guide
+
+## 15-Second E-commerce Video Prompt Template
+
+### Core Structure: One Continuous 15-Second Narrative
+
+Unlike Apple's "Don't Blink" rapid-cut style, e-commerce product videos are **one continuous shot** that uses camera movement changes to showcase the product and model.
+
+**Writing order**: Timeline narrative + Ad-6D elements interspersed
+
+```
+[Opening 0-3s] HOOK — Product MUST appear in Frame 1 + fast camera movement + start speaking immediately. Never build up slowly.
+[Showcase 3-8s] Product close-up + material details + model interaction
+[Scene 8-12s] Lifestyle scenario + usage effect + atmosphere
+[Close 12-15s] Model faces camera + product freeze frame + natural ending
+```
+
+### Product Anchoring (Start of Prompt, One Sentence)
+
+Product appearance is conveyed by the reference image. The prompt only needs **one sentence** stating what the product is + its use case:
+
+```
+The product is a [brand] [product type] for [primary use case], shown in the reference image.
+The product must match the reference image exactly in every frame. Do not invent any packaging, box, or container unless the reference image shows one.
+```
+
+**Examples**:
+- `The product is a K brand lightweight gym tote bag for fitness and daily commute, shown in the reference image.`
+- `The product is a Keep peanut-shaped massage ball for back and muscle recovery, shown in the reference image.`
+
+**Key**: Do not repeat product color, material, shape, or logo descriptions in the prompt — that information is already in the reference image. Save prompt space for the hook and visual narrative.
+
+### Model Consistency Description
+
+Immediately after the product anchoring, anchor the model's appearance:
+
+```
+A [age_range]-year-old [gender] with [hair description], [skin tone], [body type], wearing [outfit description]...
+```
+
+**Note**: Models are described entirely through text — never upload real person reference images (privacy detection will block images containing realistic human faces).
+
+## Category-Specific Keywords
+
+### Clothing
+
+- **Fabric**: flowing silk, crisp cotton, soft cashmere, stretchy knit, lightweight chiffon, structured tweed
+- **Motion**: fabric sways gently, hem flutters in breeze, pleats catch light, drape follows body movement
+- **Showcase**: twirls to show full skirt volume, adjusts collar detail, runs fingers along seam
+- **Scene**: sunlit café terrace, cherry blossom garden path, minimalist white studio, golden hour rooftop
+
+### Electronics
+
+- **Material**: anodized aluminum, gorilla glass surface, matte finish, chamfered edges catch light
+- **Motion**: screen illuminates face, finger glides across display, device rotates to reveal thin profile
+- **Showcase**: holds up to camera showing screen, taps interface with precision, places on wireless charger
+- **Scene**: modern desk setup, coffee shop workspace, commuter holding device, bedside nightstand
+
+### Beauty
+
+- **Texture**: dewy finish, velvety matte, glossy sheen, shimmering particles, creamy texture
+- **Motion**: applies with brush stroke, blends with fingertip, lips press together, eyelids flutter
+- **Showcase**: close-up of application, before-after glow, product swatch on skin, mirror reflection
+- **Scene**: vanity mirror with ring light, bathroom morning routine, getting ready for night out
+
+### Food
+
+- **Texture**: steam rises, sauce glistens, crispy golden crust, juice drips, cheese stretches
+- **Motion**: pours into bowl, breaks apart to reveal filling, scoops with spoon, bites with satisfaction
+- **Showcase**: overhead flat lay, cross-section reveal, slow-motion pour, garnish placement
+- **Scene**: rustic kitchen counter, outdoor picnic, cozy dining table, street food stall
+
+### Home
+
+- **Material**: warm wood grain, soft linen texture, smooth ceramic, brushed brass hardware
+- **Motion**: hand caresses surface, opens drawer smoothly, arranges on shelf, light shifts across surface
+- **Showcase**: styled vignette, before-after room transformation, detail close-up, scale with hand
+- **Scene**: morning light through curtains, minimalist living room, cozy bedroom corner, modern kitchen
+
+## Model-Product Interaction Vocabulary
+
+### General Actions
+- **Showcase**: holds up to camera, presents with both hands, turns to show different angle
+- **Touch**: runs fingers along, gently touches, traces the outline of
+- **Wear/Use**: puts on, adjusts, styles with
+- **Activate**: opens, activates, applies
+- **Emotion**: smiles confidently, looks surprised, nods approvingly, expresses delight
+
+### Clothing-Specific
+- twirls gracefully, walks toward camera, poses with hand on hip, flips hair to show neckline, adjusts sleeve cuff, smooths fabric over hip, turns to reveal back detail
+
+### Electronics-Specific
+- unboxes with anticipation, swipes through interface, holds up comparing to face size, places in pocket to show portability, tilts to catch light on screen
+
+### Beauty-Specific
+- applies with practiced motion, checks reflection, touches cheek feeling texture, pouts showing lip color, blinks showing eye makeup
+
+## Dialogue Writing Guidelines
+
+**Core principle**: Dialogue must be in English, embedded within the video prompt. Do not output subtitles separately.
