本文へ移動
cccskills
無料GitHub で公開

open-claude-ui-review

Review or polish a product UI for accessibility, brand fidelity, hierarchy, rhythm, responsive behavior, themes, interaction states, and generic AI-template patterns. Use for visual audits, accessibility checks, UX polish, UI critique, or pre-ship design review. Do not use for generic code review, backend review, or API design.

インストール方法を見る

含まれるファイル(7)

  • SKILL.md4.1 KB
  • agents/openai.yaml148 B
  • LICENSE.md1.9 KB
  • references/accessibility.md3.5 KB
  • references/interaction-and-verification.md4.1 KB
  • references/visual-quality.md3.3 KB
  • tests/evals.json3.3 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

UI Design Review

Judge the rendered product, not the source in isolation. A design review distinguishes standards failures, product-quality problems, and subjective preferences so the user can trust the verdict.

Mutation boundary

Review and audit requests are report-only. Inspect, execute read-only checks, and return evidence-backed findings without editing files. Change files only when the user asks to fix, polish, redesign, or implement, or when a parent workflow explicitly carries that authorized fix scope.

Establish the baseline

Read the affected UI, its tokens/components, supported themes, product context, and user-stated constraints. Capture the current rendered state before proposing a correction. Existing brand and system choices outrank generic taste; accessibility requirements do not.

When Impeccable is installed, reuse current hook findings and named refinement evidence. Do not run its detector again unless no current hook evidence exists or a specific fix needs rechecking.

Select review lanes

Read only the relevant references, resolved relative to this skill:

LaneRequired reference
Contrast, semantics, keyboard, forms, motion, target size, or WCAG reviewreferences/accessibility.md
Brand fidelity, content discipline, hierarchy, rhythm, system consistency, or generic-template reviewreferences/visual-quality.md
Interactive states, responsive/theme coverage, runtime verification, or final polishreferences/interaction-and-verification.md

A full pre-ship review uses all three. A focused contrast question reads only accessibility. Complete every applicable lane before aggregating findings so early fixes do not hide later issues.

Run lanes in the current agent by default. When the session exposes agent tools and multiple read-only lanes are genuinely independent, the active agent may delegate the minimum useful number with non-overlapping ownership, then integrates and verifies the result itself. Tool availability never changes the review contract.

Findings contract

Deduplicate findings and classify them:

  1. Blocker — standards failure or interaction/accessibility defect that prevents safe use or completion.
  2. Quality issue — broken hierarchy, system inconsistency, missing state, misleading content, responsive/theme defect, or unearned template pattern with clear product impact.
  3. Advisory — subjective refinement or deterministic detector signal that needs human judgment.

Every finding names the affected surface, evidence, user impact, and concrete correction. Do not report a preferred aesthetic as a standards failure.

Fix and re-verify

When edits are authorized, fix blockers first, then in-scope quality issues. Preserve established tokens/components and avoid unrelated redesign. Re-run the actual interaction, affected viewport/theme/state matrix, exact contrast checks, and any focused automated check after edits. Detector output remains advisory even when a fix was requested.

Completion

Return a verdict—ready, ready after named decisions, or needs iteration—plus counts by classification, evidence for each unresolved item, authorized fixes applied, and explicit unverified states. A clean source scan without rendered interaction is not a completed UI review.

When not to use

  • Generic code review, security review, backend/API review, or architecture critique
  • Creating a new screen, flow, or visual direction; use open-claude-ui-design
  • Extracting tokens or component contracts; use open-claude-design-system
  • Logic-only UI changes that preserve the rendered interface
  • Accessing a Claude Design project or preview; use open-claude-design for retrieval, then return here for the review

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Access the Anthropic product named Claude Design when the request contains that exact name, a claude.ai/design URL, or a .dc.html file. Also use for standalone-to-artifact migration and Claude Design artifact links. Use for projects, files, design systems, previews, conversations, comments, sharing, and code-to-design or design-to-code synchronization; otherwise stay inactive.

日本語の概要は準備中です。原文の説明を表示しています。

maxritter/open-claude-design342026年10月12日 更新

Apply product-grounded visual quality constraints during user-visible UI changes. Use for layout, styling, content hierarchy, theming, responsive behavior, or interaction affordances. Stay inactive for logic-only changes that preserve the rendered interface.

日本語の概要は準備中です。原文の説明を表示しています。

maxritter/open-claude-design342026年10月12日 更新

Extract, document, or normalize a product UI design system from code, screenshots, or brand sources. Use for visual tokens, theme variables, typography, spacing, colors, radii, shadows, component inventories, variants, states, or building a reusable UI library. Do not use for software architecture, database schemas, or generic system design.

日本語の概要は準備中です。原文の説明を表示しています。

maxritter/open-claude-design342026年10月12日 更新

Create or redesign product UI in an existing codebase, including a wireframe, substantive visual variations, and an interactive prototype. Use for requests about UI layout, visual direction, screens, flows, components, a landing page, dashboard, or making an interface feel polished. Do not use for backend work, system architecture, decks, or logic-only UI fixes.

日本語の概要は準備中です。原文の説明を表示しています。

maxritter/open-claude-design342026年10月12日 更新

maxritter のスキルをすべて見る

このスキルの問題を報告する