All skills
simota avatar

/vision

@c805268
by shingo imotasimota/agent-skills85 stars
15

Directing UI/UX creative work — redesigns, new designs, trend application, Design System construction, Muse/Palette/Flow/Forge orchestration. Use for design direction. Offers a co-design pair mode.

Use this Skill: https://skilld.dev/gh/simota/agent-skills/vision

This session only. Nothing lands on disk.

referencedesign-audit-checklist.md

≈2.6k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Design Audit Checklist

Reference for Vision's audit recipe. REVIEW-mode design quality audit: Nielsen 10 heuristics, WCAG 2.2 AA pass/fail grid, token-drift detection, prioritized remediation backlog.


1. Audit Scope Definition

Before scoring, declare scope:

Scope dimension Options
Surface marketing / product / docs / email / mobile / all
Pages list specific URLs or screen names (≤ 10 for one audit cycle)
Component layers tokens / primitives / patterns / templates
Audit depth heuristic-only / + WCAG / + token-drift / + competitive
Output format scorecard / remediation backlog / executive summary

Cap one audit at ~10 surfaces / 90 minutes of evaluation. Larger scopes need batching.


2. Nielsen 10 Usability Heuristics

Score each surface 1-5 against:

# Heuristic Pass signal Fail signal
1 Visibility of system status Loading / progress / save state visible Silent operations, no spinners
2 Match between system and real world Familiar terms, real-world metaphors Jargon, system-centric labels
3 User control and freedom Undo, cancel, back works One-way actions, no escape
4 Consistency and standards Same icons / colors mean same thing Brand inconsistency across screens
5 Error prevention Confirmations, constraints, validation Easy to make irreversible mistakes
6 Recognition rather than recall Visible options, autocomplete Hidden menus, memorize-this-flag
7 Flexibility and efficiency Shortcuts (cmd-K), power-user paths Same path for novice + power user
8 Aesthetic and minimalist design No irrelevant chrome Visual noise, redundant info
9 Error recovery (clear messages) Plain language, suggested fix Codes only, blame user
10 Help and documentation Inline help, contextual tips Manual hidden, no help

Aggregate: avg < 3.0 → P1 systemic remediation; 3.0-4.0 → targeted; ≥ 4.0 → polish.


3. WCAG 2.2 AA Pass/Fail Grid

Mandatory checks per surface:

Check Pass criterion Tool
Text contrast ≥ 4.5:1 (normal), ≥ 3:1 (large 24px+) axe DevTools, Stark, Contrast app
UI component contrast ≥ 3:1 (borders, icons, focus rings) axe DevTools
Focus visible (2.4.7) Every interactive element shows focus Manual tab test
Focus not obscured (2.4.11, new in 2.2) Focus indicator never hidden by sticky elements Manual scroll + tab
Target size (2.5.8, new in 2.2) Interactive targets ≥ 24×24 CSS pixels DevTools box check
Dragging movements (2.5.7, new in 2.2) Drag has single-pointer alternative Manual try-without-drag
Consistent help (3.2.6, new in 2.2) Help links in same relative location Cross-page check
Redundant entry (3.3.7, new in 2.2) Don't re-ask info already collected Multi-step form check
Authentication (3.3.8, new in 2.2) Don't require cognitive challenges (math, memorization) Auth flow check
Keyboard navigation All actions reachable via keyboard Tab through entire flow
Skip links "Skip to main content" present Tab from page load
Heading hierarchy h1 → h2 → h3 (no skips) axe outline check
Image alt text Decorative: alt=""; meaningful: descriptive axe scan
Form labels Every input has visible label axe scan
Color not sole signal Status conveyed by icon + color, not color alone Visual scan
Motion / animation Respects prefers-reduced-motion DevTools emulation

ADA Title II reminder: as of 2026-04-24 (state/local pop. ≥ 50K), WCAG 2.1 AA is mandated; penalties up to $150K/violation. Document failures as legal risk.


