agentleFS
aiappsgbbaiappsgbb/awesome-gbb/skills/web-experience-design/SKILL.md

web-experience-design

Skill8 starsChanged 12 days agoMITView on GitHub

Design and improve task-centered web experiences: executive HTML reports, analytical comparisons, catalogs, dashboards, interactive tools, documentation, informational sites and navigable HTML presentations. Turn complex content into clear views, useful visual explanations and working navigation before styling. USE FOR: web experience design, information architecture, content hierarchy, too much text, long scrolling pages, findability, list-detail flows, responsive navigation, meaningful icons and motion, SITE.md, DESIGN.md, browser UX review, printable or offline HTML. Use independently or with an approved visual tool; no mandatory theme, framework, model or external service. DO NOT USE FOR: backend-only work, cloud deployment, native mobile apps, PowerPoint authoring, prose-only editing or formal accessibility certification.

We split this skill into the rules it gives an agent

14 rules. About 0 are common, 13 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈13

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask only material unanswered questions, one at a time through the available question tool.

    line 32
  2. Distinctive:

    Read the real material before choosing a layout.

    line 48
  3. Distinctive:

    Do not default to a long all-open page, tiny text, twenty tabs, nested scrolling, scroll hijacking or a carousel of essential information.

    line 78
  4. Distinctive:

    Use contracts to record the smallest durable experience contract.

    line 93
  5. Distinctive:

    Read craft before visual implementation.

    line 99
  6. Distinctive:

    No universal palette, serif pairing, card grid, sidebar or mandatory animation.

    line 102
  7. Very specific:

    Document approved visual rules in the project's existing system or a focused DESIGN.md.

    line 110
  8. Distinctive:

    Prefer the existing stack and native semantics.

    line 122
  9. Distinctive:

    Keep ordinary page scrolling usable.

    line 131
  10. Distinctive:

    Follow verification. Execute the agreed tasks against the rendered result with real content and record expected versus observed outcomes.

    line 138
  11. Distinctive:

    Never call a missing browser check a pass, an automated walkthrough a user study, a linter a usability certificate, or a self-assigned …

    line 146
  12. Distinctive:

    Stop for a scope/budget decision if another pass makes no material progress.

    line 150
  13. Distinctive:

    Stop temporary servers you started unless the user requested retention.

    line 159
  14. Distinctive:

    Before promoting this method to a global default, run the acceptance scenarios: report, comparison/tool and a different informational …

    line 161

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 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

Web Experience Design

Make the next useful action obvious, the important content understandable, and the detail easy to retrieve. A complete document is not automatically a usable interface. Attractive screenshots are not evidence of a working experience.

Merged source; release pending: output acceptance and cross-runtime discovery remain pending; see validation boundaries.

Choose the scope

Identify the target, task and write boundary before changing anything. Reuse the brief, existing design system, approved decisions and repository conventions. Ask only material unanswered questions, one at a time through the available question tool. Do not restart discovery or ask for arbitrary taste preferences.

RequestPath
Audit or plan onlyInspect and propose; do not fix, install or publish.
Small refinementPreserve structure, behavior and brand; state the local intent, make the bounded change, verify affected paths. No full redesign workshop.
New experience or structural redesignComplete the workflow below; obtain direction approval before substantial implementation.

Choose the primary visitor task: understand, decide, operate, find, act, or present. Mixed experiences may use different modes in different views. These are routing hints, not templates. Load only the relevant part of architecture.

1. Establish content and a baseline

Read the real material before choosing a layout. Record the reader's questions, required content, evidence, consequential caveats, actions and delivery constraints. Separate facts, interpretation and proposed copy. Never invent claims, sources, testimonials, prices, user research or customer results to fill a composition.

Draft interface copy from those facts, not from generic product language. Before direction approval, run the interface copy pass in craft: headlines, labels, cards, empty states, errors and alt text in the user's language, applying gbb-humanizer short-copy guidance where installed. Keep operational UI terse; preserve facts, sources, technical names and copy the user already approved.

For a redesign, inspect the current rendered experience where possible. Define 3-5 representative tasks, or fewer for a genuinely smaller change, with explicit starting state and observable outcomes before editing. Include returning to information, not only a first visit. If a user wants a before/after but supplied no disappointing artifact, request one; do not choose an easy artificial baseline.

Preserve an original or use version control. Record actual viewport, URL/state, paths, search effort, hidden essentials and context loss. Do not infer human task times from automation. If the app cannot run, label source-only findings and unverified visual/interaction behavior rather than claiming a browser audit.

