SEO + Accessibility Audit

Pairs SEO and accessibility in one composite because they share root causes: both fail when content isn't discoverable, semantic, or properly structured. A site that fails for screen readers usually also fails for search bots, and vice versa.

Auto-invocation triggers

• User says "audit my site", "check SEO", "accessibility audit", "is this site good"
• Pre-launch / pre-redesign reviews
• Quarterly site health check
• After a search rankings drop or accessibility complaint
• Before submitting a site to government/enterprise procurement (a11y compliance often required)

Workflow

Phase 1 — Broad scan (always, in parallel)

Dispatch in parallel:

audit-website (squirrelscan, 230+ rules across 15 categories)

• SEO: meta tags, Open Graph, schema markup, sitemap, robots, canonical, internal links
• Accessibility: alt text, ARIA, color contrast, focus order, semantic HTML
• Technical: HTTPS, performance, broken links, redirects
• Returns: structured findings with severity + URL

seo-audit (focused SEO framing on top of audit-website)

• Indexability: robots.txt, sitemap.xml, noindex tags
• On-page: title/meta uniqueness, heading hierarchy, keyword targeting
• Content: thin content, duplicate content, content-to-code ratio
• Returns: SEO backlog with remediation order

Phase 2 — Deep accessibility (always, requires running site)

playwright-best-practices axe-core scan:

• Run axe-core against each top-level route
• WCAG 2.1 AA + AAA flags
• Returns: per-route a11y violations with element selectors

If the site isn't accessible from the dev machine: skip Phase 2, mark as PARTIAL, recommend running locally with the site up.

Phase 3 — Design-system pattern review (always)

ui-audit accessibility-focused mode:

• Color palette: contrast ratios across actual UI states
• Typography: font sizes, line heights, readability scores
• Layout patterns: focus order, keyboard navigation, skip links
• Component patterns: form labels, error messaging, ARIA usage
• Returns: pattern-level recommendations distinct from per-page violations

Phase 4 — Reconcile findings

Combine all four sources into one severity-ranked list, organized into 3 buckets:

Bucket A — SEO only (search-discoverability issues)

• Missing meta descriptions
• No canonical tags
• Missing schema markup
• Slow page load (Core Web Vitals)

Bucket B — Accessibility only (assistive-tech issues)

• Color contrast failures
• Missing alt text on decorative images
• Keyboard traps
• ARIA misuse

Bucket C — Both (highest priority — impacts both audiences)

• Missing/empty <title> tags (SEO + screen readers)
• No heading hierarchy (SEO + navigation)
• Images without alt text (SEO + screen readers)
• No semantic HTML (<main>, <nav>, <article>) (SEO + screen readers)
• Inaccessible forms (SEO can't index, users can't submit)
• JS-only content (search bots and screen readers both miss it)

Bucket C findings are always CRITICAL — they fail for both audiences with one root cause. Fix these first; ROI is doubled.

Phase 5 — Severity ranking + remediation plan

For each finding:

• Severity: CRITICAL (blocks audience) / HIGH (degrades experience) / MEDIUM (best practice) / LOW (polish)
• Effort: 5min / 30min / 2h / day+
• Affected scope: single page / template / sitewide

Sort by (impact × audiences) ÷ effort for the action plan.

Phase 6 — Capture

Write report to ~/.claude/projects/<slug>/memory/seo_a11y_audit_<domain>_<date>.md. Trend visible across periodic runs.

If pre-launch and any CRITICAL/HIGH found: surface as a launch blocker.

Reconciliation

SEO + ACCESSIBILITY AUDIT — <domain> — <date>

Tools run:  audit-website ✓, seo-audit ✓, axe-core ✓ (N routes), ui-audit ✓
Pages scanned: N

CRITICAL — Bucket C (both audiences):
  ✗ <finding>     (5min, sitewide)    Fix: <action>
  ✗ <finding>     (30min, single page) Fix: <action>

CRITICAL — Bucket A (SEO):
  ✗ <finding>

CRITICAL — Bucket B (A11y):
  ✗ <finding>

HIGH (X), MEDIUM (Y), LOW (Z): see report.

Top 5 actions (impact × audiences ÷ effort):
  1. <action> (fixes 1 CRITICAL Bucket C, 5min)
  2. <action> (fixes 3 HIGH Bucket B, 30min)
  3. ...

Memory: <report path>
Launch blocker: YES / NO

Outputs / Evidence

• Reconciled severity-ranked report split by bucket
• Effort-sorted remediation plan
• Memory file for trend tracking
• Launch-blocker verdict if pre-launch context

Failure / Stop Conditions

• Site not reachable → audit-website + seo-audit fail; partial report from ui-audit only with note
• Playwright cannot run (CI/headless env without browser): skip Phase 2, mark as PARTIAL, recommend manual axe DevTools scan
• audit-website CLI not installed: stop with install instruction; this skill needs squirrelscan to function

Memory Hooks

• Read prior reports for trend (e.g., "color contrast HIGH flagged 4 quarters running — push design system fix")
• Read documented false-positive list (e.g., "decorative SVG with empty alt is intentional, suppress")
• Write report as primary output