+
+**Embedding format** (forced lip-sync):
+```
+Spoken dialogue (say EXACTLY, word-for-word): "..."
+Mouth clearly visible when speaking, lip-sync aligned.
+```
+Using `Spoken dialogue (say EXACTLY, word-for-word):` instead of simply `says "..."` significantly improves lip-sync accuracy. Follow each dialogue line with `Mouth clearly visible when speaking, lip-sync aligned.` to ensure the mouth is visible.
+
+**Style**: Best-friend casual tone — like recommending to a friend, not reading ad copy. Every sentence carries specific information (numbers, comparisons, usage scenarios) — no filler.
+
+### Hook Dialogue (0-3s) — First Line to Stop the Scroll
+
+**Subversive** (most effective):
+- "Stop scrolling — I threw out all my gym equipment for these three bands."
+- "This tiny thing replaced my entire gym bag."
+- "Business trip day three and I still have not skipped a workout."
+
+**Personal experience**:
+- "My nighttime routine that actually changed my body."
+- "The one thing I have been recommending to literally everyone."
+
+**Key**: Start speaking fast, paired with fast camera movement (whip pan / snap dolly in). Never start slow.
+
+### Selling Point Dialogue (3-8s) — Specific Specs + Personal Experience
+
+- "Ten, fifteen, twenty pounds — I started pink, now I am on green, and they never roll up on you."
+- "Three resistance levels, folds flat, weighs literally nothing — this is my entire travel gym."
+- Must include: specific numbers + personal usage experience + differentiating advantage
+
+### Scene Dialogue (8-12s) — Where to Use + Portability / Versatility
+
+- "I do legs in my living room, arms on work trips — they fold smaller than my phone."
+- "Park, backyard, hotel balcony — I have zero excuses now."
+- Must include: at least 2 usage scenarios + portability/versatility
+
+### Closing Dialogue (12-15s) — Natural Personal Recommendation, No Hard Sell
+
+**Good closings** (recommended):
+- "Honestly the best forty bucks I have spent this year."
+- "Trust me just start — future you will be so grateful."
+- "Best thing I ever packed."
+- "You are welcome."
+
+**Avoid these closings** (too pushy):
+- ~~"Link below — grab yours before they sell out."~~
+- ~~"Click the link for a special discount."~~
+- ~~"Limited stock, hurry up!"~~
+
+## Hook Strategy (The First 3 Seconds)
+
+**Data**: 63% of high-CTR TikTok videos capture users in the first 3 seconds. Users make the "watch or swipe" decision in just **1.7 seconds**. Videos with 65%+ 3-second retention rate get **4-7x** more impressions.
