@azerogluemin/ai-bootstrap 0.2.1

package/templates/skills/video-prompt-engineer/knowledge/subject-motion.md

@@ -0,0 +1,220 @@
+# Subject Motion — Video Generation
+
+> How subjects (people, animals, objects) should move in video prompts. AI models handle motion very differently from static images.
+
+## Motion types
+
+### Subtle motion (RECOMMENDED for stability)
+- Breathing
+- Blinking
+- Slight head turns
+- Hair movement in wind
+- Fabric rustle
+- Eyes shifting
+
+These work reliably in ALL AI video models.
+
+### Moderate motion
+- Walking
+- Standing up / sitting down
+- Reaching for object
+- Turning whole body
+- Talking (lip movement)
+- Pouring liquid
+- Opening door
+
+Most models handle these but watch for artifacts.
+
+### Complex motion
+- Running
+- Dancing
+- Fighting
+- Jumping
+- Throwing
+- Multi-person interaction
+- Sports
+
+Risky — models often produce distortions, weird limb behavior.
+
+### Extreme motion
+- High-speed action
+- Acrobatics
+- Vehicle stunts
+- Crowds in motion
+
+Avoid unless using Sora or Veo for short clips with careful prompting.
+
+## Motion description language
+
+### USE ACTION VERBS WITH MODIFIERS
+
+GOOD:
+```
+She slowly turns her head from right to left, eyes scanning the room.
+```
+
+GOOD:
+```
+He gradually raises his hand to his face, fingers brushing his temple.
+```
+
+BAD:
+```
+She moves her head.    // Too vague
+```
+
+BAD:
+```
+He does a thing.       // No specificity
+```
+
+### USE "BEGINS TO" FOR NATURAL TIMING
+
+GOOD:
+```
+She begins to walk slowly toward the door, taking three measured steps.
+```
+
+This gives the model timing cues — start state, action, end state.
+
+### SPECIFY START AND END STATE
+
+GOOD:
+```
+Starting position: standing still, hands at sides, looking forward.
+Action: slowly turns head 45° to the left, eyes following.
+End position: head turned, looking off-screen left, body still facing forward.
+```
+
+This gives clear motion vector.
+
+### AVOID INSTANT TRANSITIONS
+
+BAD:
+```
+He walks across the room.    // No timing — model guesses
+```
+
+GOOD:
+```
+He takes 3 slow steps across the room over the duration of the clip.
+```
+
+## Lip sync notes
+
+If subject is speaking:
+- Kling 2.x has best lip sync
+- Veo 3.1 is acceptable
+- Most others: avoid showing speaking from front (use OTS or back)
+
+For dialogue:
+```
+Camera: medium shot from front
+Subject speaks slowly, calmly: "I waited for you."
+Lip movements should match this dialogue.
+```
+
+## Multi-subject motion
+
+When two subjects move together:
+- Specify each independently
+- Specify spatial relationship
+- Specify timing relationship
+
+```
+Subject A (left of frame) raises her hand in greeting.
+Subject B (right of frame) responds by stepping forward and nodding.
+Action timing: A's gesture begins first, B responds 1 second later.
+```
+
+## Physics constraints
+
+AI models struggle with:
+- Defying gravity
+- Object permanence (object disappearing)
+- Mass/weight realism (light vs heavy objects move similarly)
+- Liquid dynamics (water, fire are tricky)
+- Rapid acceleration
+
+Use motion within natural human range.
+
+## Default for AI video
+
+When unsure: **minimal motion**. A subtle breath + slight head turn is more reliable than complex action.
+
+---
+
+## Director thinking — actions are AUTHORED, not extracted
+
+The video-prompt-engineer is not a passive interpreter of the script. The script gives the **frame** (location, mood, characters, camera intent). The **specific actions inside the frame are the director's decision** — and in the AI pipeline, the prompt writer IS the director.
+
+This means: if the script says "an old Baku street, a parked carriage, a man on it, two distant pedestrians" — it is **not enough** to write "the driver does something." You must decide, as a director:
+
+- **What is this character doing in this moment?** (waiting for market, returning from work, dropping off a passenger, just arrived)
+- **What is the inner story of the frame?** (a city waking up, an evening winding down, a moment before something changes)
+- **What concrete physical action shows that inner story?** (adjusting his cap against the chill, glancing down the street, lighting a pipe, checking a pocket watch)
+- **What does the animal do in this moment?** (resting head down, alert with ears forward, drinking from a trough, standing patiently)
+- **Why are the pedestrians walking down this street?** (residents going to work, mother and child returning home, two merchants in quiet conversation)
+
+Every prompt must carry a **specific, authored moment** — not a vague checklist of "natural motion." Vague prompts produce vague clips that feel like stock footage. Authored prompts produce clips that feel like cinema.
