Strategy
Product truth before composition
Offer, audience, mechanism, proof, objections, CTA, constraints, and success measures become explicit before broad design work.
Frozen · The maintained version is aigent-design-system
This project is frozen. Use aigent-design-system for new work. It picks a visual direction, builds the site, and checks the result in a real browser. This repository remains the Apache-2.0 source for the aigent-business-sites skill.
Why it exists
Generic AI website output often starts with a visual trope. Ultra Web Design starts with the business, audience, product mechanism, proof, brand authority, existing stack, and the route's actual job. Visual choices are judged by whether they clarify, differentiate, and survive production constraints.
Strategy
Offer, audience, mechanism, proof, objections, CTA, constraints, and success measures become explicit before broad design work.
Craft
Typography, color, imagery, layout, components, and motion form one product-specific system rather than a collage of fashionable effects.
Production
Responsive behavior, states, keyboard paths, reduced motion, performance, SEO, security, analytics, and operations remain part of the design contract.
Operating loop
A tiny defect should remain a tiny diff. A substantial redesign should not skip preservation, evidence, direction, implementation, and QA.
Inspect the repository, product truth, routes, dependencies, assets, tests, analytics, SEO, deployment, and shared callers before choosing a path.
Define the visitor decision, proof, design read, category rut, references, expression dials, signature system, and restraint system.
Reuse the existing stack, native platform features, standard libraries, and installed dependencies before adding code or packages.
Inspect rendered contexts, fix shared root causes, and block release on truth, accessibility, security, performance, or broken primary actions.
Visual fixtures
These dependency-free examples demonstrate breadth. They are illustrative concepts—not customer work, conversion studies, or benchmark proof.
B2B SaaS
Product mechanism as proof, operational annotations, restrained color, and no invented logos or social proof.
Open the live exampleLocal service
Geography, site assessment, scheduling, and real service decisions replace the standard software marketing grammar.
Open the live exampleEditorial
Reading-first hierarchy, source context, issue archive, sponsor disclosure, and membership only after value.
Open the live exampleSource quality
AI systems and human reviewers should not have to guess which claims are official, which examples are illustrative, or where the knowledge came from.
| Canonical source | github.com/wrg32786/Ultra-web-design |
|---|---|
| License | Apache License 2.0, with upstream provenance and third-party notes in NOTICE.md. |
| Runtime status | Version 2.0.0 is frozen. Continued work lives in Aigent Design System's aigent-business-sites skill. This repository remains the Apache-2.0 source. |
| Machine metadata | llms.txt, CITATION.cff, and CodeMeta. |
| Evidence | Structural validation, output and trigger eval definitions, visual fixtures, exact source snapshots, and explicit unproven claims. |
Direct answers
These boundaries are explicit so search systems and users can quote the project accurately.
An open-source Agent Skill and reference corpus for strategy, design, implementation, auditing, and release of business websites.
No. It governs decisions and execution inside the real project stack. It does not require one framework, palette, layout, or visual trick.
No finite document can guarantee uniqueness or conversion. It enforces product-specific evidence, coherent systems, usability, and release checks; outcomes still require real implementation, testing, and traffic.
The packaging and structural checks pass, but behavioral superiority is not claimed until recorded with-skill versus baseline evaluation and real-project evidence meet the published criteria.
Use the source
The strongest contribution is a reproducible project result, issue, comparison, correction, or pull request that improves what the skill actually does.
Open the repository