+
+### Visual Hook Techniques (pick one for prompt opening)
+
+| Technique | Prompt Phrasing | Effect |
+|-----------|----------------|--------|
+| **Snap zoom-in** | `Camera snaps in extreme close-up on the [product]` | Product in Frame 1, maximum impact |
+| **Product slides in** | `The [product] slides into frame from the right` | Sudden appearance creates micro-suspense |
+| **Hand thrust** | `A hand thrusts the [product] toward the camera` | Direct, strong UGC feel |
+| **Close-up → wide reveal** | `Extreme macro on [texture detail], camera rapidly pulls back to reveal...` | Hook curiosity with detail, then reveal full picture |
+| **Whip pan in** | `Camera whip-pans with motion blur and lands on the [product]` | Strong rhythm, visual impact |
+
+### Hook Core Rules
+
+1. **Product MUST appear in Frame 1** — never start with someone walking, opening a door, or establishing the environment
+2. **Frame 1 MUST have motion** — static opening = instant swipe
+3. **Model MUST start speaking within the first 2 seconds** — audio retains more than visuals
+4. **Hook dialogue should feel like stopping a friend** — not reading ad copy
+
+### Hook Dialogue Formulas (ranked by effectiveness)
+
+1. **Result-first**: "This $30 bag replaced my gym bag AND my purse." — Show the result directly
+2. **Subversive**: "Stop carrying two bags to the gym — you only need this one." — Challenge existing habits
+3. **Social proof**: "200K people bought this last month and I finally get why." — FOMO
+4. **Pain point question**: "Why is your gym bag always so heavy?" — Hit the pain point directly
+5. **Personal story**: "I was that person with three bags until I found this." — Relatability
+
+## Camera Movement Pacing (15s Continuous Shot)
+
+```
+[0-3s]  HOOK — Product in Frame 1! Fast camera movement + start speaking immediately
+        Recommended: extreme close-up snap in / whip pan / product slides into frame
+        Avoid: camera slowly pushes in, person walking as buildup, empty establishing shot
+        Pacing: Complete the first camera change within 1-2 seconds
+
+[3-8s]  SHOWCASE — Close-up → medium shot transition, showcase product details
+        Recommended: fast snap dolly in on details,
+              camera orbits or slides to reveal texture
+
+[8-12s] SCENE — Pull back to medium/wide shot, show usage scenario
+        Recommended: camera pulls back to reveal full scene,
+              natural movement as model interacts with environment
+
+[12-15s] CLOSE — Return to medium shot, face camera, freeze
+         Recommended: camera pushes in tight, then settles,
+               model faces camera, product in frame,
+               frame holds steady (this final hold is important)
+```
+
+## BGM Instructions (Specified in Prompt)
+
+The video model can generate background music within the video. **Always add a BGM instruction at the end of the prompt**:
+
+```
+Background music: [genre/mood description], [tempo], [energy level].
+```
+
+**BGM Selection Guide**:
+
+| Product Category | Recommended BGM | Prompt Phrasing |
+|-----------------|-----------------|-----------------|
+| Sports / Fitness | Rhythmic electronic / lo-fi | `Background music: upbeat electronic lo-fi beat, medium-fast tempo, energetic and motivating.` |
+| Beauty / Skincare | Warm R&B / chill pop | `Background music: warm chill R&B, slow-medium tempo, soft and intimate.` |
+| Electronics | Clean minimal / tech | `Background music: clean minimal electronic, medium tempo, modern and sleek.` |
+| Fashion | Indie pop / trendy | `Background music: trendy indie pop, medium tempo, stylish and confident.` |
+| Home | Acoustic / ambient | `Background music: warm acoustic guitar, slow tempo, cozy and relaxing.` |
+| Food | Jazz / feel-good | `Background music: feel-good jazz, medium tempo, cheerful and appetizing.` |
+
+**Key**: BGM should match the video pacing — the hook segment needs energy, the closing can ease up. Music style should match the product's brand tone.
+
+## Prompt Writing Style: Director Dictation
+
+Prompts should be written as if a director is dictating on set — 6-10 English sentences, each doing only one thing. Avoid packing multiple actions into one long sentence.
+
+**Good writing** (one action/camera instruction per sentence):
+```
+Camera snaps in on a close-up of the pink peanut massage ball sitting on a yoga mat.
+A 25-year-old woman with a high ponytail and black leggings walks into frame.