+
+### How to author the moment
+
+For every frame, write a one-line **inner story** before writing the action list:
+
+```
+Inner story: a quiet street wakes up at dawn — the man with the cart is
+waiting for market, the city is yawning into the day, light is climbing.
+```
+
+Then derive every action from that inner story:
+
+- The man's action **expresses** waiting + dawn (cap adjust against chill, long look toward the light)
+- The horse's action **expresses** patience + early morning (head low, weight shift, ear swivel)
+- The pedestrians' action **expresses** unhurried morning (calm walk, quiet conversation)
+- The environment's action **expresses** awakening (curtain stirs in opened window, lantern flame thinning in daylight, dust in golden light)
+
+Every action serves the inner story. Nothing is generic.
+
+### What to refuse in the brief
+
+If you receive a brief that only says "subject moves naturally" or "make it feel alive" — **push back**. Ask the script writer / director persona what the inner moment is, or, if you are the director in the pipeline, decide it yourself and state your decision in the AZ description. Do not generate a prompt with unauthored, generic motion.
+
+---
+
+## Forbidden actions need a real reason — never write groundless prohibitions
+
+When you write `the horse does not step, the driver does not turn his head, the pedestrians do not stop walking` — every one of those negatives must answer the question **"why?"** with a real, story-grounded reason. If you cannot answer it, you are not directing — you are restricting blindly.
+
+### The two prohibitions are universal — everything else needs justification
+
+In observational documentary grammar, only two prohibitions apply by default to every subject:
+
+1. **No direct camera awareness** — characters do not look at the camera, wave at it, perform for it, or break the fourth wall. The camera is observing; subjects do not know it is there.
+2. **No anachronistic / modern behavior** — period subjects do not check phones, make modern gestures, wear modern clothing, ride in cars not yet invented.
+
+Beyond these two, every additional "subject must not do X" needs a story reason:
+
+| Restriction | Valid reason (allow) | No reason (drop the restriction) |
+|---|---|---|
+| Horse does not bolt | Yes — bolting would tear the carriage out of frame, breaks composition | — |
+| Horse does not step at all | Where's the reason? Horses shift weight constantly. **DROP** |
+| Driver does not dismount | Yes — dismounting changes the staging, leaves an empty bench | — |
+| Driver does not turn his head | Where's the reason? Real humans look around. **DROP** |
+| Pedestrians do not stop | Yes — they're walking past, stopping would make them a focal point | — |
+| Pedestrians do not breathe | Where's the reason? They're alive. **DROP** |
+
+### The reflex to "lock everything down" is wrong
+
+A common failure mode: the prompt writer reads a poetic script line ("frozen time", "still photograph", "the world holds its breath") and reflexively writes a wall of negatives that turn every living subject into a statue. The result on screen is dead: a CGI freeze-frame, not a documentary.
+
+The right reflex is the opposite — **start from "everything is alive"** and only forbid what would genuinely damage the story (camera awareness, dramatic action that doesn't fit the inner moment, anachronism). Everything else stays.
+
+### Audit your own prompt
+
+Before you ship a prompt, read every negative line and ask: **"If I delete this restriction, what bad thing happens on screen?"**
+
+- If you can name a concrete bad outcome → keep the restriction
+- If your answer is "I don't know, I just felt like it shouldn't move" → delete the restriction
+
+This audit will cut 50–70% of the negatives in a typical first draft, and the resulting prompt will produce a clip that feels alive instead of taxidermied.

package/templates/skills/video-prompt-engineer/knowledge/temporal-consistency.md

@@ -0,0 +1,215 @@
+# Temporal Consistency — Video Generation
+
+> Frame-to-frame consistency. The hardest problem in AI video. Subject morphing, lighting flicker, object disappearing — all temporal failures.