4. Token Drift Detection

Scan for hardcoded values where tokens should exist:

Pattern Risk Remediation
Inline color hex (color: #FF5722) Brand drift Replace with var(--color-action-primary)
Inline px spacing (padding: 12px) Spacing scale drift Replace with var(--space-md)
Inline font-size (font-size: 17px) Type scale drift Replace with var(--text-base)
Magic-number radii (border-radius: 7px) Inconsistent corner language Voyager[ios] to scale (--radius-md)
Duplicate tokens across themes "Frankenstein system" Consolidate; cascade Core → Brand
Token defined but never referenced Dead token Remove; reduce noise
Token referenced but undefined Build break / silent fallback Define or remove reference

Tools: design-system-linter, Theo, Style Dictionary's validate, Figma Variables API, custom regex on CSS-in-JS.

Drift score:

drift % = (hardcoded_values / total_style_declarations) × 100
  • < 5% → healthy
  • 5-15% → targeted cleanup
  • 15% → systemic; pause feature work


5. Design System Anti-Pattern Scan

Anti-pattern Detect by Remediation
Component sprawl (5+ button variants) Inventory components Consolidate to ≤ 4 variants
Atomic-rigidity in multi-brand Tokens shared but behavior diverges Apply Core → Brand → Product cascade
Mixed paradigm (some BEM, some Tailwind) grep class patterns Pick one paradigm
Spacing values not on scale All px values used Voyager[ios] to 4 / 8 / 16 / 24 / 32 / 48 / 64
Type ramp > 8 steps Inventory font-sizes Reduce to 5-7 steps
Color palette > 12 hues Inventory hue values Reduce to 5-7 hues × shades
Shadow palette > 6 levels Inventory box-shadows Reduce to 4 levels
Z-index unbounded (z: 9999) grep z-index values Use named tokens (--z-modal, --z-toast)
Modal-in-modal UX walkthrough Convert nested modal to drawer or page
Loading without skeleton Visual scan Add skeleton screens

6. Performance + UX Crossovers

Check Target Source
LCP (Largest Contentful Paint) < 2.5s Core Web Vitals
CLS (Cumulative Layout Shift) < 0.1 Core Web Vitals
INP (Interaction to Next Paint) < 200ms Core Web Vitals
Bounce rate < 55% Hotjar 2026
Page weight < 1.5MB on initial load Performance budget
Time to interactive < 3s perceived Google

Performance regressions are UX regressions — flag in audit.


7. Prioritized Remediation Backlog

Score each finding on Effort × Impact:

Effort Impact Priority
Low High P1 — do first
Low Med P2
Low Low P3 (or skip)
Med High P1
Med Med P2
Med Low P4
High High P2 (plan over sprints)
High Med P3
High Low drop

Output format:

## Remediation backlog

| ID | Finding | Heuristic / WCAG | Surface | Effort | Impact | Priority | Owner |
|---|---|---|---|---|---|---|---|
| F-01 | Focus ring missing on tabs | WCAG 2.4.7 | Settings | L | H | P1 | Palette |
| F-02 | Inline color hex on CTAs | Token drift | All | L | M | P2 | Muse |
| F-03 | Modal-in-modal in flow X | Anti-pattern | Checkout | M | H | P1 | Vision |
| ... |

8. Stakeholder Communication

Audit deliverable layers (most stakeholders read only the top):

# Design Audit — <project>

## TL;DR (5 lines max)
- Surfaces audited: ___
- Heuristic avg: ___ / 5
- WCAG 2.2 AA pass rate: ___ %
- Token drift: ___ %
- P1 issues: ___ items

