agentleFS
Razee4315Razee4315/coldframe/.claude/skills/video-rules/SKILL.md

video-rules

Skill4 starsChanged 38 days agoMITView on GitHub

Rules for product, launch, explainer and social videos (Remotion or otherwise). Use before planning, scripting, designing or animating any video, and run the checklist before rendering. Covers story structure, what to avoid so the video doesn't look generically AI-made, text, UI, data, audio, captions, CTA, length and export formats.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Write these four lines at the top of the plan.

    line 14
  2. Distinctive:

    Every number must be real and sourced.

    line 81
  3. Distinctive:

    Never imply a comparison you haven't measured.

    line 82

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Video rules

The goal is a video that looks designed by a person with taste, not assembled from the default AI toolkit. Restraint beats motion: if everything moves, nothing stands out.

This skill says what to show and what to avoid. motion-direction says how to make it move like a real motion-design film (camera, depth, transitions, two-pass workflow with styleframes), and sound-design covers audio. Use all three.

1. Decide before designing

Write these four lines at the top of the plan. If one is missing, ask.

  • Goal: one action (sign up, book a demo, install, understand).
  • Audience: one group (developers and CEOs need different videos).
  • Message: one sentence. If it needs "and", cut something.
  • Placement: where it plays. This sets length, aspect ratio and pacing (see section 9).

2. Story

Lead with the outcome or the problem, never the logo.

BeatShare of a 60 s videoWhat happens
Hookfirst 3 sShow the problem, a bold true claim, or the end result
Problem~15%The pain, shown, not described
Solution~10%Product name + one-line promise
Demo~45%2–3 moments, each showing a result, not a feature
Proof~15%A real number, quote or customer
CTAlast 5–7 sOne action + URL
  • Viewers decide in about 3 seconds. Spend them on the problem, not the brand.
  • One clear workflow beats a feature list. Feature overload is the main failure, more than low polish.
  • Hooks: one promise, present tense, simple words, nothing you can't back up.

3. Don't: the generic AI look

Each of these is fine once, on purpose. Together they read as "made by a template".

  • Opening with a logo animation or "Hi, welcome to…"
  • Near-black background + white text + purple/indigo accent + Inter. Use the real brand palette and fonts.
  • Glow, bloom, lens flares, floating particles or bokeh as decoration. (One motivated light sweep or a brief burst behind a result is fine; see motion-direction.)
  • Spring overshoot "pop-in" on every element, every element staggered. (Overshoot is for confirmations only.)
  • The same soft fade between every scene.
  • A self-drawing underline or highlighter swipe in every scene. Max one per video, two if long.
  • Word-by-word yellow-keyword captions bouncing in (the CapCut/Hormozi preset). If captions are needed, use calm 2–4 word chunks held 600–900 ms.
  • Rotating 3D laptop/phone mockups as filler.
  • Fake cursors wandering around, arrows and boxes on everything.
  • Count-up numbers for things that aren't the key stat.
  • Generic upbeat stock music with no sound effects, or sound generated from code (sine beeps, noise "whooshes").
  • Flat slides: text and boxes fading in on a still background, no camera, no depth. That reads as a slideshow, not a film.
  • AI-generated footage of the product itself, or anything showing features the product doesn't have.
  • Raw, full-screen dashboards that are unreadable on a phone.

4. Do: text and motion

  • Fonts: one or two. Big enough for a phone. Strong contrast.
  • Words on screen: 7 or fewer at a time. One idea per screen.
  • Hold time: text stays readable at least 1.5 s after it lands.
  • Emphasis: one key word per line, by colour or weight. Pick one emphasis technique per video (underline, highlighter, strike-and-replace, or colour pop) and use it only where it carries meaning.
  • Motion language: one easing family for the whole video (e.g. expo-out). Calm entrances: mask reveal, blur-to-sharp, short slide. Motion only where it explains something.
  • Camera and depth: give the film a camera (push-ins, arcs, parallax across 3+ depth planes). Build it as described in motion-direction.
  • Transitions: mix them and give each a reason. Hard cut on a beat, match cut (a shape becomes the next scene), zoom into the element that matters, wipe with a brand shape. Never the same transition every time.
  • Pacing: something changes every 2–4 s (a cut, a zoom, a new element), but no scene is busy.

5. Product UI

  • Show the real product. Rebuilt/stylised UI is fine if it matches what exists.
  • Crop and zoom into the part that matters. Spotlight it, dim the rest.
  • Make UI responses snappier than real life (about 3×), not slower.
  • A cursor only when it teaches something: large, with a click ripple and a soft click sound.
  • One success moment (a check, a finished state) after the key action. End on calm and relief, not a feature pile.

6. Data

  • One number per screen, held 1.5 s or more. Round it (2.4×, not 2.387×).
  • Count-ups use tabular (fixed-width) digits so they don't jitter.
  • Bars grow from zero and are labelled directly (no legend). Never cut the axis to exaggerate.
  • Every number must be real and sourced. Put a small source line on screen ("Measured on …, 2026").
  • Never imply a comparison you haven't measured.

7. Audio

Follow the sound-design skill. In short: a soft music bed the user picked, a real recorded sound for every visible click, keystroke, toggle, landing, success and camera move, one impact for the big moment, about -14 LUFS with true peak ≤ -1 dBTP, and an audio sample approved by the user before the full render. Never synthesise music or effects in code. The video must still work muted; README and website-hero versions ship silent. Burn in captions if there's a voiceover (calm 2–4 word chunks) and ship an SRT.

8. Brand and CTA

  • Brand colours and fonts everywhere. Same motion style from first scene to last.
  • Full logo only on the end card (optional small corner mark elsewhere).
  • CTA is specific ("Install in one command", "Start free, no card") and the URL is large and readable.
  • End card: logo, one line, CTA, URL. Nothing else.

9. Length and formats

PlacementLength
TikTok / Reels6–15 s
YouTube Shortsup to 60 s
LinkedInunder 45 s
Homepage / README demo20–90 s
Feature page2–3 min
  • Formats: 16:9 (YouTube, web), 1:1 (feeds), 9:16 (Reels, TikTok, Shorts). Keep text out of the platform's button zones (right edge and bottom on vertical).
  • 1080p minimum. 30 fps for most, 60 fps for smooth UI and cursor motion.
  • Custom thumbnail: the outcome or a face, 3–5 words.
  • Make cutdowns (15 s, 6 s) from the main video instead of one video everywhere.

10. Checklist before rendering

Answer each with yes. Fix anything that isn't.

  • The first 3 seconds show a problem or result, not a logo.
  • One goal, one audience, one message, written down.
  • Every claim and number on screen is true and measured.
  • No glow, particles, or default dark-purple look. Real brand palette and fonts.
  • At most one underline/highlight technique, used once or twice.
  • Transitions are mixed and each has a reason.
  • No screen has more than 7 words. Every text holds 1.5 s or more after landing.
  • Readable on a phone (check a frame at 390 px wide).
  • sound-design checklist passed: music bed, a sound for every action, measured loudness, sample approved.
  • motion-direction checklist passed: directions → beat sheet → styleframes approved; camera and depth; one signature transition.
  • Captions burned in if there's speech.
  • Ends on one specific CTA with a readable URL.
  • Rendered in every format the placement needs.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.