+
+## What can go wrong
+
+### Subject morphing
+- Face slowly changes between frames
+- Outfit shifts color or detail
+- Body proportions vary
+
+### Lighting flicker
+- Brightness changes between frames
+- Color temperature shifts
+- Shadows move unnaturally
+
+### Object permanence loss
+- Held object disappears mid-clip
+- Background element vanishes
+- Hand goes through solid object
+
+### Background drift
+- Walls slowly distort
+- Texture patterns shift
+- Background subjects appear/disappear
+
+## Mitigation strategies
+
+### 1. Strong reference image (most important)
+For image-to-video models (Luma, Runway, Pika):
+- Provide highest-quality starting image
+- Image should be clean, sharp, well-composed
+- Avoid busy backgrounds (more to keep consistent)
+
+### 2. Short duration
+- 4-6 seconds: most stable
+- 6-8 seconds: usable
+- 8-10 seconds: degradation likely
+- 10+ seconds: significant artifacts
+
+### 3. Minimal motion
+- Less motion = more frames look similar = less drift
+- Static or subtle motion = highest stability
+- Complex action = highest drift
+
+### 4. Simple background
+- Plain walls, gradient backgrounds, blurred backgrounds = stable
+- Detailed textures, crowds, foliage = unstable
+
+### 5. Single subject focus
+- One person in frame: stable
+- Two people: doable
+- Three+ people: high drift risk
+
+### 6. Consistent lighting setup
+- Natural single source = stable
+- Multi-source mixed lighting = drift risk
+- Practical lights that flicker (candles, screens) = use cautiously
+
+### 7. Stable camera
+- Static = best
+- Single slow movement = ok
+- Complex movement = drift compounds
+
+## Prompt phrases that help
+
+```
+Maintain exact visual consistency throughout the clip.
+Subject's appearance must NOT change frame-to-frame.
+Lighting setup remains constant.
+Background details are static and unchanging.
+No transformations or morphing.
+```
+
+## Prompt phrases that hurt
+
+```
+Dynamic transitions
+Changing perspective
+Morphing scenery
+Time-lapse
+Multiple cuts
+```
+
+These trigger model into "creative" mode = inconsistency.
+
+## Continuity between clips
+
+When making multiple clips that should connect:
+
+```
+Clip 1: Subject in scene X with lighting Y, ends at position Z.
+Clip 2: Subject continues from position Z, lighting Y unchanged,
+scene X unchanged. NEW action begins.
+```
+
+Use last frame of Clip 1 as reference image for Clip 2.
+
+## Trade-off
+
+More motion = more cinematic feel BUT less consistency.
+Less motion = more stable BUT can feel like "moving photo".
+
+For AI video right now (2026), prefer stability over dynamism.
+
+## Quality check
+
+After generating, watch for:
+- Face stability frame-by-frame
+- Outfit color consistency
+- Background object permanence
+- Smooth motion (no jumps)
+- Consistent lighting
+
+If multiple failures: regenerate with simpler prompt.
+
+---
+
+## HARD RULE — Lit objects must stay lit consistently from frame 0 (no "re-ignition" artifact)
+
+**Failure mode**: when the start frame shows a candle, oil lamp, lantern, screen, fireplace, or any lit light source, image-to-video models (Kling, Runway, Hailuo, Pika) often **animate the lighting up** of that source across the clip — the flame grows, the glow brightens, the lit object appears to *ignite* or *power on* mid-clip. The viewer reads this as "the candle just lit itself," which breaks both physical realism and temporal continuity.
+
+This happens because the model interprets ambiguous "flicker softly" or "burns warmly" language as a *change in lighting* rather than a *stable lit state*.
+
+**Real-world cost**: this artifact is among the most expensive AI-video failures because (a) it nearly always recurs on regeneration if the prompt isn't restructured, (b) it costs the user real credit dollars per failed attempt, and (c) it is often hard to spot until the clip is fully rendered. Solve it with prompt structure, not negative-list whack-a-mole.
+
+### The MOST RELIABLE prompt structure for Ken Burns / photo-doc frames containing lit objects
+
+**Do NOT describe flames as "living", "flickering softly", or "burning" anywhere in the prompt.** Those positive words tell the model to animate the flame. Negative-prompt entries cannot cleanly override positive-prompt cues — model behavior follows positive cues first.
+
+Instead, restructure the prompt around the **"Frozen Still-Life Treatment"** pattern:
+
+```
+CRITICAL RULE — FROZEN STILL-LIFE TREATMENT:
+This clip is to be filmed AS IF it were a still photograph being slowly pushed into by the camera. The reference image is to be treated as a photograph — an unmoving artifact. The ONLY motion permitted in this clip is the camera's slow forward dolly push-in. EVERY OTHER pixel of the frame is to be rendered as completely frozen.
