taste-banners
Skill4 starsChanged 16 days agoMITView on GitHub
Designs social covers, ad creatives, website heroes, email headers, Open Graph cards and print banners as export-ready HTML/SVG, always in three genuinely different art directions from one token set, with platform safe zones, crop tests and Playwright PNG export. Use for "banner", "OG image", "social cover", "LinkedIn/X/YouTube header", "Instagram post/story", "display ad", "hero image", "email header", "poster".
In this repository12 other files it gives its agents.
- dist/AGENTS.mdAGENTS.md
- dist/copilot-instructions.mdCopilot instructions
- taste-auditSkill
- taste-brandSkill
- taste-briefSkill
- taste-componentsSkill
- taste-directionSkill
- taste-guardSkill
- taste-identitySkill
- tastemakerSkill
- taste-slidesSkill
- taste-tokensSkill
The file
Files
Taste Banners: one message, three real directions
Reads brand-guidelines.md and design-tokens.json (run brand Light and tokens Minimal first if missing).
Platform sizes change. Use references/sizes.md, verify against the platform's current spec when web access exists, and say "unverified" otherwise.
Procedure
- Brief (6 lines, infer; ask max 2 questions): goal, platform(s), audience, the single message, one CTA, mandatory elements (logo, legal line, date, price).
- Three directions from
references/directions.md, each a different idea (not a recolor), with one line each for concept, hierarchy, color use and eye path. - Compose the largest size first using the rules below, then reflow (never scale) to other sizes.
- Build one HTML file per direction x size:
banners/<campaign>-<direction>-<w>x<h>.htmlplusbanners/index.htmlpreview grid. - Check safe zones, crop behaviour, contrast, file-size budget. Export PNG at 2x with
scripts/export-banners.mjswhere Node + Playwright exist.
Composition rules
- Hierarchy: headline, then proof line or visual, then CTA, then logo. Max 3 text levels.
- Copy: headline at most 7 words (tiny formats 4); verb-led CTA of 2-3 words; formulas from taste-brand (PAS, BAB, 4U). Never invent statistics, testimonials or discounts (C03).
- Grid: margins at least 5% of the shorter side; align to a baseline; token spacing.
- Contrast: text 4.5:1 (3:1 for 24px+ bold); photos get a scrim (surface color at 60-80%) or a text panel.
- Type: brand fonts, 2 weights max, tighter tracking at display sizes, no text over busy image areas.
- Color: 60/30/10; the accent is for the CTA only. No default AI gradient (V01), blobs (V04) or glow (V09).
- Imagery: supplied, original or generated abstract/illustrative only; no identifiable people without rights; no third-party logos.
- Motion (HTML ads): max 15s or 3 loops,
prefers-reduced-motionrespected, final frame carries the whole message. - Alt text for every exported image in the notes.
Artboard pattern
<section class="banner" style="width:1200px;height:630px" role="img" aria-label="<alt text>">
<div class="bg"></div>
<div class="content"><h1>...</h1><p>...</p><span class="cta">...</span></div>
<svg class="logo" aria-hidden="true">...</svg>
</section>Fixed pixel artboard for exact export; preview pages scale it with transform: scale() only.
Quality gate
- One message, one CTA, headline within limit.
- Three distinct directions with rationale.
- Text inside safe zones; crop-tested per platform.
- Contrast verified; scrim on photos.
- Tokens only; no invented claims or third-party marks.
- Exact dimensions; within file-size budget; alt text written.
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
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.