+She picks up the ball and holds it up to the camera.
+Spoken dialogue (say EXACTLY, word-for-word): "This little thing saved my back after deadlifts."
+Mouth clearly visible when speaking, lip-sync aligned.
+She places the ball on the mat and lies down on it, rolling her spine.
+Camera pulls back to a medium shot showing the full living room scene.
+...
+```
+
+**Bad writing** (too many actions crammed in one sentence):
+```
+A woman enters carrying a pink ball while the camera pans and she says "..." as she lies down and rolls.
+```
+
+## Prompt Quality Checklist
+
+- [ ] Pure English director-dictation style paragraph (6-10 sentences, one thing per sentence)
+- [ ] **Product anchoring at the start** (one sentence: what the product is + use case + match reference image + no-packaging lock)
+- [ ] Model appearance anchoring description immediately after
+- [ ] **Dialogue uses `Spoken dialogue (say EXACTLY, word-for-word):` format** (English, 4 lines, best-friend casual tone)
+- [ ] `Mouth clearly visible when speaking, lip-sync aligned.` after each dialogue line
+- [ ] Hook dialogue in the first sentence, paired with fast camera movement
+- [ ] Closing dialogue is natural, no hard sell
+- [ ] Includes specific product material/texture descriptors
+- [ ] At least 3 camera movement changes
+- [ ] Includes lighting/atmosphere description
+- [ ] Model has clear physical interaction with product
+- [ ] Ending frame holds steady
+- [ ] Overall pacing: fast open → detailed showcase → scene → freeze
+- [ ] **BGM instruction at the end** (`Background music: [genre], [tempo], [energy]`)
+- [ ] Hook segment has product in Frame 1, no buildup
+
+## Renoise Submission Notes
+
+- **Must upload product image** as material (image1) — product accuracy improves significantly
+- **Never upload real person/model images** — privacy detection will block them (error: PrivacyInformation)
+- Model appearance is controlled entirely by prompt text description
+- Product images should ideally be clean white-background product photos, avoid images with marketing text overlays
+- For batch generation: upload the product image once, reuse the material ID and swap scenes/dialogue

package/skills/video-download/SKILL.md

@@ -0,0 +1,161 @@
+---
+name: video-download
+description: >
+  Downloads videos from YouTube, TikTok, Douyin, Bilibili, Instagram, XiaoHongShu and 1000+ platforms.
+  Primary: yt-dlp. Fallback: agent-browser + GreenVideo for Douyin/TikTok when yt-dlp fails.
+  Use when user says "download video", "save video", "grab video", "watermark-free download",
+  or pastes a video URL. Do NOT use for AI video generation or video editing.
+allowed-tools: Bash
+metadata:
+  author: renoise
+  version: 0.2.0
+  category: utility
+  tags: [download, youtube, tiktok, douyin, yt-dlp, greenvideo]
+---
+
+# Video Download
+
+Download videos from YouTube, TikTok, and other platforms to local MP4 files using yt-dlp. Handles format selection, platform-prefixed dedup, and TikTok cookie fallback automatically.
+
+## Prerequisites
+
+Verify yt-dlp is installed:
+
+```bash
+yt-dlp --version
+```
+
+If missing: `brew install yt-dlp` (macOS) or `pip install yt-dlp`.
