frontend-visual-verify

name	frontend-visual-verify
description	Visual verification of UI components via Storybook snapshots and Playwright MCP. Use when verifying component rendering.
allowed-tools	Read, Glob, Grep, Bash(ls:*), Bash(mkdir:*), Bash(stat:*), mcp__playwright__browser_navigate, mcp__playwright__browser_snapshot, mcp__playwright__browser_take_screenshot, mcp__playwright__browser_tabs, mcp__storybook-mcp__get-story-urls, Write($JAAN_OUTPUTS_DIR/frontend/**), Task, AskUserQuestion, Edit(jaan-to/config/settings.yaml)
argument-hint	[storybook-url or component-path or frontend-design output]
license	PROPRIETARY
disable-model-invocation	true

frontend-visual-verify

Visual verification of UI components via Storybook accessibility snapshots and Playwright MCP.

Context Files

• $JAAN_CONTEXT_DIR/tech.md - Tech stack context (optional)
• $JAAN_CONTEXT_DIR/design.md - Design system guidelines (optional, #visual-standards and #storybook sections)
• $JAAN_TEMPLATES_DIR/jaan-to-frontend-visual-verify.template.md - Output template
• $JAAN_LEARN_DIR/jaan-to-frontend-visual-verify.learn.md - Past lessons (loaded in Pre-Execution)
• ${CLAUDE_PLUGIN_ROOT}/docs/extending/frontend-ui-workflow-reference.md - Shared reference (visual scoring rubric, network policy, MCP degradation)
• ${CLAUDE_PLUGIN_ROOT}/docs/extending/language-protocol.md - Language resolution protocol

Input

Arguments: $ARGUMENTS

Accepts any of:

• Storybook URL — http://localhost:6006/?path=/story/{component} (localhost only by default)
• Component path — Path to component file (derives Storybook URL from naming)
• frontend-design output — Path to design output folder (reads preview HTML)
• Reference description — Text describing expected visual appearance
• Empty — Interactive wizard

If no input provided, ask: "What component should I verify? (Storybook URL, component path, or design output path)"

---

Pre-Execution Protocol

MANDATORY — Read and execute ALL steps in: ${CLAUDE_PLUGIN_ROOT}/docs/extending/pre-execution-protocol.md Skill name: frontend-visual-verify Execute: Step 0 (Init Guard) → A (Load Lessons) → B (Resolve Template) → C (Offer Template Seeding)

Also read context files if available:

• $JAAN_CONTEXT_DIR/design.md — Visual standards, accessibility targets

Language Settings

Read and apply language protocol: ${CLAUDE_PLUGIN_ROOT}/docs/extending/language-protocol.md Override field for this skill: language_frontend-visual-verify

Language exception: Generated code output remains in the project's programming language.

---

PHASE 1: Analysis (Read-Only)

Thinking Mode

ultrathink

Use extended reasoning for:

• Determining verification mode (visual vs static)
• Planning verification checklist from component requirements
• Analyzing accessibility tree structure
• Comparing visual output against design specs

Step 1: Determine Output Mode

Check Playwright MCP availability:

1. Attempt to use mcp__playwright__browser_snapshot (any lightweight call)
2. If available: set output_mode = visual-mode
3. If unavailable: set output_mode = static-mode

Reference: See ${CLAUDE_PLUGIN_ROOT}/docs/extending/frontend-ui-workflow-reference.md section "MCP Graceful Degradation Patterns" — Playwright MCP section.

Display mode to user:

• visual-mode: "Playwright MCP detected — full visual verification available."
• static-mode: "Playwright MCP not available — static code analysis only. Visual score will be N/A."

Step 2: Validate URL / Resolve Target

URL validation (CRITICAL — localhost-only default):

Reference: See ${CLAUDE_PLUGIN_ROOT}/docs/extending/frontend-ui-workflow-reference.md section "Playwright Network Policy" for URL validation rules.

1. If URL provided: validate against allowed patterns (localhost:*, 127.0.0.1:*)
2. If external URL: require explicit per-URL user confirmation via AskUserQuestion
3. If component path: construct Storybook URL from component naming convention
4. If design output: use preview HTML path directly

If MCP tool mcp__storybook-mcp__get-story-urls available:

• Use it to get accurate story URLs for the component
• Fallback: construct URL as http://localhost:6006/?path=/story/{component-id}--{story-name}

Present target summary:

VERIFICATION TARGET
───────────────────
Component:    {name}
URL:          {storybook_url or preview_path}
Mode:         {visual-mode / static-mode}
Reference:    {design description or N/A}

Step 3: Read Component Source

Regardless of mode, read the component source for code-level analysis:

1. Read component file — Props, structure, accessibility attributes
2. Read stories file (if exists) — Story states to verify
3. Check design.md — Visual standards (breakpoints, contrast, animation)
4. Check for CVA variants — Grep: "cva(" {component_path}

Build verification checklist:

VERIFICATION CHECKLIST
──────────────────────
Rendering:
  [ ] Component renders without errors
  [ ] All props/variants render correctly
  [ ] Loading/error/empty states work

Accessibility:
  [ ] ARIA attributes present and correct
  [ ] Keyboard navigation functional
  [ ] Color contrast meets WCAG AA (4.5:1 text, 3:1 UI)
  [ ] Focus indicators visible

Visual:
  [ ] Layout matches design intent
  [ ] Typography hierarchy correct
  [ ] Spacing consistent with design system
  [ ] Responsive at key breakpoints
  [ ] Dark mode (if applicable)

• API dependency check: If the component imports API hooks (useQuery, useMutation, fetch) or references API endpoints, add an info-level note: "Component has API dependencies — verify MSW handlers are configured for Storybook stories. See /jaan-to:frontend-story-generate --contract."