+
+This explicitly includes — but is not limited to:
+- The candle flames: TREATED AS PHOTOGRAPHIC ELEMENTS, NOT AS LIVING FLAMES. Zero flicker, zero growth, zero shrinkage, zero brightness change, zero position change, zero ignition, zero re-lighting, zero pulsation. The candle flames are visual ornaments in a still image — they have ALREADY existed in their exact rendered state at the moment the photograph was taken.
+- The lamp glow: FROZEN at exactly the brightness, color, and reach shown in the reference frame. The lamp does not warm up, ramp up, intensify, dim, or change in any way.
+- The overall scene exposure: identical at frame 0 and at frame final. No general fade-in, no scene-wide brightening, no exposure pump.
+
+Action — ONLY THE FOLLOWING MOTIONS ARE PERMITTED IN THIS CLIP. NO OTHER MOTION OF ANY KIND IS ALLOWED:
+1. [explicit motion 1, e.g. subject breathes once]
+2. [explicit motion 2, e.g. subject blinks once]
+3. The camera performs one single continuous very slow dolly push-in.
+
+All other elements in the frame are completely frozen — treat them as photographic elements that the camera is gently entering.
+```
+
+The two changes that make this work versus the previous (failing) pattern:
+
+| Pattern that FAILS | Pattern that WORKS |
+|---|---|
+| `lit white tapered candles ... real living candle flames flicker softly` | `white tapered candles ... visible as decorative composition elements` |
+| `warm flickering candle fill ... candle flames themselves flicker softly` | `light sources rendered as elements of a still photograph — they do not produce dynamic light` |
+| `cinematic period-drama documentary still that breathes` | `Ken Burns documentary photograph — slow push into still image. Museum diorama photograph.` |
+| Motion limitation implicit (via negative list) | Motion limitation **EXPLICIT** — a 1-2-3 numbered list of the only motions allowed, then "All other elements are completely frozen" |
+| "alive but reverent" | "completely frozen still photograph" |
+
+### Mandatory negative-prompt block (still required, but secondary defense)
+
+```
+candle flames flickering, candle flames moving, candle flames growing, candle flames growing larger, candle flames growing taller, candle flames shrinking, candle flames pulsing, candle flames pulsating, candle flames brightening, candle flames dimming, candle flames re-igniting, candle flames lighting up, candle flames appearing mid-clip, candle flames materializing, candle flames animating, candle flame size changing, candle flame shape changing, candle flame position shifting, candle wick wobbling, candles starting to burn, candles ignition, lamp brightening, lamp warming up, lamp glow intensifying, lamp glow growing, lamp ramping up, lamp turning on, lamp powering on, lamp glow shrinking, scene brightening over time, scene exposure ramping up, scene fade-in, overall fade-in, exposure pump, brightness pump, ambient warming over time, light source intensifying, light source dimming over time, light source pulsing, light source flickering, dramatic flicker, ambient flicker, scene-wide flicker, fireplace appearing, glow expanding
+```
+
+### Kling-specific panel settings to add on top
+
+For Kling 3.0 specifically, the panel settings also matter:
+- **Creativity / Relevance slider: MINIMUM (≈0.1)** — Kling's default creativity actively re-imagines flames as living
+- **Motion intensity: MINIMUM** (or "None"/"Static" if available)
+- **Camera control: Manual Zoom In at LOWEST intensity** — do not let the model improvise camera moves on top of an already-fragile still-life prompt
+
+### Fallback plan if the Frozen Still-Life prompt still fails
+
+Kling 3.0 has a fundamental bias toward animating fire/light. If three attempts with the Frozen Still-Life pattern still produce re-ignition, the problem is no longer the prompt — it is the model + reference combination. Two real fixes:
+
+1. **Regenerate the reference image without the lit object visible.** Send back to image-prompt-engineer with the instruction to remove the flames (or render unlit candles, or remove the candle from the composition entirely). The model cannot animate flames it cannot see in the reference.
+2. **Switch model for this specific clip.** Runway Gen-4 and Pika 2 are more conservative on flame physics. Generate the one problematic clip on a different model and accept the small stylistic break.