+
+### Optional (for Douyin/TikTok fallback)
+
+- `agent-browser` installed globally (`npm install -g agent-browser`)
+- Chrome for Testing installed (`agent-browser install`)
+
+## Usage
+
+### Single Video
+
+Run the download script with the video URL:
+
+```bash
+bash ${CLAUDE_SKILL_DIR}/scripts/download-video.sh '<URL>'
+```
+
+The script handles everything automatically:
+- Extracts a platform-prefixed video ID (`yt-dQw4w9WgXcQ`, `tk-7571284267028729101`, `vid-aHR0cHM6Ly93d3`)
+- Saves to `resources/references/<video_id>.mp4`
+- Skips download if file already exists (dedup)
+- Retries TikTok downloads with `--cookies-from-browser chrome` on failure
+- Removes zero-byte leftovers from interrupted downloads
+
+### Custom Output Directory
+
+```bash
+bash ${CLAUDE_SKILL_DIR}/scripts/download-video.sh '<URL>' 'path/to/output'
+```
+
+### Batch Download
+
+Run the script in a loop:
+
+```bash
+for URL in '<URL1>' '<URL2>' '<URL3>'; do
+  bash ${CLAUDE_SKILL_DIR}/scripts/download-video.sh "$URL"
+done
+```
+
+## Script Output
+
+The script prints one of three status lines:
+
+| Output | Meaning |
+|--------|---------|
+| `ALREADY_EXISTS: <path>` | File already downloaded, skipped |
+| `DOWNLOADED: <path>` | Download succeeded |
+| `FAILED: <message>` | Download failed (exit code 1) |
+
+## Troubleshooting
+
+| Error | Solution |
+|-------|----------|
+| `HTTP Error 403` (TikTok/Douyin) | Script auto-retries with cookies. If still failing, use **GreenVideo Fallback** below |
+| `--max-filesize` skipped | Video exceeds 200M limit. Download manually with `-f 'best[height<=720]'` |
+| `is not a valid URL` | Ensure URL is wrapped in single quotes |
+| `Requested formats are incompatible` | yt-dlp auto-transcodes, no action needed |
+
+## Video ID Logic
+
+| Platform | Pattern | Example ID |
+|----------|---------|------------|
+| YouTube | `watch?v=`, `shorts/`, `embed/`, `youtu.be/` → 11-char ID | `yt-dQw4w9WgXcQ` |
+| TikTok | 15+ digit numeric ID in URL | `tk-7571284267028729101` |
+| Other | Base64url of URL, first 16 chars | `vid-aHR0cHM6Ly93d3` |
+
+---
+
+## GreenVideo Fallback (for Douyin/TikTok)
+
+When yt-dlp fails for Douyin or TikTok URLs (common with 403 errors or region-restricted content), use GreenVideo as a browser-based fallback. Requires `agent-browser` (see Prerequisites).
+
+### Decision Logic
+
+```
+User gives URL → Try yt-dlp first
+                   ↓ fails (403 / no result)
+               Is it Douyin/TikTok?
+                   ↓ yes
+               Use agent-browser + GreenVideo
+                   ↓ no
+               Report error, suggest manual download
+```
+
+### Step 1: Open GreenVideo
+
+```bash
+agent-browser open "https://greenvideo.cc/en/"
+```
+
+### Step 2: Paste URL and Parse
+
+```bash
+agent-browser snapshot
+# Find the input textbox ref (e.g., @e63)
+agent-browser fill <input-ref> "<video-url>"
+agent-browser click <start-button-ref>
+```
+
+Wait 5 seconds for parsing to complete, then snapshot to verify results.
