v3.3.0 — the accessibility layer for AI coding agents

First tagged release (series 3.0→3.3 shipped in one week — see CHANGELOG for everything).

Highlights

• 12 MCP tools + 5 prompts + a skill: audit → fix → document → watch, WCAG 2.2, es/en
• Rendered audit (Playwright): computed contrast with alpha compositing, 2.5.8 target size, focus indicator, :focus/:hover state contrast, same-origin iframes
• 0-100 weighted score + ARIA validity (roles, refs, values) + autocomplete 1.3.5 + link purpose 2.4.4
• SARIF 2.1.0 export → GitHub code scanning / PR annotations
• Accessibility error budget: only NEW findings block deploys (SRE-style, baseline expiry)
• Honest badge as accessible SVG — score, date, screening scope; never claims conformance
• WCAG-EM conformance ladder prompt: express → guided (agent-verified) → conformance (human-signed)
• Light same-domain crawl (pages parameter), accessibility-tree snapshots with tree diff
• Legal statements (EAA / RD 1112/2018), text-over-image pixel contrast, zero core dependencies

Install: claude mcp add a11y-toolkit -- uvx --from a11y-toolkit a11y-toolkit-mcp (PyPI pending trusted-publisher activation) — or today from this tag: uvx --from git+https://github.com/kinti/a11y-toolkit@v3.3.0 a11y-toolkit-mcp