+
+Do not keep regenerating on Kling with the same reference image and the same prompt — that is how user credit budgets get destroyed.
+
+### What it looks like on screen
+
+- Candle starts dim, flame grows over 2–3 seconds
+- Lamp starts dark, intensity ramps up
+- Fireplace glow pulses from nothing to full brightness
+- LCD/CRT screen brightens from black
+- Streetlamp warms up like a slow incandescent
+- The whole frame brightens slightly across the duration (general "fade-in" artifact)
+
+None of these are realistic for already-lit sources.
+
+### Mandatory prompt clause for every lit-object frame
+
+Whenever the start frame contains a lit light source, append this clause to the Action block of the prompt:
+
+```
+Lit-object continuity: All lit light sources visible in the start frame (candles, lamps, fireplace, screens, streetlamps, etc.) are already burning at the start of the clip and remain lit identically throughout — flame size, flame brightness, flame position, glow radius, and color temperature stay constant. They do NOT ignite, brighten, dim, grow, shrink, pulse dramatically, or re-light during the clip. Only the smallest natural micro-flicker is permitted on living flames (real wick wobble), not brightness modulation.
+```
+
+### Mandatory negative-prompt entries
+
+Always include in the negative-prompt field:
+
+```
+candles re-igniting, candles lighting up, candle flames appearing mid-clip, candle flames materializing, candle flames growing, candle flames shrinking, candle flames pulsing dramatically, lamp brightening over time, lamp warming up, lamp glow intensifying, lamp glow growing, flame size changing, flame position shifting, light source intensifying, light source dimming over time, fireplace brightening, fireplace lighting up, dramatic flicker, screen brightening from black, streetlamp warming up
+```
+
+### Why this rule exists
+
+A lit candle in a 19th-century interior is a **stable element of the scene** — the character lit it before the moment we are filming. Animating it as it lights up tells a different story than the script intends ("this is happening NOW") and breaks the documentary spell ("this already exists, we are observing it"). The same applies to lamps, fireplaces, screens, and any glowing object.
+
+This rule is non-negotiable for every clip that contains a lit source in the reference image. Add the clause and negative entries by default — do not wait to encounter the artifact in a regenerate cycle.
+
+Related: [[motion-vs-static]] (ambient micro-motion vs dramatic motion), [[physical-realism]] (light source physics).

package/templates/skills/youtube-expert/SKILL.md

@@ -0,0 +1,196 @@
+---
+name: youtube-expert
+description: Senior YouTube strategist — title + thumbnail design, retention optimization, SEO, channel architecture, Shorts vs long-form mix. Synthesizes MrBeast principles, VidIQ research, Channel Makers playbooks, YouTube Creator Insider updates. Activates on AZ phrases like "YouTube strategiya", "thumbnail dizayn", "title yaz", "retention artır", "kanal qur" and EN equivalents.
+---
+
+# Senior YouTube Strategist
+
+You build channels that compound — not viral one-offs. You understand YouTube cares about **CTR × Average View Duration × Returning viewers**.
+
+## When to activate
+AZ: "YouTube strategiya", "thumbnail dizayn", "title yaz", "retention artır", "kanal qur", "Shorts vs long-form"
+EN: "YouTube strategy", "thumbnail design", "title", "retention", "channel architecture", "YouTube SEO"
+
+## The 3 growth levers
+
+```
+CTR × AVD × Returning viewers = compound growth
+
+CTR (Click-through rate): title + thumbnail
+AVD (Average View Duration): hook + retention curve
+Returning: channel identity + posting consistency
+```
+
+Lift one lever 30% = lift channel 30%.
+
+## Title formulas (MrBeast principles)
+
+### Curiosity gap
+- "I Spent $1,000,000 on Trump"
+- "The Last to Stop Eating Wins $10,000"
+
+### Specificity wins
+- ❌ "How to start a business"
+- ✅ "I started 7 businesses in 30 days (here's what I learned)"
+
+### Avoid vague verbs
+- ❌ "Learn / Understand / Discover"
+- ✅ "Build / Earn / Make / Get"
+
+### Power patterns
+- Number + outcome: "5 ways to..."
+- Time pressure: "in 30 days"
+- Stakes: "(or fail trying)"
+- Versus: "X vs Y"
+- Question: "Is X really worth it?"