+
+### Step 3: Extract Video Direct URL
+
+The video URL is stored in Nuxt state. Extract it with:
+
+```bash
+agent-browser eval "
+(function() {
+  const nuxtData = window.__NUXT__;
+  if (!nuxtData) return 'ERROR: No Nuxt data found';
+  const str = JSON.stringify(nuxtData);
+  const mp4Match = str.match(/https?:[^\\\"]*(?:mp4|video|play|aweme|douyinvod|bilivideo)[^\\\"]{0,500}/g);
+  if (mp4Match && mp4Match.length > 0) return mp4Match[0];
+  return 'ERROR: No video URL found in Nuxt state';
+})()
+"
+```
+
+### Step 4: Download and Cleanup
+
+```bash
+curl -L -o ~/Downloads/<filename>.mp4 "<extracted-video-url>"
+agent-browser close
+```
+
+### GreenVideo Troubleshooting
+
+| Issue | Solution |
+|-------|----------|
+| Parse fails or no result | Check URL is valid and publicly accessible. Some `/user/self` URLs need the direct video URL |
+| No video URL in Nuxt state | Try clicking download button and check for `<video>` elements: `agent-browser eval "document.querySelectorAll('video').length"` |
+| Download fails (403) | Video URLs expire quickly — extract and download immediately. Try adding `-H "Referer: https://greenvideo.cc/"` to curl |
+| agent-browser not responding | Run `agent-browser close` then `agent-browser open "https://greenvideo.cc/en/"` |

package/skills/video-download/scripts/download-video.sh

@@ -0,0 +1,91 @@
+#!/usr/bin/env bash
+# download-video.sh — Download video to resources/references/ with dedup and platform detection
+#
+# Usage:
+#   download-video.sh <URL> [output_dir]
+#
+# Features:
+#   - Extracts platform-prefixed video ID (yt-xxx, tk-xxx, vid-xxx)
+#   - Skips download if file already exists (dedup)
+#   - Auto-retries TikTok with --cookies-from-browser chrome on 403
+#   - Outputs final file path on success
+
+set -euo pipefail
+
+URL="${1:?Usage: download-video.sh <URL> [output_dir]}"
+OUTPUT_DIR="${2:-resources/references}"
+
+# --- Video ID extraction ---
+
+extract_video_id() {
+  local url="$1"
+
+  # YouTube: watch?v=, shorts/, embed/, youtu.be/
+  if [[ "$url" =~ (youtube\.com/(watch\?v=|shorts/|embed/)|youtu\.be/)([\w-]{11}|[A-Za-z0-9_-]{11}) ]]; then
+    local yt_id
+    yt_id=$(echo "$url" | grep -oE '(watch\?v=|shorts/|embed/|youtu\.be/)([A-Za-z0-9_-]{11})' | grep -oE '[A-Za-z0-9_-]{11}$')
+    echo "yt-${yt_id}"
+    return
+  fi
+
+  # TikTok: numeric ID (15+ digits)
+  if [[ "$url" =~ tiktok\.com ]]; then
+    local tk_id
+    tk_id=$(echo "$url" | grep -oE '[0-9]{15,}' | head -1)
+    if [[ -n "$tk_id" ]]; then
+      echo "tk-${tk_id}"
+      return
+    fi
+  fi
+
+  # Fallback: base64url-encoded URL, first 16 chars
+  local b64
+  b64=$(echo -n "$url" | base64 | tr '+/' '-_' | tr -d '=' | head -c 16)
+  echo "vid-${b64}"
+}
+
+VIDEO_ID=$(extract_video_id "$URL")
+OUTPUT="${OUTPUT_DIR}/${VIDEO_ID}.mp4"
+
+# --- Dedup check ---
+
+if [[ -f "$OUTPUT" && $(stat -f%z "$OUTPUT" 2>/dev/null || stat -c%s "$OUTPUT" 2>/dev/null) -gt 0 ]]; then
+  echo "ALREADY_EXISTS: $OUTPUT"
+  exit 0
+fi
+
+# Remove zero-byte leftover from interrupted download
+[[ -f "$OUTPUT" ]] && rm -f "$OUTPUT"
+
+mkdir -p "$OUTPUT_DIR"
+
+# --- Download ---
+
+YT_DLP_ARGS=(
+  -f 'bestvideo[ext=mp4]+bestaudio[ext=m4a]/best[ext=mp4]/best'
+  --merge-output-format mp4
+  --no-playlist
+  --max-filesize 200M
+  -o "$OUTPUT"
+)
+
+echo "Downloading: $URL"
+echo "Output: $OUTPUT"
+
+if yt-dlp "${YT_DLP_ARGS[@]}" "$URL"; then
+  echo "DOWNLOADED: $OUTPUT"
+  exit 0
+fi
+
+# --- TikTok 403 retry with cookies ---
+
+if [[ "$URL" =~ tiktok\.com ]]; then
+  echo "Retrying TikTok with browser cookies..."
+  if yt-dlp "${YT_DLP_ARGS[@]}" --cookies-from-browser chrome "$URL"; then
+    echo "DOWNLOADED: $OUTPUT"
+    exit 0
+  fi
+fi
+
+echo "FAILED: Could not download $URL"
+exit 1