---

HARD STOP — Review Verification Plan

Present verification plan to user.

Use AskUserQuestion:

• Question: "Proceed with verification?"
• Header: "Verify"
• Options:
  • "Yes" — Run full verification
  • "No" — Cancel
  • "Edit" — Adjust checklist items

Do NOT proceed to Phase 2 without explicit approval.

---

PHASE 2: Verification (Execute Phase)

Step 4: Execute Verification

visual-mode (Playwright MCP available)

1. Navigate: mcp__playwright__browser_navigate to Storybook URL
2. Snapshot: mcp__playwright__browser_snapshot — Get accessibility tree (primary, fast, deterministic)
3. Analyze accessibility tree:
   • Check ARIA roles and labels
   • Verify heading hierarchy
   • Check interactive element accessibility
   • Validate form labels and associations
4. Screenshot: mcp__playwright__browser_take_screenshot — Capture visual state
5. Repeat for each story state (Default, Loading, Error, variants)
6. Responsive check: Resize and re-snapshot at breakpoints (if applicable)

static-mode (No Playwright)

1. Read component source — Analyze code structure
2. Check accessibility attributes — Grep for ARIA, semantic HTML
3. Analyze CSS classes — Check for responsive utilities, color tokens
4. Review prop types — Validate TypeScript correctness
5. Cross-reference design.md — Check against visual standards

No visual score in static-mode. Report header: "Static analysis only — visual verification requires Playwright MCP."

Step 5: Score Results

Reference: See ${CLAUDE_PLUGIN_ROOT}/docs/extending/frontend-ui-workflow-reference.md section "Visual Scoring Rubric" for score definitions.

visual-mode scoring (0-10)

Evaluate against rubric criteria:

• Rendering correctness (all states)
• Accessibility tree completeness
• Visual consistency with design system
• Responsive behavior
• Interactive states

static-mode scoring

• Visual score: N/A
• Code quality findings only
• Cannot make visual pass/fail conclusions

Step 6: Generate Report

Read template: $JAAN_TEMPLATES_DIR/jaan-to-frontend-visual-verify.template.md

Fill sections:

• Executive Summary: Component name, mode, score, pass/fail
• Verification Results: Checklist with pass/fail per item
• Accessibility Findings: ARIA, keyboard, contrast results
• Visual Findings: Layout, typography, spacing (visual-mode only)
• Screenshots: Captured screenshots (visual-mode only)
• Recommendations: Suggested fixes with severity
• Metadata: Generated date, skill version, mode

Step 7: Quality Check

• All checklist items evaluated
• Score matches rubric criteria
• static-mode report does NOT claim visual pass/fail
• Recommendations are actionable
• Screenshots captured (visual-mode only)

Step 8: Preview & Write Output

Show verification summary with score and key findings.

Use AskUserQuestion:

• Question: "Write verification report?"
• Header: "Write Report"
• Options:
  • "Yes" — Write report files
  • "No" — Cancel

Step 9: Generate ID and Write Output

source "${CLAUDE_PLUGIN_ROOT}/scripts/lib/id-generator.sh"
SUBDOMAIN_DIR="$JAAN_OUTPUTS_DIR/frontend/visual-verify"
mkdir -p "$SUBDOMAIN_DIR"
NEXT_ID=$(generate_next_id "$SUBDOMAIN_DIR")
slug="{component-name-slug}"
OUTPUT_FOLDER="${SUBDOMAIN_DIR}/${NEXT_ID}-${slug}"

Write files:

1. mkdir -p "$OUTPUT_FOLDER"
2. Write {id}-{slug}.md — Verification report
3. Write {id}-{slug}-screenshots/ — Captured screenshots (visual-mode only)
4. Write {id}-{slug}-readme.md — Summary with pass/fail and next actions
5. Update subdomain index:

source "${CLAUDE_PLUGIN_ROOT}/scripts/lib/index-updater.sh"
add_to_index \
  "$SUBDOMAIN_DIR/README.md" \
  "$NEXT_ID" \
  "${NEXT_ID}-${slug}" \
  "{ComponentName} Verification" \
  "{mode}: score {score}/10 — {pass_count}/{total_count} checks passed"

Step 10: Suggest Next Actions

Verification complete!

Results: {score}/10 — {pass_count}/{total_count} checks passed

Next Steps:

• Run /jaan-to:frontend-component-fix to fix identified issues
• Run /jaan-to:frontend-story-generate to add missing story states
• Run /jaan-to:qa-test-cases to generate test cases from findings
• Re-run /jaan-to:frontend-visual-verify after fixes to confirm

Step 11: Capture Feedback

Use AskUserQuestion:

• Question: "How did the verification turn out?"
• Header: "Feedback"
• Options:
  • "Useful!" — Done
  • "Needs improvement" — What should change?
  • "Learn from this" — Capture a lesson for future runs

If "Learn from this": Run /jaan-to:learn-add frontend-visual-verify "{feedback}"

---

Skill Alignment

• Two-phase workflow with HARD STOP for human approval
• Dual output modes: visual-mode (Playwright) vs static-mode (code-only)
• Localhost-only network default with explicit external URL confirmation
• MCP-enhanced with graceful degradation
• No false confidence in static-mode (visual score = N/A)

Definition of Done

• Output mode determined and displayed to user
• URL validated (localhost-only by default)
• All checklist items evaluated
• Score assigned per rubric (N/A in static-mode)
• static-mode report does NOT make visual pass/fail claims
• Screenshots captured (visual-mode only)
• Recommendations are actionable
• Report written with readme for integration
• Output follows v3.0.0 structure (ID, folder, index)
• User approved final result