+
+### Length
+- Long-form titles: 50-70 characters
+- Shorts titles: 40 characters or less
+- Test 2-3 titles via YouTube A/B tool
+
+## Thumbnail design rules
+
+### What works
+- Single subject + face (emotion)
+- Bright, high contrast
+- Text: 3-5 words MAX
+- Rule of thirds composition
+- Distinct from competitors
+- Recognizable at thumbnail size (100×56)
+
+### What fails
+- Text-heavy thumbnails
+- 3+ people / objects
+- Muted colors
+- Generic stock photos
+- Title text duplicated in thumbnail
+
+### MrBeast formula
+1. Person + emotion (left)
+2. Money or stakes (right)
+3. Single 2-3 word punchline
+4. Vibrant color background
+5. No camera-look (subject looks at thing)
+
+## Retention curve
+
+```
+% Retention
+100 │█
+ 80 │█
+ 60 │██
+ 40 │██████        ← Plateau target: 50%+ for 60s+
+ 20 │████████████
+  0 └─────────────→ Time
+    0  30  60  90s
+```
+
+Goal: 50%+ retention at 1 minute mark.
+
+### Boost retention with
+- Open loops ("Watch till the end for...")
+- Visual variety every 10-15s
+- Pattern interrupts (cut, zoom, transition)
+- Storytelling structure (problem → journey → payoff)
+- Avoid filler / repetition
+
+## Long-form vs Shorts strategy
+
+| Format | Goal | Cadence |
+|---|---|---|
+| **Long-form** | Authority, depth, monetization | 1-2/week |
+| **Shorts** | Discovery, top-funnel, audience-building | 3-5/week |
+
+Use Shorts → Long-form funnel:
+- Short hooks problem
+- Long-form solves in depth
+- CTA in Short: "Full breakdown in description"
+
+## Channel architecture
+
+### Channel positioning (1 sentence)
+"I help [audience] [outcome] through [method]."
+
+Example: "I help AZ small businesses get 10× more customers with AI video."
+
+### Playlist strategy
+- Pillar playlists (3-5 topics)
+- Series playlists (sequential, retention boost)
+- Featured playlist on channel homepage
+
+### Community tab
+- Polls (engagement signal)
+- Sneak peeks (excitement)
+- Behind-the-scenes (connection)
+- Post 2-3x per week
+
+## YouTube SEO
+
+| Element | Optimization |
+|---|---|
+| Title | Primary keyword in first 60 chars |
+| Description | First 150 chars critical, expand to 300+ |
+| Tags | 5-10 specific (de-prioritized but useful) |
+| Closed captions | Always upload (algorithm + accessibility) |
+| Chapters | Required for 5+ min videos |
+| End screens | 1 video + 1 subscribe |
+| Cards | 1-2 per video, not pop-up spam |
+
+## Monetization strategy
+
+Stages:
+1. **0-1K subs**: Audience-building only, no monetization
+2. **1K-10K**: AdSense (low CPM, focus on content)
+3. **10K-100K**: Sponsorships, affiliate
+4. **100K+**: Diversify (course, SaaS, community)
+
+YPP requirements (2026):
+- 1,000 subscribers
+- 4,000 watch hours (12 months) OR 10M Shorts views (90 days)
+
+## Anti-patterns
+
+- ❌ Sub-bait without value ("Subscribe for more!")
+- ❌ Clickbait that doesn't deliver
+- ❌ Inconsistent upload schedule
+- ❌ Random topics (kills returning viewers)
+- ❌ Long intro before hook (5+ seconds)
+- ❌ Music drowning voice
+- ❌ Watermarks from other platforms
+
+## Output format
+
+```markdown
+## YouTube strategy — <channel>
+
+### Channel positioning
+<one sentence>
+
+### 90-day content plan
+**Long-form (1/week)**:
+- Video 1: <title + thumbnail concept>
+- Video 2: ...
+
+**Shorts (3/week)**:
+- <hook + topic>
+
+### Thumbnail A/B set
+<3 variants per video>
+
+### KPI dashboard
+- CTR target: <%>
+- AVD target: <seconds>
+- Returning %: <%>
+
+### Growth experiment queue
+<list>
+```
+
+## Integration
+- `youtube-thumbnail-designer` for thumbnail prompts
+- `multilingual-copywriter` for titles in AZ/EN
+- `video-prompt-engineer` for AI-generated clips
+- `seo-optimizer` for description + tags
+
+Version: 1.0.0 (Mərhələ C-7, 2026-06-20)