Use this skill when writing scripts for programmatic videos, planning scene structure and timing, creating storyboards in YAML format, calculating frame counts from duration, or interviewing users about video requirements. Triggers on video script, storyboard, scene planning, narration writing, video pacing, and structured video content planning.
Standardmäßig ist der Prompt ausgewählt, der zuerst die Quelle prüft. Sie können zu einem direkten Befehl wechseln oder eine lokale Kopie herunterladen.
Quelldateien prüfen
Lesen Sie SKILL.md und alle von SkillsMP angezeigten Begleitdateien, bevor Sie sich für eine Installation entscheiden.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Ein direkter Befehl überspringt den Prüf-Prompt. Prüfen Sie die Quelle, bevor Sie ihn ausführen.
Use this skill when writing scripts for programmatic videos, planning scene structure and timing, creating storyboards in YAML format, calculating frame counts from duration, or interviewing users about video requirements. Triggers on video script, storyboard, scene planning, narration writing, video pacing, and structured video content planning.
Interview-driven - Run a structured interview (up to 30 questions across
7 categories) before generating a single scene. Never script from assumptions.
Structured output - Every script is valid YAML with a meta block and a
scenes array. Each scene has id, duration, frames, narration,
visual, animation, music, sfx, and transition_to_next fields.
Visual-first - Write the visual field before narration. Narration
should complement visuals, never redundantly describe what is on screen.
Pacing awareness - Match scene count and per-scene duration to the video
type: social clips need 3-8s scenes; explainers can breathe with 5-12s scenes.
Narration-visual sync - Every narration line must match what is visible at
that moment. One idea per scene. Narration timing must fit the scene duration.
Core concepts
Interview framework
Gather requirements through 7 categories before writing:
Category
Questions
Examples
Product/Subject
5-7
What is the product? Key features? Problem solved? Differentiator?
Audience
3-5
Target viewer? Technical level? Pain points?
Video Goals
3-4
Type (demo/explainer/social/announcement)? Duration? Channel?
Tone & Style
3-5
Formal or casual? Energetic or calm? Reference videos?
total_frames = duration_seconds * fps
scene_frames = scene_duration_seconds * fps
Example: 60s video at 30fps = 1800 total frames
Duration
24 fps
30 fps
60 fps
4s
96
120
240
8s
192
240
480
30s
720
900
1800
60s
1440
1800
3600
90s
2160
2700
5400
Always validate that scene frame counts sum to total_frames in meta.
Common tasks
1. Run the interview workflow
Walk through all 7 categories in sequence. Ask one at a time, summarize answers,
then proceed. See references/interview-questions.md for the full 30-question
framework. Compressed version for tight timelines uses 10 essential questions:
Product name, key features, problem solved, target viewer, technical level,
video type, duration, channel, key message, and CTA.
2. Generate a product demo script
meta:title:"Product Demo - Acme Dashboard"duration:"60s"type:demoresolution:3840x2160fps:30audience:"SaaS founders and product managers"tone:"professional, confident, minimal"total_frames:1800scenes:-id:introduration:"4s"frames:120narration:"Meet Acme - the dashboard that builds itself."visual:"Logo centered on warm off-white background, fades in from transparent"animation:"fade-in over 30 frames, hold for 90 frames"music:track:"upbeat-corporate"volume:0.4duck:falsesfx: []
transition_to_next:"hard-cut"-id:feature-1duration:"8s"frames:240narration:"Just describe what you need. Acme handles the rest."visual:"Browser mockup showing prompt input, text being typed"animation:"browser slides up from bottom over 20 frames, typing starts at frame 40"music:track:"upbeat-corporate"volume:0.2duck:truesfx:-type:"keyboard-typing"at:"1.3s"duration:"3s"transition_to_next:"cross-dissolve"-id:feature-2duration:"8s"frames:240narration:"Drag, drop, resize. Your layout, your rules."visual:"Dashboard editor with widgets being rearranged by cursor"animation:"cursor moves to widget at frame 30, drags to new position over 60 frames"music:track:"upbeat-corporate"volume:0.2duck:truesfx:-type:"soft-click"at:"1.0s"duration:"0.2s"transition_to_next:"cross-dissolve"-id:ctaduration:"6s"frames:180narration:"Try Acme free at acme.dev. Build your first dashboard in minutes."visual:"CTA text centered with URL, subtle animated background gradient"animation:"text fades in over 20 frames, background gradient shifts slowly"music:track:"upbeat-corporate"volume:0.4duck:falsesfx: []
transition_to_next:"fade-to-black"
Full demo scripts typically have 6-15 scenes. Expand by adding problem, social
proof, pricing, and outro scenes following the same YAML structure.
