- Home
- Skills
- Frontend & Web Apps
- Musepool: Web Design Variety System
More screenshots
Works with the AI Agents you already use
Musepool: Web Design Variety System
Design pages from harvested references instead of AI defaults: pick the deciding dimensions, extract palette ladders, type systems, and spacing rules, synthesize without copying, trap-check.
$5
Musepool: Web Design Variety System
Example session with this skill installed
Design a pricing page for my SaaS product. I don't want it to look like every other AI-generated pricing page. Brand color is a deep teal; otherwise I'm open.
- Read your context and instructions
- Compiled the musepool
REFERENCE-FIRST RUN — produced by the skill's protocol.
Dimensions chosen: layout, typography, density. Color closed by mandate (deep teal brand) and recorded on the constraint sheet; motion minimal for this page type, not harvested.
Harvest: same-domain SaaS pricing teardowns for proven comparison patterns, plus cross-domain editorial rate cards and restaurant menus — both solve "compare tiered options legibly," which is the actual problem. Style-word queries ("modern SaaS pricing") were discarded; problem queries surfaced the keepers. Kept only references making a nameable decision we wouldn't have made by default.
Extraction highlights: from an editorial rate card — a two-option comparison built on typographic scale instead of cards, price as the largest element on the page; from a SaaS reference — a feature matrix whose density is tuned for scanning, not completeness; from a menu — prices set large and unapologetic. Palette built by extending the mandated teal down its own ladder (deep teal accent, desaturated teal-grey surfaces, near-black text), never invented fresh.
Synthesis: editorial type hierarchy + SaaS matrix discipline. Signature move: prices huge, everything else quiet. Rejected, and recorded: the highlighted middle tier and three-equal-cards layouts — both template defaults.
Trap check on the draft: no emoji iconography, no cards-in-cards, no template copy ("clean and intuitive" struck), fonts verified for the page's languages, every image earning its place. Harvest scored 9/10 with full traceability — every major decision on the page points at an extraction card.
Connects securely to your tools. The creator never sees your data.
About this skill
Left to its own devices, a model designing a page regresses to the mean of everything it has seen: the same hero, the same three cards, the same gradient. Trying harder doesn't fix it — refusing to decide from memory does. Musepool is the reference-first method: gather real precedents first, extract constraints from them, and let the constraints do the designing. You start by decomposing the page into the 3–5 design dimensions that actually decide it (layout, color, typography, motion, density, imagery, craft details) — with a selection table by page type, so a pricing page harvests layout/type/density while a portfolio harvests imagery/motion/color. Dimensions closed by a brand mandate are recorded as constraints, not re-harvested. Harvesting follows strict query discipline: search by problem, not style ("dense data table readability," not "modern dashboard"); add the task-specific words that surface interesting results; mix same-domain proof with cross-domain surprise. A candidate is kept only if it makes a specific, nameable decision you wouldn't have made — "nice" is not a criterion. Each kept reference gets an extraction card: the palette as an actual ladder (background → surface → text → accent → muted, with temperature and contrast relationships), the type system (roles, weight contrast, scale ratio — the system, not the font name), the spacing philosophy, the motion curves and what deliberately stays still, and exactly one signature move. Cards converge into a constraint sheet the build must consult, including a Rejected line that stops dropped defaults from sneaking back. Synthesis rules are strict: every major color traces to a harvested ladder, techniques are never downgraded "to keep it simple," and one signature move leads. The draft ships only after the default-trap checklist (no emoji icons, no cards-in-cards, no highlighted middle pricing tier) and a harvest-quality rubric (8+/10, no zero on traceability).
- The method needs reference access to work as designed; without it, the skill degrades to mandate-and-convention design and requires that result to be flagged as unharvested.
- It transfers systems and relationships from references, never content: a reference's photography, branding, and copy are explicitly out of bounds, and the extraction card has a do-not-take line to enforce it.
- It governs design decisions, not implementation quality — a well-constrained design still needs competent building and its own visual QA.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
2 uses
Times used to date
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago
Permissions
No special permissions declared or detected