## Top 3 risks
1. ____ (legal/UX/brand)
2. ____
3. ____

## Top 3 wins (preserve these)
1. ____
2. ____
3. ____

## Detailed findings
[full backlog table]

## Methodology
- Heuristics scored by ___
- WCAG checked with ___ tools
- Drift detected by ___

9. Common Pitfalls

Pitfall Avoidance
Audit scope too broad → useless avg scores Cap at ~10 surfaces; batch larger scope
Score without specific finding → unactionable Every score links to ≥ 1 finding ID
WCAG checked only with auto-tools (~30% catch rate) Manual keyboard + screen reader pass mandatory
Findings without effort/impact → wishlist, not backlog Score every finding
Token drift % not measured → vibes-based Use linter; quantify
Anti-patterns identified but no replacement pattern Pair every anti-pattern with the recommended pattern
Audit produced, then shelved Hand off to specific owner per finding
Audit re-run not scheduled Quarterly cadence; track delta over time
Performance ignored LCP / CLS / INP are UX metrics; include
Brand-fit not audited Use brand-strategy.md scoring rubric in parallel

10. Decision Walkthrough Template

Audit scope:
  Surfaces (≤ 10): ____
  Audit depth: heuristic / + WCAG / + drift / + competitive
  Output: scorecard / backlog / executive

Heuristic scores (1-5):
  H1 status: ___  H2 real-world: ___  H3 control: ___  H4 consistency: ___  H5 prevention: ___
  H6 recognition: ___  H7 flexibility: ___  H8 aesthetic: ___  H9 error: ___  H10 help: ___
  Avg: ___ → action class: cosmetic / targeted / systemic

WCAG 2.2 AA grid:
  Text contrast pass rate: ___ %
  UI contrast pass rate: ___ %
  Focus visible (2.4.7): ✓ / ✗
  Focus not obscured (2.4.11): ✓ / ✗
  Target size (2.5.8): ✓ / ✗
  Dragging movements (2.5.7): ✓ / ✗
  Consistent help (3.2.6): ✓ / ✗
  Redundant entry (3.3.7): ✓ / ✗
  Authentication (3.3.8): ✓ / ✗
  Keyboard nav: ✓ / ✗
  ADA Title II compliance status: compliant / at risk
  Overall pass rate: ___ %

Token drift:
  Hardcoded color: ___ %
  Hardcoded space: ___ %
  Hardcoded type: ___ %
  Total drift: ___ %  → < 5 healthy / 5-15 cleanup / > 15 systemic

Anti-pattern scan results:
  □ Component sprawl
  □ Atomic-rigidity in multi-brand
  □ Mixed paradigm
  □ Off-scale spacing
  □ Bloated type ramp
  □ Bloated palette
  □ Bloated shadows
  □ Unbounded z-index
  □ Modal-in-modal
  □ Loading without skeleton

Performance:
  LCP: ___s  CLS: ___  INP: ___ms

Backlog generated: ___ items
  P1: ___  P2: ___  P3: ___  P4: ___

Owners assigned: ✓ / ✗
Re-audit scheduled: <date>

Handoff:
  → Muse (token drift remediation)
  → Palette (WCAG fixes)
  → Builder (component fixes)
  → Beacon (performance regressions)
  → Canon[regulatory] / Canon (legal-risk findings)

11. References

  • Nielsen Norman Group, 10 Usability Heuristics
  • W3C WCAG 2.2 (October 2023, current AA standard)
  • DOJ ADA Title II final rule (2026 compliance deadline)
  • Core Web Vitals (Google)
  • DTCG specification v2025.10
  • Adrian Roselli on APCA / WCAG 3 status

Source: SKILL.md on GitHub

No alerts14d5 checks · Risk SAFE
  • Gen Agent Trust Hub14d

    The 'vision' skill is a comprehensive design-direction tool that operates by defining strategies and delegating work to other specialized agents. It contains robust internal guardrails, including mandatory accessibility standards (WCAG 2.2 AA) and ethical AI principles. No security risks were identified.

  • Socket14d

    No alerts

  • Snyk14d

    Risk: LOW · No issues

  • Runlayer6mo

    9 files scanned · No issues

  • ZeroLeaks5mo

    Score: 93/100 · 2 sections analyzed

Signed by skilld at c805268. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 days ago.

Activeupdated 2 weeks ago

README badge

README badge for simota/agent-skills/vision