2. Decide the structure

Map each required content unit to an initial view, detail, source or complete reading/export view. Remove redundancy only with a reason; moving content must not make it disappear from search, links or export.

Explicitly choose among addressable dynamic views, multiple pages, a hybrid, or linear reading. Explain the choice in terms of the tasks. Do not default to a long all-open page, tiny text, twenty tabs, nested scrolling, scroll hijacking or a carousel of essential information. Long-form reading is valid when intentional, not the only way to retrieve a specific answer.

Specify navigation labels, current location, direct links, Back/Forward, filter and selection state, focus/return behavior, narrow-screen changes, and applicable search, empty and error states. A header link and a footer CTA performing the same navigation must preserve the same relevant context.

Show a monochrome wireframe with enough real content to expose density and wrapping. Offer two structurally different alternatives only when a real tradeoff exists, not two palettes of the same page. Obtain explicit direction approval; if already approved, reuse it. With no answer mechanism, stop at a labeled proposal unless the user has already authorized those decisions.

Use contracts to record the smallest durable experience contract. Reuse an existing file; do not create a bureaucracy of duplicate briefs, manifests and acceptance receipts.

3. Give the experience a visual language

Read craft before visual implementation. Preserve an existing brand; otherwise derive type, spacing, composition, imagery and component character from the subject and use context. Distinctive does not mean unfamiliar. No universal palette, serif pairing, card grid, sidebar or mandatory animation.

Choose how to explain each idea: concise text, comparison table, diagram, chart, image or interaction. Use semantic icons with visible labels where meaning is ambiguous. Define purposeful motion for state, continuity and feedback, with a reduced-motion alternative. Do not satisfy these requirements with decorative arrows and a page-load fade.

Document approved visual rules in the project's existing system or a focused DESIGN.md; token values and their implementation must agree. The contract reference distinguishes a structured design document from a task/navigation contract.

External generators and other design skills are optional executors, not authorities that may replace the approved structure. Consult integrations only when using one. Never silently upload content, install tools, enable hooks, create accounts or expand scope.

4. Build the approved scope

Prefer the existing stack and native semantics. A dynamic experience does not require a heavy framework; multiple pages do not require a backend. Use actual content and data shapes, including long labels and missing values.

Implement the navigation/state contract across every entry point, including deep links, alternate audiences, browser history and recovery paths. No fake controls, placeholder destinations, simulated loading or fabricated availability. Wire charts, tables, status and sources to the same underlying content.

Keep ordinary page scrolling usable. On mobile, change relationships between views where necessary instead of stacking the entire desktop. If offline or print is required, implement that delivery mode deliberately; hidden, unfetched or virtualized content is not automatically present in a complete export.

5. Verify the experience

Follow verification. Execute the agreed tasks against the rendered result with real content and record expected versus observed outcomes. Inspect desktop, narrow screens and intermediate states, not only the hero. Check keyboard/focus, applicable contrast and reflow, icons/motion, assets, links, meaningful data displays and recovery; verify print/offline when required. Reread rendered copy after material edits; a lexical check is not proof of natural voice.

Distinguish observed, source-inferred, not checked, and blocked. Never call a missing browser check a pass, an automated walkthrough a user study, a linter a usability certificate, or a self-assigned score a quality measurement.

Batch findings and fixes coherently, then repeat affected tasks and regressions. Stop for a scope/budget decision if another pass makes no material progress; do not keep generating pages or polishing without an end condition. Unresolved required behavior blocks a success claim. Report it plainly.

Handoff and reuse

Lead with the delivered outcome, supported by the paths actually exercised. Identify changed behavior, material limits and skipped checks without dumping every log into the interface or chat. Keep development notes out of shipped pages. Stop temporary servers you started unless the user requested retention.

Before promoting this method to a global default, run the acceptance scenarios: report, comparison/tool and a different informational identity. Include the user's real baseline when one exists. Obtain acceptance; do not overwrite unrelated personal skills or global instructions. Source availability, discovery, invocation, workflow application and output quality are separate checks.

Reference map

Load when neededPurpose
architectureDeliverable routing, structural choices and content editing
contractsExperience and visual contracts without duplicate sources of truth
craftInterface copy, typography, composition, icons, visual explanation and motion
verificationTask evidence, accessibility, responsive and delivery checks
integrationsOptional tools, compatibility, privacy, cost and authority
evidenceSources, dates, applicability and conflicts; not mandatory upfront research

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.