agentleFS
BintzGavinBintzGavin/helios/skills/guided/brand-video/SKILL.md

guided-brand-video

Skill130 starsChanged 38 days agoApache-2.0View on GitHub

Guided, brand-aligned marketing videos rendered with Helios — promo or hype, explainer or walkthrough, product demo or feature showcase, testimonial or social proof, launch or release announcement, and vertical social clips (Reels, TikTok, Shorts). Extracts the brand from a repo or URL, locks the format, randomizes structure and staging, and designs an original look from a seed so videos don't look alike. Optional beat-synced music and Internet Archive B-roll.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Keep each video in its own folder, e.g.

    line 12
  2. Very specific:

    If they name or describe a style, a medium or a reference look, note it for step 5.

    line 22
  3. Distinctive:

    Write a brand profile: value proposition, audience, tone, palette, type, emotional position.

    line 28
  4. Very specific:

    Write one candidate from each of the three random lenses the roll drew, plus one that drops an assumption all the obvious answers share.

    line 58
  5. Distinctive:

    Use archive-broll. Grade every clip into the video's palette and materials so footage reads as part of the look.

    line 77
  6. Very specific:

    Check two or three early stills of the real page as style frames: do they look like the chosen candidate and like nothing on the obvious …

    line 96

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Brand video

One pipeline for six formats. The format profile fixes structure and pacing (references/formats.md). A random roll decides how the story is staged. The look is designed fresh for each video from the brand and a seed, the same way make-video does it. Known looks from motion-styles are opt-in: use one only when the user asks for it.

Build the page under the make-video contract (one HTML page, window.renderAt(t), every frame a pure function of t) and render, review and check as it describes. Read motion-design-rules before writing the spec.

Keep each video in its own folder, e.g. videos/<slug>/: seed.txt, direction.json, spec.md, video.html, audio/, broll/, video.mp4. Commit direction.json: the next video reads it to avoid repeating this one.

1. Gather context

⛔ Ask together, then wait:

  1. "Are we in your brand's repository, or should I research a URL (e.g. your marketing site)?"
  2. "Do you want generated background music (ElevenLabs; you'll need an API key set up), or do you have a track?"
  3. "Should I pull B-roll from the Internet Archive (public-domain and Creative Commons film)?"

If they name or describe a style, a medium or a reference look, note it for step 5. Don't read .env files or check keys; trust the answers.

2. Brand

From a repo: Tailwind config or CSS variables, global styles, design tokens, logo, favicon, hero sections, marketing copy, CTAs, UI components, existing motion. From a URL: the same visible signals. Add the format's extra signals from its profile.

Write a brand profile: value proposition, audience, tone, palette, type, emotional position. The brand fixes facts, logo, product UI, and the colours and fonts that must appear. It doesn't dictate a visual style.

3. Lock the format

Pick the profile in references/formats.md and apply its defaults: duration, aspect, arc, timing anchors, limits.

4. Roll the staging

Draw the seed with the apply-randomness skill if it's installed (save its output as seed.txt); otherwise:

LC_ALL=C tr -dc 'A-Za-z0-9' < /dev/urandom | head -c 1024 > videos/<slug>/seed.txt

Then roll:

node <this-skill>/scripts/roll-direction.mjs --format promo \
  --seed-file videos/<slug>/seed.txt --avoid . --out videos/<slug>/direction.json

It picks one option on every axis in references/direction-deck.json: structure, opening, ending, peak position, palette approach, ground, scale, camera path, layout, pacing, entrance emphasis, accent device, footage role, and one twist (a constraint or device; some twists break a motion-design rule on purpose). It also draws three idea lenses for step 5 and lists earlier videos' looks. It skips what the last two videos used, and never pairs picks that overlap. Add --no-broll if B-roll was declined; --lock axis=option for anything the user or brand requires; --style <slug> when step 5 uses a style card.

Honor the picks. Up to two vetoes are allowed for a concrete brand or format conflict, re-rolled with --lock and written in the spec. "Something else would look better" isn't one.

5. Design the look

Default: an original look, found with references/look-ideation.md:

  1. Write down the obvious answers: the pun on the product's name, the category's usual look, the stock looks (dark gradient with floating orbs or particles, glass cards, word-by-word fade-ups, neon glow on black), and the recent_looks in direction.json. They're off limits unless the brief asks for one.
  2. Write one candidate from each of the three random lenses the roll drew, plus one that drops an assumption all the obvious answers share.
  3. Cut candidates that collapse into an obvious answer or into each other.
  4. Let roll-direction.mjs --pick <n> choose among the survivors, and record the result in picks.look.label.

Only when the user asks for a known look, use the motion-styles library its own way:

  • They name or describe a style or medium ("make it risograph", "like an old blueprint"): open that one card, and re-run the roll with --style <slug> so the history records it.
  • They ask for options or "a known look": follow motion-styles to offer at most three styles plus one original direction, and build what they pick.

With a card, its invariants win: if it bans a rolled pick (Swiss bans bounce, risograph bans dissolves, a locked-camera style can't track), re-roll that axis with --lock to an option the card allows. Map brand colours into the card's colour rules. Never copy the card's demo composition.

Without being asked, you may offer once, in one line, that known styles exist.

6. Music (if wanted)

Generate an instrumental with ElevenLabs matching the brand tone, the format's music direction, the rolled pacing and the runtime, or use the user's track. Analyse it with helios analyze (or make-video's fallback) and key timing to it as make-video's music section and motion-design-rules Rule 7 describe. Without music, build the timing grid from the rolled pacing.

7. B-roll (if wanted and broll isn't none)

Use archive-broll. Grade every clip into the video's palette and materials so footage reads as part of the look.

8. Creative specification

One spec in a single code block, also saved as spec.md. Don't restate the prompt. Include:

  • Duration, aspect, resolution, FPS, render mode
  • Brand summary
  • Format, every rolled pick, locks, vetoes and why
  • The obvious answers, the four look candidates with their lenses, what was cut and why, the pick, and the seed pattern → choice map (or the style card the user asked for)
  • How the twist shows up (and which motion-design rule it breaks, if any), and what happens at the peak
  • Scenes with exact timestamps, each keyed to a sound or time (W/B/H/F/T, Rule 7)
  • Voiceover and on-screen text
  • Per scene: layout, hero, layers, entrance, transition out
  • Elements that persist across scenes
  • Audio direction and assets (logo, UI, B-roll clips with ranges, fonts)

9. Build, review, render

Check two or three early stills of the real page as style frames: do they look like the chosen candidate and like nothing on the obvious list? Then build, render, sheet, verify and check as make-video describes.

CLI notes: still and sheet capture at 1920×1080 unless you pass --width and --height. still, sheet and verify take no --mode (only render does), and --fps only applies to sheet --strip. check, analyze and verify --cues need CLI 0.47.0 or later; on an older CLI, use the fallbacks in make-video's troubleshooting table. Say in the report which checks didn't run.

Report the output path, format, seed, picks, the chosen look, locks and vetoes, B-roll credits, and anything unverified.

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.