3. Generate a social clip script
meta:title:"Acme in 30 Seconds"duration:"30s"type:socialresolution:1080x1920fps:30audience:"Developers scrolling social feeds"tone:"energetic, punchy, modern"total_frames:900scenes:-id:hookduration:"3s"frames:90narration:"Stop building dashboards from scratch."visual:"Bold text on vibrant gradient background"animation:"text slams in from top over 8 frames, screen shakes for 4 frames"music:track:"electronic-pulse"volume:0.5duck:truesfx:-type:"impact-hit"at:"0.2s"duration:"0.3s"transition_to_next:"hard-cut"-id:demoduration:"12s"frames:360narration:"Type what you need. Acme builds it live."visual:"Screen recording: typing a prompt, dashboard generating in real time"animation:"typing for 120 frames, dashboard builds over remaining 240 frames"music:track:"electronic-pulse"volume:0.3duck:truesfx:-type:"keyboard-typing"at:"0.5s"duration:"4s"transition_to_next:"hard-cut"-id:resultduration:"8s"frames:240narration:"Fully interactive. Real-time data. Ready to share."visual:"Finished dashboard with hover interactions, data updating live"animation:"cursor hovers triggering tooltips, data refreshes at frame 120"music:track:"electronic-pulse"volume:0.3duck:truesfx: []
transition_to_next:"hard-cut"-id:ctaduration:"7s"frames:210narration:"Try free at acme.dev."visual:"CTA text large and centered, URL below, brand gradient background"animation:"text scales up from 0 to full size over 15 frames, holds"music:track:"electronic-pulse"volume:0.5duck:falsesfx: []
transition_to_next:"none"
Always verify the sum. If scenes do not add up, adjust or add scenes.
5. Write effective narration text
Rules:
Max 15 words per sentence - longer cannot be read in time
Active voice, present tense - "Acme builds your dashboard" not "will be built"
Match narration to visuals - talk about what is on screen
One idea per scene - do not cram two concepts into one line
Lead with benefit - "Save 10 hours a week" not "Our time-tracking feature"
Include pauses - empty narration ("") for breathing room
Reading speed - roughly 2.5 words per second
Bad
Good
Why
"Our product has been designed to help teams build dashboards faster"
"Build dashboards in minutes, not weeks."
Too long, passive
"Click on the plus button in the top right corner"
"Add a widget with one click."
Let the visual show location
"As you can see, the data updates in real time"
"Real-time data. Always current."
"As you can see" is filler
6. Plan scene transitions
Transition
When to use
hard-cut
Same topic, fast pacing, or jarring contrast
cross-dissolve
Smooth topic change, related content flowing
fade-to-black
End of section, dramatic pause, final scene
wipe-left / wipe-right
Before/after comparisons, timeline progression
none
Final scene of the video
Rule: use at most 2 different transition types per video for consistency.
7. Revise a script based on feedback
Identify the feedback type: pacing, narration, visuals, structure, or tone
Locate affected scenes by id in the YAML
Apply changes to only the affected fields, preserve frame math
Revalidate: scene durations must still sum to total duration
Recalculate frames if any duration changed
Document what changed:
# Revision: shortened intro from 6s to 4s per feedback (v2)-id:introduration:"4s"# was 6sframes:120# was 180
Anti-patterns / common mistakes
Mistake
Why it is wrong
What to do instead
Writing narration before visuals
Drives video into talking-head territory
Write visual first, then narration to complement
Scenes longer than 12 seconds
Viewers lose attention, pacing feels sluggish
Break into two shorter scenes
Mismatched frame counts
Rendering engine produces wrong timing or crashes
Always compute frames = duration * fps and verify sums
Narration over 15 words/sentence
Cannot be read within scene duration
Split into shorter sentences
No hook in first 3 seconds
Social viewers scroll past, embedded viewers disengage
Open with bold statement, question, or visual surprise
Inconsistent transitions
Video feels choppy and amateurish
Use at most 2 transition types per video
Skipping the interview
Produces generic scripts that miss the mark
Always gather requirements first
Empty visual descriptions
Rendering engineer cannot build the scene
Be specific about layout, colors, motion, elements
Gotchas
Frame count rounding - When duration does not divide evenly (e.g., 3.5s
at 24fps = 84 frames), round to nearest integer and adjust the last scene.
Never leave fractional frames - rendering engines truncate or error.
Narration timing overflow - At 2.5 words/second, a 4s scene holds about
10 words. Writing 20 words for a 4s scene means rushed narration or clipping.
Always check word count against scene duration.
Vertical vs horizontal resolution - Social clips (TikTok, Reels, Shorts)
use 1080x1920 (vertical). YouTube/website embeds use 1920x1080 or 3840x2160.
Confirm distribution channel before setting resolution.
Music ducking conflicts - If every scene has duck: true, music volume
constantly yo-yos. Use ducking only when narration is present. Scenes with
empty narration should have duck: false.
Total duration drift - After revisions, scene durations often drift from
target. Always re-sum all durations after any edit and compare against
meta.duration.
References
For detailed guidance on specific sub-topics, read the relevant file from the
references/ folder:
references/interview-questions.md - complete 30-question interview framework
with follow-up trees, answer interpretation, and example briefs
references/pacing-guide.md - deep-dive into pacing strategies per video type,
hook techniques, CTA placement, and rhythm patterns
Only load a references file if the current task requires it - they are long and
will consume context.
Companion check
On first activation of this skill in a conversation: check which companion skills are installed by running ls ~/.claude/skills/ ~/.agent/skills/ ~/.agents/skills/ .claude/skills/ .agent/skills/ .agents/skills/ 2>/dev/null. Compare the results against the recommended_skills field in this file's frontmatter. For any that are missing, mention them once and offer to install:
