KIN Design
Use KIN as a product contract, not as a visual theme preset.
Required context
- Locate the KIN repository root from this skill directory and read
DESIGN.mdandDELIVERY.mdcompletely. - Read
components/catalog.mdandcomponents/terminology.mdwhen the task adds, selects, renames, audits, or claims completion of a component. - Read
components/ai-assistance.md,review-and-approval.md,background-work.md, orcharts-and-analysis.mdwhen AI output, evidence, approval, side effects, durable jobs, media review, or analytical charts are in scope. - Read
principles/verification.mdbefore claiming accessibility, browser, motion, zoom, RTL, or component-completion evidence. - Read the consuming project's
AGENTS.md,DESIGN.md,PRODUCT.md, package manifest, global styles, token source, component primitives, and relevant route files when present. - Preserve the user's explicit request, real product behavior, data model, accessibility, routes, analytics, and established brand assets.
- Read only the routed reference below. Do not load every reference.
If the KIN contract is unavailable, say so and do not claim KIN compliance.
Design read
Before changing code, state internally or in a concise progress update:
Mode: <build | redesign-preserve | redesign-overhaul | audit | review>
Surface: <information-site | product-workspace | ecommerce | canvas | other>
Primary user task: <one sentence>
Register: <calm/dense/editorial/operational/etc.>
Constraints: <brand, data, platform, accessibility, performance>
Ask one focused question only when two plausible answers would materially change the product, brand, information architecture, or data behavior. Otherwise proceed with the safest evidence-based interpretation.
Route the task
- New page, application, component, or feature: read
references/build.md. - Existing interface redesign: read
references/redesign.md. - KIN adoption, migration, or evidence recording: read
references/adoption.mdbefore the applicable build or redesign workflow. - Compliance, accessibility, performance, or anti-slop audit: read
references/audit.md. - Design critique without implementation: read
references/review.md. - Product-family and visual-register selection: also read
references/registers.md.
Source precedence
Resolve conflicts in this order:
- User's explicit current request.
- Safety, accessibility, privacy, and legal requirements.
- Existing product behavior, real data, routes, and documented brand decisions.
- Project-local
AGENTS.md,DESIGN.md, tokens, and component contracts. - KIN normative
MUSTandMUST NOTrules. - KIN defaults and examples.
- Third-party component defaults and aesthetic references.
Do not silently change routes, navigation labels, data meaning, form fields, analytics identifiers, permissions, brand assets, or public API behavior.
Core execution loop
- Inspect the real project and identify the source of truth.
- Separate facts, assumptions, user choices, and generated suggestions.
- Reuse existing tokens and primitives before creating new ones.
- Plan information hierarchy and states before styling.
- Implement the smallest coherent change.
- Verify realistic content, empty/error/loading/stale states, light/dark themes, target viewports, keyboard use, touch where relevant, and both normal and reduced motion. Label automated and manual evidence separately.
- Inspect rendered screenshots whenever a runnable UI exists.
- Report evidence, deviations, unresolved issues, and rollback.
For consuming projects with kin.config.json, run the structural adoption check before implementation. During an audit, the optional KIN CLI may locate candidates, but every match must be confirmed in source context and rendered UI before it becomes a finding.
Stop conditions
Stop and request direction when the task requires an undocumented brand replacement, destructive data or route migration, fabricated metrics or AI behavior, a new external dependency with material product impact, or a choice that changes the product's information architecture beyond the requested scope.
Completion contract
Do not say a design is compliant only because the code builds. A completed response identifies:
- design and product decisions;
- files and dependencies changed;
- commands and tests actually run;
- screenshots or visual checks performed;
- accessibility and responsive checks;
- known issues and intentional exceptions;
- rollback path when the change is material.